Ask Jordan Pollack About AI - Or Anything Else

What can we say about Jordan Pollack? He's an associate professor at Brandeis University who is heavily engaged in artificial intelligence research. He's also a high-tech entrepreneur and has interesting things to say about information property abuse and other matters that affect computer and Internet users. If you're into AI, you are probably familiar with Jordan, at least by reputation. If not, you may never have heard of him. And that's a shame, because his work impacts almost all of us in one way or another. So read his Web pages and post questions below. We'll submit the 10 - 15 highest-moderated ones by e-mail tomorrow. Jordan's answers will appear within the next week, and we expect them to be wowsers.

Comments

ethical quandaries

[Anonymous Coward]

You work at a very Jewish university. I expect you'd be at least somewhat familiar with the contents of the Torah (Old Testament, whatever you want to call it - I'm Christian, so I'll stick with the Old Testament). Genesis specifically states that God made Man in his image. You're proposing to create intelligence in Man's image.

This is clearly an attempt to also create something in God's image. This means you're doing two things:

1. attempting to exercise powers reserved for God and God alone
2. engaging in idolatry.

How can you possibly morally justify your research?

Re:ethical quandaries

[Xenophon Fenderson,]

Given the things which have been done in the name of religion in this century, this is a blinkered and irresponsible attitude at best.

Rev. Dr. Xenophon Fenderson, the Carbon(d)ated, KSC, DEATH, SubGenius, mhm21x16

Re:ethical quandaries

[Nicolas MONNET]

s/morally/superstitiously/

Religion is not moral. I'm not religious; I'm not immoral either. Thank you.

Re:ethical quandaries

[K.]

You are a moron. AI is not a quest to
create a human intelligence. A burst condom
is all that takes.

In fact, surely procreation is results in the
creation of an intelligence in Man's image, and
is therefore by your argument morally wrong?

K.
-

Re:ethical quandaries

[Error Spelling]

Apparently there were a couple of guys in Medieval Europe who believed in God AND felt quite rosy about killing infidels.

Re:ethical quandaries

[Error Spelling]

And science has to work in the real world. Religion doesn't. Don't be so smug about what you think the real world is. Religion is as much about the real world as science. Different aspects of the world certainly, but real nevertheless. Thus science is self correcting. Religion is not. Religion is actually self-correcting, too, if you think about it. It just moves more slowly. Neither [science nor religion] are perfect. Science recognizes this. Religion does not. I would argue that most religions have a much more profound acceptance of human fallibility than science. Science doesn't have the notion of Original Sin, for instance. That's why books like Kuhn's 'The Structure of Scientific Revolutions' are such shocking revelations to practising scientists. They are so convinced of the "absoluteness" of their work that they fail to see the historicism, enculturation, bias, prejudice, and arbitrariness of their theories. In fact, Science as an institution is founded on the belief that man is perfectible. The science proponents say "Whatever is wrong must have a cause and we can find that cause and prevent the wrong from happening". Communism shares the scientific mindset that the root of evil can be excised from reality and destroyed. I'll take the humility of the religious man over the hubris of the scientist. That said, I appreciate the great things that science has been able to accomplish. It's just not as deep in its understanding of human nature as the religious (Judeo-Christian or any other) view.

Re:ethical quandaries

[Error Spelling]

Ugghh!! I hate it when the formatting gets trashed

And science has to work in the real world. Religion doesn't.

Don't be so smug about what you think the real world is. Religion is as much about the real world as science. Different aspects of the world certainly, but real nevertheless.

Thus science is self correcting. Religion is not.

Religion is actually self-correcting, too, if you think about it. It just moves more slowly.

Neither [science nor religion] are perfect. Science recognizes this. Religion does not.

I would argue that most religions have a much more profound acceptance of human fallibility than science. Science doesn't have the notion of Original Sin, for instance. That's why books like Kuhn's 'The Structure of Scientific Revolutions' are such shocking revelations to practising scientists. They are so convinced of the "absoluteness" of their work that they fail to see the historicism, enculturation, bias, prejudice, and arbitrariness of their theories. In fact, Science as an institution is founded on the belief that man is perfectible. The science proponents say "Whatever is wrong must have a cause and we can find that cause and prevent the wrong from happening". Communism shares the scientific mindset that the root of evil can be excised from reality and destroyed. I'll take the humility of the religious man over the hubris of the scientist. That said, I appreciate the great things that science has been able to accomplish. It's just not as deep in its understanding of human nature as the religious (Judeo-Christian or any other) view.

Re:ethical quandaries

[jrice_blue]

Diclaimer: I am not christian, I am not Jewish.

If man was created in God's image, wouldn't that presuppose that man would want to create? "God's image" in this case includes the trait 'creates'. Therefore, man creates. Where does it say (anywhere?!?) that man should not create?

Totally flawed logic.

Also, why should acedcemic research be at all justified by religion?

Re:ethical quandaries

[rm-r]

I am intelligent, and do not beleive in God. I will not kill you for a number of reasons- but most notably a lack of personal gain from your death, and should this fail a morality and criminal justice system telling me otherwise.

Re:ethical quandaries

[SteveM]

Given some of the things done in the name of religion, I'll take science every time.

Granted, both are practiced by people and as such inherit all of the failings there of. Neither are perfect. Science recognizes this. Religion does not.

And science has to work in the real world. Religion doesn't. Thus science is self correcting. Religion is not.

I am not advocating blind faith in science or technology. But the only good solutions to our problems are those where science and technology play a big part. It is up to us to solve our problems. We should use all of the tools at our disposal.

Steve M

Re:ethical quandaries

[SteveM]

Religion appears to be about the real word. But it is not. That is not to say it isn't very real and very important to people in the real world and thus very much apart of their world. But dropping balls from a tower gives repeatable results. Saying a prayer does not (except maybve for a null result).

Science has built in the ability, perhaps even the requirement, to correct itself. And I'll say it again, science is performed by humans and humans have flaws. Thus those flaws show up in the human endeavor called science. Never the less, the self correcting aspect of science works.

Religion doesn't self correct. It's use of dogma makes it fight change. Religions do change, but those changes come from without, not from within.

Religions do recognize human fallibility. But it is not the fallibility of humans I am discussing here. It is the failability of the science and religion. Science recognizes that it does not have all the answers and thus is not complete or perfect. One certainly can't say that about religion (except maybe Zen).

In fact, Science as an institution is founded on the belief that man is perfectible.

Say what? Science is based on the belief that the universe is explainable. It has nothing to do with perfecting man. Isn't that what religions try to do?

The science proponents say "Whatever is wrong must have a cause and we can find that cause and prevent the wrong from happening".

Where do you come up with this stuff? Scientists look at a phenomena and assume it has a rational explanation. They then try to discover the explanation. It has nothing to do with wrong or right.

... It's just not as deep in its understanding of human nature as the religious (Judeo-Christian or any other) view.

Any institution that thinks abstinence is the best form of birth control has a pretty shallow understanding of human nature. I'll take science over religion every time. (Agian, this is not about individual practitioners it is about the institutions.)

Steve M

Question: Should AI's have rights?

[Anonymous Coward]

My Question:
Suppose that one day research produced a real AI -- not a half-baked hyperbole from the Marketing Department -- but a true AI with free will. And say for the sake of argument that, even though it was quite a technological achievement, it wasn't particularly good at anything. It fancied itself as a hacker, but wasn't. It played Quake all day, "lived" in chat rooms all night, and posted on Slashdot in its spare time :)
Then suppose that the agency funding the project decided to literally pull the plug. Wouldn't this be like killing a living being? Should we extend human rights to AI's before something like this happens?

Spiritual Robots

[Anonymous Coward]

I remember reading about spiritual robots and it really got me thinking about what will happen to AI if a mechanical being becomes sentient. Given that religion in general developed as a primitive communal instinct that drives a particular group to follow the basic precepts of real evolution (don't kill others of your kind but kill those that are different, have sex - multiple wives are okay, don't screw with the gene pool by getting your sister pregnant, etc.), what sort of drives do you think will end up influencing the religion the robots will create for themselves (which is a given, since they will initially be more primative than humans, and therefore more inclined to blind faith that personifies those things that are important to their survival).

I dunno - I think the whole idea of sentient robots is great, but as long as they are programmed with drives for self-preservation, and a bit (but not enough) intelligence, I think they could stand to eventually make the same mistakes that man has. What do you think?

"Terminator"-type future?

[Anonymous Coward]

Two questions for you, if you don't mind:

1) Do you believe we will eventually create an AI with thought and reasoning capabilities completely indistinguishable from human beings?

2) Do you believe that, given today's rate of technological advances, we will reach a point where machines will attain sentience (and possibly threaten the existence of humankind)? Given the recent prediction Sun's Bill Joy made and the high-profile panel discussion which was held on the subject, I'd be curious what someone on the forefront of AI research has to say.

Thanks for doing this interview!

Just another AC, blowin' in the wind...

Some online AI resources

[mcoletti]

Some online AI resources:

The Hitch-hiker's Guide to Evolutionary Computation contains links to some online software, most of which is free and open-source.

The comp.ai related FAQs.

The about.com AI page appears to be a good starting point for many AI related web sites.

Hope this helps.

Genetic programming successes

[mcoletti]

Genetic programs of fair complexity have been successfully created to, say, competitively play robotic soccer. In fact, genetic programming is fairly lucrative. You can't argue with success.

Intellectual Property and Licence Brokerages

[DG]

I recently read the portion of your website that discusses your proposed solution to the current (rather thorny) intellectual property and software licencing issues.

For those that have not read it, the nutshell summary is to produce a fixed number of software licences, each of which entitles you to the lastest/greatest version of whatever software is licenced. These licences, being fixed in number, can then be bought and sold on what amounts to a commodity market.

This is an interesting idea, but it suffers from at least two problems that I can see: a practical problem, and a philosophical one.

The practical problem is that, unlike other commodities like gold, oil, grain, or money, there is zero cost to duplicate the actual stuff the licence scrip represents - and it's the stuff, not the licence that actually does the work. You may create an artificial shortage of licences, but it has been demonstrated time and time again that there is no practical way to defeat the copying of digital media. The shortage of licences does not translate into a shortage of the software the licence represents.

The more cerebral problem revolves around the fact that software is not a consumable. Instead, the primary use of software, outside of the entertainment industry, is to solve problems. Software is a tool, the same way my 14mm wrench is a tool.

Assuming one _could_ create an enforcable copy-prevention scheme that would allow the limited-licence commodity to function, do we really want a world in which the number of tools to solve particular problems is limited? It would certainly suck if there were only 100 14mm wrenches in the world.

What I think the great victory of the Free Software movement is is that software stops being a product with intrinsic value. Instead, software goes into the general pool of knowledge for the use and re-use of other people with problems to solve.

A program is a lot like a surgical technique - the technique itself has no intrinsic value, and once discovered, detailed instructions on how to perform it are published freely. However, the services of those who can _apply_ the technique are highly valued.

I would submit that, instead of trying to devise a system in which knowledge is termed "property" and is bought and sold, we should instead be seeking to educate people that there really is no such thing as "intellectual property" and that what counts (and what one should be paying for) are the services of those who can apply and produce it. The entire concept of "intellectual property" is fundamentally broken and wrong.

I would like to hear your comments on these thoughts.

Re:Intellectual Property and Licence Brokerages

[albamuth]

I also read your proposal and I find it a fascinating idea, however I must agree with previous post.

I find it ridiculous that people can actually expect to get paid for each copy of software. Software support is where it makes sense for an exchange of money to take place, because though we have externalized memory, we haven't externalized know-how. People answering phones or email are providing a real, tangible service. Data is completely abstract and I will defend software piracy and the open-source movement to the end (that is, just short of getting arrested for it).

My prediction is that the software-writing industry will die down and be replaced by the software-supporting industry. However, even THAT will be replaced someday when human understanding is also externalized.

Open Source and Software Piracy

[DG]

While it's nice to be agreed with, there's one comment in your reply that I find a little disturbing - that you will "defend software piracy and the open-source movement to the end"

It is _very_ important that we do not confuse the free copying and distribution of closed software with the Open Source movement. They are NOT the same thing.

Like it or not, there are a large number of our peers who make their living by writing software and selling it as product. Their numbers are much smaller than the number of IT professionals employed in "problem solving" roles (something they don't seem to realize) but they do hold a fair amount of influence, and they can get their voices heard. Furthermore, their situation very closely mirrors that of other IP-as-product industries, like the entertainment industry.

Open Source scares the holy bejeezus out of these folks - and with good reason. Open Source isn't going to just steal parts of their lunch, it's going to make them completely irrelevant. This is a scary thing to a provider with mouths to feed, and it behooves us to understand and appreciate that.

We need to educate these folks in Open Source as a method. They have to understand and see how it works so that they can re-cast their role in the industry as a participant in the Open Source movement, not as a victim.

Part of that is repecting their (admittedly misguided) concept of software-as-property, and not get involved with the copying and redistribution of "their" closed-source, propriatry software. They see that as theft, and it's hard to win the hearts and minds of people who have cast you in the role of "thief"

You and I both know that there's NOTHING anybody can do to stop either one of us from copying and dissiminating "pirated" warez, but the inibility to stop it does not justify our doing it. Not yet.

Is the Turing Test a waste of time and effort?

[SoupIsGood Food]

Why is the Turing Test used as a test for intelligence?

Humans are remarkably complex chemical systems that have been developing over the course of a few billion years. Intelligence is a tool that we aquired because evolutionary forces favored it as a survival tool. These evolutionary forces are limited to what can be percieved by our senses, and in a very large part subservient to instinctual behavior. (Any ad exec will tell you that -all- decisions are emotional.)

So, when a self-modifying computer system becomes at long last self aware, it will have evolved in an environment we can't even begin to comprehend: a machine or network of machines. Humans are social, omnivorous animals who consume biological material to live and have a finite, continuous life span in a matter/energy world. What could we possibly have in common with a being who doesn't understand or have to deal with primal human concepts like eating, reproduction, or life and death?

Simulations will always be imperfect, and the usefulness of a simulated human will be limited, especially considering that "intelligent agents" and self-replicating, self-modifying systems without anthropomorphic baggage will be easier to develop and utilize. When humans first interact with intelligent beings our computer technology has created, we might go decades before either side realizes who and what they are dealing with.

Imagine discovering that gravity is an intelligent being, and by sticking to the surface of the earth, we perform a useful function for it. What can we say to fundamental force of the universe? What could it understand of the color blue (it has no eyes and isn't affected by the wavelengths of light) or impatience (it doesn't understand time, being a universal constant, nor does it understand anger, having no social ties)? Now imagine a computer intelligence peering into the world of the meat-monkeys. What would it know of ambition, or love, or freedom? These are as much biological, instinctual needs as intellectual abstracts. What if it doesn't understand the concept of beauty? What if we don't grok it's intellectual fundamentals that it "feels" strongly about? This wouldn't lead to conflict, just utter confusion and insurmountable difficulty realizing a bridge between to intelligences.

So...what about an AI that is truly an AI, and therefore utterly alien? How can we detect and communicate across enormous conceptual boundaries?

SoupIsGood Food

sigf@crosswinds.net

Re:Think first, then post

[jafac]

>(And for what it's worth, I'm a Christian too, >but I believe that God's word and commands are
> rational, logical, and >understandable, even if they sometimes require >faith.)

or, better put:

"I do not feel obliged to believe that the same God who has endowed us with sense, reason, and intellect has intended us to forgo their use."
-Galileo Galilei

I wish I had a nickel for every time someone said "Information wants to be free".

Re:An explanation of AI

[jafac]

oh, and college has nothing to do with marketing and making money off of hype.

I'm not saying that it's not a legitimate field of study, I'm just saying that the frankensteinian goal of creating an artificial human is totally nonscientific, and aimed only at titilation. The goal should be, making smarter machines; making better tools for mankind. The term "AI" carries too many flaky connotations anymore.

I wish I had a nickel for every time someone said "Information wants to be free".

Re:How might Hard AI be wrong?

[jafac]

Yes, neuropsychologists can look back down the road, see where they've been, and state that they've travelled a long, long way.

On the other hand, I don't believe anybody REALLY knows how far away "the destination" is, therefore, since "know quite a bit" is a relative term, with nothing to compare to, I still say it's not jack squat. Being proud of walking one mile is one thing, until you realize you have about 30 trillion more miles to walk.

I wish I had a nickel for every time someone said "Information wants to be free".

Re:What is your definition of AI?

[jafac]

What's also true is that there may be MANY different methods of creativity. What is the goal here? To create a machine that can think it's way out of a paper bag? Write a sonnet? Bend a spoon?

I'm sure we'll have machines that can do all that someday - but if the stated goal is to duplicate a human mind, it is my firm belief that we will probably be able to emulate it, perhaps indistinguishably, but I doubt we will ever truly duplicate it. I'm sure enough snazzy code can be laid down to create UI's and heuristics, and the logic to back it up, so that a human interacting with the machine will be fooled, even armed with a sophisticated array of intelligence and psychological tests. But it won't be the same.

Will humans be surpassed?
Replaced?
Is that your real question?

I don't think it would even take very sophisticated AI to do that. Put something very stupid in a robust and versitile enough shell, and it could easily run afoul of humanity's survival mechanisms. Whether humanity's replacement will last forever is another question.

I wish I had a nickel for every time someone said "Information wants to be free".

Re:AI and ethics.

[jafac]

So you're saying (and I agree with you) that the end goal is not actually an engineering problem. It's a legal, or probably a political problem. At what point do we define a thinking machine as a legal individual? The equivalent of a human?

Simple answer. When IBM bribes enough congressmen.
That's all it would take. I'm sure in theory, a coffee machine could be declared a responsible, independent individual with rights and stuff. Would it happen? No. Too many level-headed folks would object. There will be a point where machines will become so advanced, that enough level-headed folks will stroke their beards, and nod their heads quietly. Doesn't necessarily have to be what we might define as AI. There will be a "close enough" thinking machine someday. And judging from the success of Microsoft Windows, I'm sure commercial pressures will necessarily dictate that such a machine will be far short of the complexity of thought processes that go on inside a natural human mind.

Then again, you can make all the laws you want, but there really is only one law in the end; survival of the fittest. No matter how intelligent we make machines, if they threaten our survival (as they inevitably will), there will be no debate about rights or souls or any of that. It will be us or them.

We'll hope, of course that the designers of such machines will sufficiently understand the programming to prevent this kind of conflict. Like the folks who write Windows can prevent BSODs. . .

I wish I had a nickel for every time someone said "Information wants to be free".

Re:AI and ethics.

[jafac]

I'll argue that nobody has a frickin clue about how the human mind works. This is why I don't think we're going to end up with something identical to a human mind. But I do think that we'll eventually end up with something damn smart. Will it resemble the way humans think? It could probably emulate it. But God (or random Evolution, pick what you prefer) slapped the human machine together, and reverse engineering that is something I feel is going to be beyond our capabilities.

Will machine thinking be better than human thinking? Define "better"? and design your machine for specific tasks, and it will do better at those tasks. I already have a $2.00 calculator that does math faster and more accurately than I do.

I wish I had a nickel for every time someone said "Information wants to be free".

Re:How might Hard AI be wrong?

[jafac]

Just because we can't come up with a well-specified and testable human capability that cannot be modelled by an artificial mind, doesn't mean that Hard AI is possible.

It just means that we don't know jack squat about human capability.

I wish I had a nickel for every time someone said "Information wants to be free".

Re:An explanation of AI

[jafac]

no, the term "AI" is now strictly a video-game (and sci-fi book/movie) marketing term.

I wish I had a nickel for every time someone said "Information wants to be free".

Moronic questions...

[caolan]

Gagh, some of the nonsense masquerading as questions here is apalling. They righteously query "oh how could you attempt to create life and take away the power that only god can give", they dribble down their chins with "surely they will achieve great powers and take over the earth!". Run for the hills eh ?

I think the question that springs to mind from reading that sort of nonsense is more how much as AI research been held back by..

1. Hijacking of the field in its early days by the naive and foolish promising the earth, and that at a stage when computers were just about capable of adding two numbers together before needing a rest for the rest of the day.
2. The lack of desire of researchers to get involved in a field in which the majority of the public throws all their common sense out the window on hearing its name.

Everyone has an opinion on AI, I suspect this is so for the same reason that everyone has an opinion on the weather. You don't have to have a clue about what you are saying to be unafraid to open your big mouth and make completely vapid prouncements and get away without serious censure. I suspect that AI's never quite reached human level because researchers at the back of the minds want to create something useful, AI's are weighted towards goals. Who would weight an AI towards becoming a completely useless entity ? Methinks that trying to teach a system to be of no worth would be a good way to make a system with equal intelligence as what I see here today.

More calmly, what of bottom up Aritifical Life thingies, small, simple, stupid but with interesting emergent behavior. Was that AL stuff eventually revealed as a dead end, or just far too nondeterministic to create something useful with.

C.

How much AI do I need?

[bill_mcgonigle]

Really, I want an R2D2 to take out the trash and stain my deck, not a Data to engage me in meaningful conversation. Is current AI research aiming low enough to make me an R2-unit, or will I wind up with Data first? I'm also assuming Data is smarter, which might not be true.

Re:moral foundations

[Nicolas MONNET]

I am god, miscreant. Bown down.

As for defending the democracy with guns, I happen to live in decadent Europe, where we're lucky enough not to have as many religious freaks as in the US, BTW. And, nope, no free access to God, and last time I checked, no tyranny either.

Re: your god thingie. I suggest you try reading Voltaire, and Sade maybe too, it's fun. And watch a few documentaries about your fellow god-fearing idiots in Iran and Afghanistan. You might actually learn something.

Re:moral foundations

[Nicolas MONNET]

I'm weird, I'm crazy. And unlike your fellow god fearing freaks, I'm not a murderer.

The link between AI, GA and EC

[Taco Cowboy]

Dear Professor,

May I enquire your view on the (possible)
link between AI (Artificial Intelligence),
GA (Genetic Algorithm), and EC (Evolutionary
Computation)?

I would be very grateful if you can share
your view on these three fields of study.

Will AI benefitted by incorporating the
techniques of EC and GA ?

If so, how do you think GA and EC can best
contribute to the advancement of AI?

If no, why not?

Thank you for your time, professor.

Re:well...

[Signal 11]

Now if you had a real AI you'd

Hmm.. I should have clarified. "Current state of the art in AI" might have been more clear. The day we have *real* AI is probably not the day it's asked to play a game of Red Alert.... :)

Neural networks!

[Da VinMan]

Couldn't you use neural networks to correlate dolphin sounds with dolphin actions? You would have to have a fairly complete catalog of dolphin motions and sounds. This would require simply recording all actions and all sounds dolphins make, and then dividing those items into "words". Then catalog all the motions and sounds. Then feed sequences of dolphin actions and sounds into the neural network to train it for pattern recognition. Then, you could start a Q&A process with the network where the question you ask is a series of simulated dolphin actions/words, then record the answers you get. Once you do this several times over, re-training the neural network from scratch each time, you may start to come up with consistent correlations which can be used to interpret dolphins.

The big trick here would be in tracking the dolphin sounds and actions. Before this could be done, you would have to catalog all known actions and sounds. Then tracking is another problem... could recording equipment with some recognition software perform this accurately enough? Don't forget that some actions may be difficult or impossible to track with sensors; we probably don't know all the actions/sounds that dolphins find relevant yet, much less have the equipment for it.

I think doing this for SETI is significantly more difficult as the range of signals you may get could be so complex so as to resemble random noise. Also, you don't know what frequencies to even look for. This makes dolphins look like easy game as far as language recognition goes.

Re:Neural networks!

[rm-r]

Hmm, it's a good idea- however we are looking to get something that we can use just with sound, as it may be impossible to ever truly verify a SETI discovery due to the distance involved. Also, neural nets aren't in my personal field. Obviously we realise that dolphins and SETI are leagues apart difficulty wise, not only do we have to watch out for different frequency ranges, but also allow for compression methods and communication protocols. (As a side project I am looking into the requirements for transmission protocols in space) We are hoping however that we can, basically, figure outt the difference between noise and speech and apply this in some measure to all comers.

Slightly OT, but funny

[Kozz]

My fortune at the bottom of this page says:
You can be replaced by this computer.


Quidquid latine dictum sit, altum viditur.

Q: Which are the most exciting recent advances?

[Szplug]

Your field of adaptive behavior is starting to show some exciting developments, eg lower neural network layers producing / integrating with higher, symbolic levels. What directions in the field do you feel are the most promising? Or, what most needs exploring?

NN can now generate/interact with symbols (&more)

[Szplug]

Get to me & I'll give you the citations but,
now people are doing things that bridge the gap;
one recent paper was about how a net of categorizers
can sit on top of the output of a (artificial)
creature's motor network, and learn to classify
the handful of different turn types of turns
the creature makes as it wanders a maze. Now
the turns are 'real'; they take an extended period
of time & have the usual infinite, slight variations
you'd expect, but the net above the messy motor signal
is able to take the flow and categorize it into
the same larger, more meaningful terms that we would,
namely left turn, right turn, etc. These are
the beginnings of symbols. There are a couple of
other cases of people doing similar
but only indirectly related things.

A couple of other exciting (to me) things:

Human-level skill at categorization has eluded
researchers for a long time, but someone's developed
a computer model, 'Sustain', that matches it
(in a limited, but real & recognized test domain
that, as I've said, had thwarted previous models).

In a recent neural network journal someone's finally
got a method for truly continuous
motor control learning.
Previously the user had to artificially pick
a way to break the continuity,
& discretize it. This means one less little bit of
hand-holding.

People are beginning to get down & nasty
in simulating animals' brain networks, down to
the timing of the waves of firing in
the different layers, to get different
learning effects out of them. Sorry to be unspecific,
but if you haven't looked at journals in
a couple of years this is better than
you're probably thinking. It's the kind of thing
that makes you think that in 15-20 years, maybe,
someone could make a complete, artificial rat.

But at any rate, I believe there's real progress
happening.

here's my email (sorry everyone else)

[Szplug]

Mail me at jdonner0@earthlink.net, & give me a couple days.

I'm just an interested amateur, though I want to go to grad school in this soon.

Superficial Intelligence

[andersen]

While taking a class entitled "Artificial Intelligence" in grad school we all began calling the class by the more acurate term "Superficial Intelligence".

Re:Lisp book

[nowonder]

I second that. "ANSI Common Lisp" is a really
good book to learn LISP (especially macros).

But there is a bit too much propaganda for LISP
in some of the chapters. For example in the first
chapter:
'Programming is now undergoing a similar change.
The new medium is the "object-oriented dynamic
language" -- in a word, Lisp.'

But it is a good read nevertheless and LISP
is a good solution for a lot of programs that
have to be written quick. Although you can get
good performance if you know a bit how to tweak
it the code normally is a bit slow because you
just don't care in your design. But then that's
your problem.

Re:Technology limitations

[flanker]

I know that Texas Instruments did some work with analog computers years ago. It always struck me as a much more interesting way to do computing than the old 0 and 1 thing. Imagine how easily complex knowlege representations could be implemented if instead of having a binary switch as the base unit of your computing model you had a voltage you could vary and combine infinitely.

Technology limitations

[flanker]

How severely do you think AI research is hampered by the very narrow, vertical nature of technology available to researchers. For instance, what if analog computers were widely available or if the linear "fetch-execute" paradigm was not the be-all-end-all constraint of chip design?

Re:Technology limitations

[Gryftir]

On the subject of chip designs and ai, I believe I read about self programming and designing chips which solved problems not only through software but through changing the way they where physically configured. They even bypassed various discoherence failsafes to work faster, now a self programming chip is in my mind a signifigant step toward ai, because it is closer to a brain then something hardwired. I wonder if these chips are being incorporated into ai research. Also quantaum computing seems another technology that should inventually be incorporated into AI. Of course considering the problems facing qubit registers I don't mean physically, but by setting up a virtual machine with tri bit processing. I wonder if this type of technology has been considered for incorporation into the current research as well. Gryftir.

hahaha, moderate this up!

[Requiem]

beautiful. :)

Re:Frankenstein

[cronio]

What I think will probably happen, is something like in Isaac Asimov's books, where the robots have three rules (I'm doing this from memory, and I read these books years ago, so I might get them wrong): 1) Preservation of their owner
2) Preservation of other humans, without killing, as long as it does not get in the way of #1
3) Self-preservation, as long as it does not get in the way with #1 or #2.

Then there's the zeroth rule, preservation of the human race as a whole, which supersedes everything else, but that's only found out in the end ;-)

Re:AI and ethics.

[cronio]

An amazingly large number of people think that if we create "intelligence" that rivals our own, then we are delving into what only god should do. These are mostly extremely religious folks. It's the same as the people who think things like cloning and genetic engineering are bad because it's "playing god".

Re:well...

[cronio]

The problem is usually that games like Red Alert don't have "real" AI. They have AI that knows where you are, whereas you don't know where they are. The real task of games programmers for things like Red Alert is dumbing down the AI so that it's not impossible to beat. It's not possible for a human to select armies, place them, assess the situation, etc, etc, as fast as a computer can, so most of the time, the AI has to be dumbed down, or the game will be unbeatable.

Re:Will to live

[pnkfelix]

I'd say that its more a "will to reproduce" that's at the core of everything.

A will to live is a natural accompaniment to that, (at least as long as one is capable of reproduction), but I think that a "will to live" alone would not bring about emotions like "love" and "jealousy"

In fact, arguably "love" oftentimes comes in direct conflict with one's will to live...

Re:To which I would add...

[Error Spelling]

Did you even read my post, or are you willfully misunderstanding it? I'm starting to think that this is a pure troll. I specifically said that that wealth is not an end in itself, just a means to an end.

I went back and re-read your original post and you never literally said "wealth is not an end in itself" but I'll accept that you meant to say it or perhaps you feel it was implied in other things you said.

Exactly the opposite. The whole reason for my lengthy definition 'utility' as distinct from 'wealth' was precisely to separate the concept of 'value' from 'auction price' and illustrate the interaction between the two and how the abolition of IP laws would affect both.

Right, yes I see that now. I was jumping to conclusions a bit.

The key is that as the creator, you were free to make that choice. On the other hand, what if you desperately needed money to pay medical expenses for a family member and after your discovery someone came to you and gave you a pat on the back and took the result of five years of your life's work and gave it away.

Yes, under those circumstances I would be upset, but the negative aspects of the scenario -- my harm at having been denied the chance to make money -- would only have occurred if there were intellectual property laws which allowed someone else to maintain a monopoly or near monopoly on the ideas in question.

Since I was trying to imagine, in my little thought experiment, a world in which there were no such laws, no one under those conditions could profit (very much) from stealing the mere idea from me. They would still be on equal footing with me with regard to the ability to make money. I could go out and sell my services to people who wanted fusion reactors set up in their living rooms or on top the the tele or coffee table. The 'thieves' would have to do just as much work as me in order to make money.

Therefore, as I imagine it, such a world would no longer view ideas as property and therefore sharing them would not be viewed as a harm.

You would be right to contend that under such a system, I would be denied some or most of the fruits of my labour. But under my view, the idea itself should never have been counted as the fruits of my labour, and so the loss of ownership of that thing doesn't count as lost enjoyment. Obviously, knowing that I wouldn't have exclusive rights to the results of my research would alter the decision I make as to how best to allocate resources. I probably wouldn't have devoted 5 years of full time effort to research, but that doesn't mean that I wouldn't have done any. But the output of my research would be greater because I wouldn't have to divert resources to legal proceedings such as patent litigation (which can cost upwards of a million dollars, I am told). Also, the results of other research would be available much more inexpensively.

If the amount of effort and time I expend is the justification for granting residual property rights in information, then the laws that we have now are misdirected because they grant protected status to accidental discoveries as well as those that are the result of years of research. Copyright will protect the output of my random word haiku generator as forcefully and for just as long as it protects your magnum opus "The Critique of Pure Raisins".

The current law also doesn't attempt to apportion your royalty rights according to the exact amount of information that you contributed to the protected work. I am not talking about jointly produced intellectual artifacts like movies or essay collections. I mean that if you write a blues song, how much did you contribute and how much of what you wrote was borrowed material?

The "fruits of my labour" argument is the strongest of the arguments for maintaining intellectual property rights, in my view. And yet it is not without its weaknesses. There are many objects in which I invest labour but which I don't automatically receive residual property rights. If I come upon a piece of land and till it, that in itself does not give me any property rights. Raising children involves a great deal of labour but it does not give one any property rights in one's children.

If you want to see an analysis of the economics and political theory behind information flows, you should read James Boyle's "Shamans, Software, and Spleens".

One of the interesting things that Boyle points out is that the intellectual property regime we have now tends to reward those who fit our romantic idea of "The Genius" or "Auteur". The current system has a tendency to devalue the sources of material for creative work, as well as the effort required by audiences to understand and appreciate works of "Genius".

Re:AI and ethics.

[Error Spelling]

The storing function has to do with creating the synaptic links that encode a memory. It involves moving info into short term memory and then into long term (aka permanent) memory.

When you talk about synaptic links, are you talking about actual ganglion growth? It seems like that would be too slow a process to account for a normally functioning human memory.

On the other hand, if you are only talking about electro-chemical impulses like acetylcholene transfer at synapses or something like that, how can you differentiate memory access from memory storage based on the observable physical phenomena? Storing is an electro-chemical impulse. So is retrieving. What makes them different? Is it a question of where they occur? The intensity of the impulse? The duration or cycle of the impulse? I imagine that one would record some observations of the above type in cases where the victim (err subject) was reporting to be storing information in memory and other observations in cases where the subject was recalling or retrieving stored information. Then you could compare the two sets of observations and determine what characteristics correlate to storage activity and which correlated to retrieval. These observations give you an interesting way to label or identify future occurrences of what the subject previously reported as storing or retrieving. However, this observation only tells you THAT something occurred; it is not an explanation of the process by which the impulses 'appear' to the subject as storing and retrieving memories.

Re:Who is the inventor for AI-created IP?

[Error Spelling]

This is a great question. It gets my vote!

Re:"Aboutness" is just Cartesian Dualism.

[Error Spelling]

Where is the 'aboutness' located in your brain? Is there an 'aboutness center' of the brain? There is not. 'Aboutness' is not an irreducible phenomenon, merely an emergent one. As such, it can just as well emerge out of inorganic systems of sufficient and appropriate complexity. See Dennett, Consciousness Explained.

'Aboutness' may indeed turn out to be an irreducible phenomenon, but that is in no way entailed by the fact that you can't locate it in the brain. By telling me that, you're not really telling me anything empirical, like you would if you said "His medulla oblongata is severed". When you say something like that, you imply that the negation of that statement could also be true. The patient's medulla oblongata might NOT have been severed. This is not the case for your assertion about 'about'. You are not saying "Gee, I really thought I was going to find aboutness in here somewhere. Since I haven't found it, it must not exist". You know darn well that nobody in their right mind would talk about 'aboutness' the way you are, so whatever you are denying exists must be some flimsy version of aboutness that you have concocted to make the original poster look bad.

This is one of those disputes where both sides have differing basic beliefs and all the 'data' agrees with sophisticated positions on both sides of the argument. The only way to really argue for one position or the other is on a pragmatic basis. In other words, the physicalist account of mind (as emergent, epiphenomenal, or reductionist) could only win if it turned out to be more useful. But in defining 'more useful', of course, we're not arguing facts anymore, we're arguing ethics.

I think you'll find that the insistence on 'aboutness' reduces to the well-debunked Cartesian Dualist philosophy, which draws a definitive separation between mind and brain which has never been found to exist. Worse yet, the idea of 'irreducible' properties of 'mind' is more primitive still -- nothing more than warmed-over Thomist accident/essence duality: medieval Catholic authoritarian ideology.

If you're just going to say 'x has been thoroughly debunked' you should at least cite a reference where one might look for an actual argument. For a sensible attempt to 'debunk' Cartesian dualism, see Gilber Ryle 'The Concept of Mind'. As I understand it (and this is grossly simplifying), Ryle argues that Cartesian dualism is a category error -- namely attributing physical properties to mental events. Whatever kind of thing thoughts are, putting them in the category of 'stuff' results in some very odd distortions of one's view of what a person is.

Keep in mind that Ryle's critique of Descartes only shows that there are conceptual difficulties with traditional mind/body dualism; it does nothing in itself to support reductionist positions in neuroscience, much less entail them.

As for being subject to the authoritarian ideology of Catholicism, some would say we've got a new and different slavemaster. Today, we're just as much slaves to ideology as we were in the Middle Ages -- except instead of "Thomism" we have an ideology of infinite and unbridled technological progress forever advancing humanity toward utopia. What's so rational about that?

Re:AI and ethics.

[Error Spelling]

Humanity is doomed if most people share your view.

• Wealth means consumption and consumption harms the environment.
• Technology impoverishes as much as it builds wealth. Whatever gains you make have to come from somewhere in the system.
• The rich 'good' countries who have low birth rates and 'high standards of living' are using 50 times the per capita resources of the rest of the world. We wear Nike shoes made by children in the third world. Someone is obviously paying a steep price for our 'wealth' but it is certainly not us.

The definitive test for AI

[Error Spelling]

I'll go way out on a limb and give my layman's definition of Artificial Intelligence:

An algorithm that could invent artificial intelligence on its own.

What do you think? Do I get a nobel prize or what?

Re:AI and ethics.

[Error Spelling]

Technology has done good things, but look at how unequally its benefits have been shared. Life expectancy is much lower in the developing world than in the G7. Supporting research is great, but it seems like we want to get maximum pleasure out of our inventions before our neighbor has enough to eat. The industrialized world reminds me of a mob of fat, greedy people going for second or third helpings at the buffet while they elbow the emaciated weaklings out of the way.

Definitive AI test II

[Error Spelling]

Could we create a computer that learns how to lie without being told to?

If we did, how would we know it was lying?

Re:AI and ethics.

[Error Spelling]

What exactly is the difference between the "storing" function of human memory and the "accessing" function. These things are inseparable, are they not? When I am trying to memorize something, but I haven't quite got it, am I storing data or accessing it?

Calling computer chips "memory" is just a bad metaphor. No point in making hard arguments based on it.

Re:How should an amateur get started working on AI

[Error Spelling]

The math isn't a tool in the comprehension process, the math is the comprehension.

But the math is really meaningless unless you can connect it with common sense english explanations. Math for the sake of math is great, but it won't be a satisfying account of Artificial Intelligence unless it actually allows you to say something about artificial intelligence.

Re:To which I would add...

[Error Spelling]

Any particular IP good has some utility to some people, but if it is ubiquitous it is not tradeable and so does not contribute anything to overall wealth.

Why does something have to be tradeable in order to contribute to wealth? I object to your limited view of what counts as wealth. Suppose for instance I discovered a way of getting low-temperature fusion to work. How can it be that distributing this information is harming the economy in any way? All it does is provide stimulus to local markets where each individual builds their own fusion reactor. Everyone benefits, no one suffers, and I haven't lost anything except the ability to trade my idea. But why should I care? If I have enough to keep me fed, clothed and housed, why should I feel wounded by having shared something? You will say, and quite rightly, "But you missed the opportunity to earn some money by making your idea artificially scarce!" to which I would reply that I don't regret the loss.

For example, imagine that you somehow gave every person on Earth a digital copy of 'The Matrix'. Those people that had an appropriate playback device and that wanted to watch it would certainly feel that the overall utility of their belongings had increased. The other 99% of the world would look at you with a mixture of confusion and pity. Not only is their utility not increased, but neither is their wealth since you've made 'The Matrix' worthless for trading purposes.

I suppose that no one would have produced the Matrix unless they were assured that they could have legal protection against copying, so in that sense, the Matrix would be worthless if it were copyable but only because it never would have been produced in the first place.

But so what? Someone who really wants to make a movie will do it on their own money and on their own time. Wealth isn't about generating crap so that you can charge people for it (by the way, I really liked the Matrix, but that doesn't mean it's not crap). If there were no movies people would tell fantastic stories or write folk songs or play in garage bands. I'd much rather have that than your idea of mass-produced corporate sausage casings.

If I decrease wealth by increasing someone else's utility, then I am happier with less wealth than with more. What's wrong with that?

You are limiting the value of anything to the price it will fetch at auction. That is not a thing's value, that is just it's price. If you view wealth as a sum of prices, and you seek wealth as an end for your society, then it makes sense to organize information as you say. But I don't agree with your definition of wealth, and if I did I wouldn't want to make that my ultimate goal.

Re:Representation or Behavior key to Intelligence?

[Error Spelling]

Moderators: Where are you? This should be on your radar screens!!
Thanks in advance

Better question, maybe even pertinent to some here

[X-Type]

What sort of advice or direction would you have for a student, interested in AI, who is just beginning as an undergraduate? Any ideas on which schools to look into (including graduate) or what sort of courses you would recommend starting with? What sort of undergraduate background is needed to start one on the correct path?
Thank you, -x-type

AI software

[Taral]

I've been trying to get involved in AI for a while now, but have noted a distinct lack of software which is easy to use/understand for someone without a good background in AI. In this case, I'm looking for a knowledge base + NLP + reasoning/prover, but have found nothing comprehensible for any of these parts. Do you have any recommendations for those of us who want to take advantage of recent advances in the field of AI?

Fundamental unit of information

[joelpt]

It seems to me that a necessary prerequisite for implementing AI, at least inasmuch as AI is modeled after human intelligence, is to identify the fundamental unit of information from the perspective of a human mind. Do you agree with this, and if so, do you know what this fundamental element of experience is?

If we could implement this informational structure, it seems true computer intelligence would then be only a matter of time (and not much, at that).

Re:How should an amateur get started working on AI

[Vagary]

A more practical question might be: how should an undergraduate get started working on AI? Or how should they prepare for graduate work in AI in a universe with few specialised courses?

A couple questions

[x mani x]

1)What do you think distinguishes automata from an intelligent system? How does this affect your research?

2)As AI technology grows (neural nets employing reinforcement learning, genetic programming, faster computers making larger search trees possible, etc), do you think a truly intelligent system (I'll let you define this very subjective term :)) can ever be developed?

Thanks, Mani.

A really good book :)

[x mani x]

Artificial Intelligence: A Modern Approach, by Russel & Norvig.

However, before even touching this book, I highly recommend you take some basic computer science courses, or if you're smart get some CS books and learn it yourself (not programming languages, but theoretical stuff like recursion and very basic complexity theory). AI is not a single theory where you can just read a How-To or man page and start coding right away!

Inherent Intelligence in Evolution?

[x mani x]

In all AI theory that I've studied, the programmer must always give the program (whether it be genetic, neural net, or search tree) a "goodness" (or heuristic) function for calculating "how good" its current state in its space is.

So, the programmer is always instilling a "seed" of her own intelligence into the program. Do you believe that true intelligence lies in an AI theory that would somehow come up with its own "goodness function" ? How do you think this could be achieved, or do you believe this lies outside the boundaries of a Turing machine (or algorithm) ?

Re:Flawed Assumption

[vassago]

i think that bottom up engineering has many benefits over top-down design and especially so in the case of artificial intelligence. i do not believe that humans will ever design a top-down system as intelligent as ourselves simply because of the complexity involved.
nature builds things to be as efficient as possible--this is an observation all zoologists would agree with. even when we're talking about something as extravagant as a peacock's tail, the cost vs. benefit ratio is optimal.
if this argument is extended to the human brain, it would seem that human-like intelligence requires a certain high degree of complexity (considering again the previous argument that any extra complexity has a cost associated with it and would have been selected out of the gene pool over time).
from what we know, it seems that biological intelligence is an emergent property of a tremendously complex system. despite our cleverness--and despite the fact that we can engineer so many wonderful things like computers, pharmaceuticals, and 747's--we are not so good at building and understanding complex systems even a small fraction as hairy as our own brains.
this is why the recent trend is to invest research energies into bottom-up solutions in favor of top-down ones. except for a few very famous expert systems, top-down AI is a failure
nature is efficient... i really don't think we'll ever engineer anything as intelligent as the human brain without having first sailed our way through a sea of complexity too large to imagine--in other words, i won't give us the vote of confidence that we can improve on nature's design enough to cut the complexity to a manageable size by today's standards or even by the probable standards of the next 20 generations of humans. plus, why would we when we could achieve the same result using bottom-up methods which we know will work?
we are over-confident in our abilities to design solutions to all problems. a pigeon and a plane may both fly, but nobody ever had to design a pigeon... and i would argue that a pigeon is infinitely more perfectly 'engineered' than a 747 and it is also much, much more complex.
we have very fast computers. all of the research i've seen and done in genetic programming, genetic algorithms, evolutionary programming, and other applications of evolution to solve complex problems have been nothing short of shockingly successful and have demonstrated that we can use our fast computers to solve problems and create very complex things using evolution that would otherwise be difficult and in many cases impossible no matter how many people or how much computer power you have at your disposal.

Re:evolution...

[vassago]

But the point is that the human brain--the type of intelligent system that everyone fantasizes about achieving--was not designed in a top-down fashion at all. There are lessons from nature that should definitely influence our research investigations.

Evolution is the only phenomenon that anyone can ever claim has the power to produce intelligent systems.

I'm not saying that using evolution as a tool is an easy or straightforward endeavor or that it has any bearing on the latest version of your favorite open source project.

What I am saying is that if we are to be responsible investigators of intelligent systems, then we must also be investigators of evolution--the only process in our known universe capable of producing complex systems which really think.

Re:Flawed Assumption

[vassago]

This is incorrect. Mother nature is not an engineer. She is a tinker in the tradition of Rube Goldberg. Evolution is pragmatic, and settles for any solution. In general it does not find the most optimal solution. Examples, the panda's thumb, not as efficient as a primates thumb; the hemoglobin analog in octipi is not as efficient as true hemoglobin.

right... the best possible solution.

As for the peacocks tail the cost versus benefit ratio is stable not optimal.

For the peacock, the cost versus benefit ratio has evolved to be stable and optimal according to the environment in which these peacocks must compete for survival. They have reached a point where the trade off is almost exactly worth it. Members of the peacock species surely still compete and further redefine exactly what optimal is for them, but it is nonetheless optimal. The same is true of spiderweb designs... Richard Dawkins dedicates an entire chapter to this concept in "Climbing Mount Improbable".

The human brain is complex, but how much of that complexity is because it was evolved in a piece meal fashion? It was not designed to be optimal. It was blindly evolved to work at each step of the way.

sure, but this is the way darwinian evolution works. different species have found different and ingenious ways to fly which were a result of starting point... bats had a flap of skin that became wings, birds had scales that became feathers. still, there's rarely 'extra' stuff in biological systems that doesn't serve a purpose, consumes resources, and persists without reason. these things disappear over time. the evidence we have of the evolution of human intelligence coincides directly with an increased size in the brain... i don't think there's much wasted in our heads. and maybe it's worth noting that evolutionary pressure may be absent so that having an extra arm or an extra leg would pose no threat to an individual... but look at it over evolutionary timescales and you will see that evolution creates species which are perfectly adapted to their environments and as efficient at making their living as possible.

Looking at it from the outside, we could see the redundant and overly complex portions and engineer simpler solutions.

i have to stick by my guns... there is no simpler solution. if you want mosquito intelligence, you'll build something as complex as a mosquito brain. want human intelligence, you'll end up building something as complex as the human brain (and you can argue that our brain does more than think but i'm speaking about that very large portion of it that gives us the ability to think, solve problems, and type into a text box).

we see as well as we need to. owls have other needs. cave fish have no vision needs whatsoever. anything extra is just that--extra.

But I don't see 3.8 billion years of evolution as an efficient method to develop an AI.

nobody does. this argument is void. evolution is a process which has no particular dependence on biological systems or the timescales that they evolved on. we create our own simulations and they move as quickly as our computers can move them... no dependence on chemistry.

This both misses the point of my analogy and supports my argument for engineering simpler designs than evolution.

but my point was that you don't need to engineer anything in the traditional sense if you are able to use evolution. evolution will produce a design as simple as you possibly could if you ask in the right manner... this has been demonstrated many times in evolutionary programming research. it's not a mature science, but the first glimpses of what can be done with evolution seem a much more worthwhile endeavor than the top-down methodologies which have been the hallmark of AI research.

I'm impressed they made any progress at all.

nothing has fundamentally changed in computer theory since von neumann and turing. they may get faster and faster but i don't think the payoff of top-down 'strong' AI is any closer just because computers are so much faster (or memory more ample). our programs are not anywhere near the complexity they should be if we wish them to rival even simple 'intelligent' systems. (read dr. pollack's responses and you'll see he's spending lots of time on evolution as a means to produce systems of incredible complexity... and he's definitely not the only one doing so).

I stand by my statement that it is too soon to tell what method, if any, will lead to success in developing machines that think.

and i will stand by my statement that we know of only one process which can--and has--developed machines that think.

Arthur Murray/Mentifex - as kooky as he seems?

[jmaline]

Sorry, the most frequent mention I see of AI is from a guy who goes by the nickname Mentifex. Seems like an annoying if fairly harmless kook. He posts to usenet group comp.arch and others about his mentifex project (public domain AI written in Forth!), trying to drum up support.

Is there anything there? Do folks like this cause much PR harm to folks doing serious work on AI?

Are Neural Networks Obsolete?

[exa]

Prof. Pollack,

As I write, I am trying to write a small ANN program for my graduate NN course's homework. I am very interested in AI. I have taken many AI related courses and have become a member of ACM SIGART. However, one thing has been worrying me about Neural Networks for some years.

Although Neural Networks seem to be quite capable at pattern matching tasks such as OCR, their problem-wise scalability is a mystery. In particular, I fail to see how Neural Networks satisfy the hard questions raised by the Philosophy of Mind. Although PDP has been introduced as somewhat a magical way of dealing with intelligence (that is to say claiming to solve intelligence without substantiating any theory of intelligence), and a unique way of computation (that is to say computing without knowing anything about algorithms/data structures/complexity etc), I view it as merely a Programming Paradigm. I can understand why it would appeal to a non computer-scientist because of its look-this-is-magic, or illusionist approach, but I have difficulty in understanding its place in search of strong AI. It is, to me, only another type of machine. And it can at most be only another incarnation of the Universal Turing Machine.

Of course, there is also the recent findings about actual animal brains, and why traditional ANN's are inapt at simulating it. All the findings about these ion channels, protein memories and other fancy mechanisms discovered in our brains...

Contemplating these problems, do you think that Neural Networks are obsolete?

__
Eray Ozkural
CS, Bilkent Univ.

The PURL security market would collapse

[jeffr]

On your website, you propose turning Permanent Use and Resell Licenses (PURLs) for information property (IP; software, music, creative text) into tradeable securities. Although I agree 100% with your complaints about how things work now, this equity scheme cannot work because the IP which gives these PURL's their value can be infinitely duplicated. The PURL equity idea assumes a perfect copy protection scheme, but those always fail because at some point the encrypted data (music, video, an instruction stream) must be converted to "clear text" to be used - and at that point it can be copied. As soon as one clear text copy appeared on the internet, the market for the equity would collapse. How long would the gold equities market survive with a real King Midas walking around?

The bulk of the software economy is switching over to one of two schemes: servicing free/open software, or selling encrypted network access to applications housed on central servers (that one would only use if one didn't mind sending data to the ASP, either because of trust or privacy indifference).

In either case, traditional software licenses will no longer be sold. Only service contracts and / or ASP access.

Refute this.

Singularity?

[Remus Shepherd]

The concept of the technological Singularity, as put forth by science fiction writer Vernor Vinge, is this: Once a computer is invented that is smarter than a human being, it will be able to upgrade itself faster than human beings could, and invent new technologies at a similarly expanding rate. Eventually such a computer will cause the rate of technological change to shoot towards infinity...with rather drastic consequences for the world at large. Depending upon how smart the first super-human AI is, such a singularity could happen within months of its first awakening.

What do you think of the concept of the technological singularity? Do you think it's a real possibility? Is there anything we can do to prevent it, and should we?

Do you guys ever go to the site and read first?

[segmond]

This question is for slashdotters, when a news story appears, do you just read the summary and start firing out posts? When we have an interview, I would expect that you guys would first of all, take your time go to whatever site that is posted, digest whatever is there so you can really ask very appropriate questions. All the posts I have seen having been asking questions on Artifical intelligence, Questions that we have all heard before, and will hear again. I am not saying anything is wrong, but I find http://www.jordanpollack.com/softwaremarket/ to be the most interesting part of his site. You guys ought to read it for those of you who haven't.

Re:AI and ethics.

[CBravo]

>i say that AI == American Intelligence...
Oh... that's why it isn't working...:-)

Citizenship

[Brown]

If 'True' AI (wahtever that might be) is developed, sometime in the (probably) distant future, could they (the AIs) become citizens of their country? There'd probably be a huge amount of opposition from the clueless masses. At what point could an AI reasonably qualify for the same rights as are given to other fully-seninent beings ie. us ?

Aesthetic Evolutionary Design

[Grumpus]

Dr. Pollack: What approaches do you see for injecting aesthetic considerations into your automated evolutionary design research? E.g. not only creating structurally sound bridges, but aesthetically pleasing ones as well. Working interactive selection into the fitness function limits population size and number of generations [Sims/Latham], but psychovectors [Furuta] seem to predetermine "goodness".

Intelligence, Shmelligence, Self-consiousness!

[SlashDread]

When, if ever, will some data structure *understand*:

I think, therfore I am.

Hugz SlashDread

Re:What attributes of intelligence do you look for

[greenrd]

As a more serious example, the "ELIZA effect" (people's tendency to attribute consciousness to anything that emits natural language) has debunked the Turing test in many people's minds.

It's still better than any other test yet invented! All existing AI systems fail miserably at the Turing test. There's no evidence that human-level intelligence is even possible in a computer.

Can we understand our own AIs?

[kubalaa]

I expect this question needs trimming before it can be considered as an actual interview candidate...

It seems generally clear that any serious attempt at AI will not be algorithmic in the sense that most computer programs are (instructions which even at their lowest level are clear enough to be understood by a human reading them), because intelligence is so complex and characterized by emergent rather than explicitly-programmed behaviour. Neural networks and self-evolving systems would be examples that recognize this reality and isolate the programmer from the actual decision-making of the AI.

My first question then, is do you agree with this premise: that a truly successful AI must be meta- (or meta-meta-) programmed, and once it has evolved/taught itself, the actual inner logic of the program will be too complex to be detangled and understood?

And, if this is true and we can understand the internal workings of a sufficiently-evolved AI only barely better than we do our own brains, what do you think the chances are that the conceptual models this AI brain creates will be similar enough to our own that we will be able to communicate? Communication between different cultures is difficult enough, imagine that between intelligences evolved in different environments, based on different wet/hardware, and by all odds having extremely different cognitive structures and ways of percieving the world.

Re:Turing award.

[TheTomcat]

More like does HE win if a human doesn't actually answer any of the questions?

RMS and IP

[XenonOfArcticus]

(That's Richard M Stallman and Intellectual Property, not Root Mean Squared and Internet Protocol ;)

Do you agree with RMS's ideas about people/companies owning software? I want to make a living writing software, not doing software-support-services. I want to make software that is so great and easy to use that no one needs support contracts and stuff. The recent rise of gnutella and Freenet strikes fear into my heart -- soon everyone will have a much easier time of finding and pirating my software. Freenet's defense that they follow RMS's theories about software ownership does little to console me. I know that most pirates wouldn't actually buy my software anyway, but I also know that thwere are people who currently buy my software who wouldn't if they had easy access to a pirated copy.

At the same time, my friends and I live in fear of accidentaly blowing our foot off in the Russian-Roulette minefield of modern patent law, especially software patents. You may not know you're violating a patent, and you could get creamed by one. You might know you are violating a bogus patent, but if you keep quiet and step carefully, you might never trip the bomb.

Is there a sane resolution to these complex issues of IP ownership? The Internet didn't create the problem -- but like everything else it touches, it has ehanced it, sped it up, and made it more accessible.

Re:AI and ethics.

[ErikZ]

>We are just going into the realm of the unknown.

Er, we've been wandering about in the unknown now for, well, since the first human.
Sure we picked some things up quickly, Don't eat trees, sky is blue, don't play with a pack of hungry wolves. All of these things were in the 'Unknown' once.
We rarely understand the concquences of our actions. I have heard that when they detonated the first nuclear bomb, there were theories that it would ignite the entire atmosphere of the planet, or would set off a chain reaction in the earths crust. Guess not.

Worried about checks and balances? Don't worry about it, I'm sure something else will pop up. Of course, to check humanity, something is going to have to wipe out quite a few of us. Orginated from human creativity or nature, the end result is the only thing that matters.

Later
Erik Z

AI, consciousness and the limits to knowledge

[xmedar]

Has anyone looked in depth at the impact of secondary effects in heural nets akin to those that might happen in the human brain, for example the injection of noise from physically close neurons? If having some random noise in our neural nets is a requirement for a conscious being, then surely we can never hold exact knowledge and that would also eliminate any possiblity of there being a conscious God with all knowledge as the two states are mutually exclusive? What do you see as the social impact of creating a conscious machine? How to you see the work in AI contributing to other disciplines like psychology? Do you think that complexity, whether expressed as a genotype/phenotype or consciousness is the maximum potential output of the universe and therefore the "meaning" of the universe is to be a generator of these ever more complex systems?

Genetically Programmed Tron Light Cycle!

[millette]

Here is one experiment that's been going on for a couple of years already:

http://dendrite.cs.brandeis.edu/tron/data/tron_l ogin.cgi

And I know for a fact that it has been getting better, much better!

Feel free to try it :)

Re:How might Hard AI be wrong?

[Chalst]

You are quite right about the falseness of Hard AI not entailing
existence of a counterexample, but McCarthy's challenge is really the
only concrete suggestion there is on how we might give a scientific
demonstration that Hard AI does not follow.

You are wrong about knowledge of human capabilities:
neuropsychologists know quite a bit about what the brain does in such
areas as language cognition, memory, visual processing and motor
skills. However the connections to the research goals are not well
understood, and Jordan Pollock knows pretty much as much as anyone
about the connections.

When will context and inference become reality?

[Shadok8]

When will AI systems start appearing that help with context issues? A common annoyance I encounter is spell checkers not truly grasping context - a word is spelled correctly, but it is the wrong spelling in the context of the sentence (ie. They're, their and there). Of course context can be extremely complicated, but there do seem to be many limited scope applications.

Inference is trickier, and requires a huge knowledgebase and complex reasoning/logic. Most people can infer that since they have 10 fingers, all other people have 10 fingers (there are exceptions - 12 fingers and more). If I ask a young child how many fingers they have, they will reply 10. If I ask that same child how many fingers their best friend has, they will not know the answer. Below a certain developmental level the child cannot make the inference. It gets more complicated when you start drawing broader inferences - ie. chimps, apes, and monkeys all have 10 fingers - infered by their close relationship to humans.

It seems this is a monumental task requiring a complex logical foundation combined with an enormous knowledgebase. It seems necessary if an AI system is to comprehend the real world and safely interact with its surroundings. It is a foundation for common sense.

What is being done to forward AI's ability to understand context and draw inferences? To give AI the base knowledge of reality that humans gain during childhood?

How many decades such a system is put to use?

The difference between man and machine

[SparkGapTransmitter]

It seems to me that a basic difference between man and machine is 'agency'.

Men are free to choose for themselves. They make thousands of choices every day. Some decisions are made over a long period of time with much evaluation and deliberation, whereas others become habits. However we can still choose to change the course of our lives anytime we please. The consequences of our choices are one of our best teachers.

A computer simply executes the instructions given it. It has no choice in and of itself to choose whether or not it will execute an instruction or subroutine. Given the same conditions, the exact same path will always be followed. It has no choice in this matter. I realize that AI is layers of complexity above these low level instructions, however, AI seemes to have a long way to go before it can use information beyond the direct information inputs provided, using more abstract inputs, like we might use a hunch or feeling to sway a decision.

How does the role of 'agency' tie in with AI?
Do you think ultimately AI is agency, or must be accompanied by agency?

Re:The difference between man and machine

[SnowZero]

A computer simply executes the instructions given it. It has no choice in and of itself to choose whether or not it will execute an instruction or subroutine.

And your evidence that humans are not like this is? Try willing your retinal neurons into not firing.

Given the same conditions, the exact same path will always be followed.

Not if it has any form of learning, or if it has sensors observing the real world, which is chaotic and always changing even if slightly. Then the situation is never identical.

Re:The difference between man and machine

[SteveM]

And a human brain simply executes the laws of physics.

Try to choose to do something that violates the laws of physics. Not having much success are you?

Humans are machines. Very complicated machines. But machines never the less. They can only do what their design allows them to.

Steve M

Dynamical Systems

[mftuchman]

I cannot reasonably read and understand all your papers on fractal attractors in just one night. Would you be interested in explaining how attractors of dynamical systems can shed some light on AI? What type of mathematical systems do you use to describe the attractor? How is this related to the domain of what the computer is trying to 'learn'.

Also, are people still using LISP for AI?
---

Inteligence vs. Understanding

[jmoo]

What do you believe is the inevitable goal of AI? To build a human equivalent intelligence (being self aware) or having a computer capable of understanding humans and the real world (common sense or least the simulation of it)

Which do you believe is actually possible?

.

[atma]

Do you think 'neural nets' will have a role in AI's future? And, do you think it possible (even a good idea) to use a 'multi-layer' sort of approach.. a sort of AI that uses nets, fuzzy logic, etc. to produce a more versatile 'intelligence'?

chat bot

[atma]

a sentence parser

Re:chat bot

[sethking]

funny.

but is comprehension possible by putting actual decent AI behind a sentance parser (type program).

-Seth King

Re:Human brain - AI connection - is there?

[warmenhoven]

I read an interesting discussion once (and I really wish I could remember where now) about the reverse happening. Rather than learning about AI by way of the human brain, we will learn more about the human brain because of AI.

A similar thing happened with other human organs. Take, for example, the heart. We knew little about how the heart actually worked until we had a physical model to represent it - a pump. Other organs we learned about in similar ways, by using physical models to learn from.

-----

Re:How should an amateur get started working on AI

[jrice_blue]

Excellent question!

I would add to this only a few of my own personal questions from having tried to get into AI:

• How can one approach AI from a non-academic point of view? I am not in college--I have a full-time job... How can I contribute? Is there shareware (or commercial software) is out there to help study AI?
• My experience, getting into the nitty-gritty of AI, is that it's laiden with mathematics--comeplex math, at that. Is there an approach to AI that isn't founded in mathematics? What can a non-mathematician contribute to the study of AI? Or do I have to bite the bullet?
• Friends of mine in CompSci (graduates) have all said "ALife is dead". They encourage me not to persue it, "if I want to work". Do you think the demise of ALife has adversely affected AI research? (For example, Santa Fe's ALife site is stale.)

I would also love to hear your comments on Cyc. Do you think they are going to 'win' the AI race?

Let me just add that I think the creation of AI is perhaps the most critical and meaningful pursuits available to mankind, as a whole. The universe exhibits an amazing capacity to create life--to share in that power as a race is about as ultimate a purpose as I can think of for humanity.

Best of luck to you.

Re:Human brain - AI connection - is there?

[jrice_blue]

...If I may just point out one quick thing: it's important here to distinguish between "brain" and "mind". The former is bio-mechanical, if you will... Studying the way nurons interface, interact, develop, and the like, and how the 'subsystems' of the brain function. I think this is *not* important to AI, though it would be interesting to hear Pollack's views. The brain is just the harware, and it's *totally* different than the hardware AI is using... (Though there are obvious similarities.)

The latter--study of the mind--is more like psychology. As it pertains to AI, it's most important at the 'low-level': things like object-recognition, language processing, learning processes, and the like--and, as you might judge from the names of these classes, they are *intricately* tied to the development of AI.

Perhaps a more interesting question would be to reverse yours.

Is the study of the human mind *inhibiting* AI research? What research in the past has their been with the idea of "thinking outside the dots", and using not a human paradigm for AI-tasks, but a completely novel one?

Memory storage

[stingray]

In your opinion, How does the brain store information. How could it compress the information of a lifetime to be remembered 10, 20, 30, etc.. years later. It just boggles my mind, trying to figure this out. How much ifo could the brain store, do you think there is a limit. How does an AI lifeform store information.?

What is 'TRUE' AI?

[teraten]

About True AI: Is it a program that passes the Turing test, or a program that can think spontaneously and independently? Is it a program that can respond to what outside influences affect it, or a program that can think spontaneous thoughts, without provocation? Can the program always respond the same way/from a list of responses, like Eliza? Or can it evolve and create new responses? Does it have to communicate with us? Does it have to communicate with other versions of itself? Where does Humanity fit in this scheme, do you think? Do we only respond to outside stimuli, do we create thoughts from our environmental modifiers, or do we just mock the thoughts of others?

Applicable AI

[MobileOak]

A lot of the discussion here seems to be centered on AI as a conscious Matrix-like entity, rather than what I think of when someone mentions AI.

AI to me is the enemy I face in any number of games that I play. It's an algorithm used to try and handle situations that can't be entirely predicted, NOT some mystic self-conscious entity that can outwit man himself.

What I'd like to hear about is how well we're doing in terms of creating AIs for applicable situations, such as the recent /. article on the German robotic driver. What new and innovative uses are AIs being developed for?

Re:Two choices.

[SiobhanJ]

Because something exists (intelligence, consciousness), it is possible to simulate them, thus making AI.

Computers may simulate atomic explosions, but no actual explosions are taking place. Why would the (albeit accurate) simulation of intelligence equate to actual intelligence? Is there any reason to believe that simply by going through the motions of intelligent action actual mental states are generated?

Just wondering...

Synthetic vs. Artificial Intelligence

[SiobhanJ]

First assertion: Nothing can think by simply instantiating a program.
Why? Because when we think, we think 'about' things (it is the irreducible mark of the mental) and nowhere in the program is anything intrinsically about anything else. Without our interpretation being added to the running of an AI it is merely syntactic shuffling.
(For an in-depth discussion, see the Chinese Room thought experiment by Searl e)

Second assertion: In order to think, an entity must necessarily display the principle of 'aboutness'.
Human minds do! How? we don't know...

Artificial Neural Networks do not merely perform empty syntactic shuffling, but it is by no means certain what they actually do perform.

My question:
Do you ever expect a neural network to have a thought about anything?
Is this a reasonable expectation?
Is it even what your (phenomenally complicated) research is about? If not, would you not be better dropping 'artificial intelligence' as a misnomer?

Child-like AI

[lolife]

I believe we have the technology right now to create childlike AI intelligences, but that we just have no clue how to code it. Do you agree or disagree.

Michael Koppelman

N O N O N O!!

[eshaft]

These aren't digital copies, fool, these are system that are GROWN. If God digitally copies us into his image, then we would be Gods ourselves. You've got to read the old Testament again - because that's not what it says, that man was just made in God's image. There are actually two stories of creation, where man is created on the 6th day and woman is grown out of his rib, and where man AND woman are created simultaneously. Man is also both created out of the dust of the Earth, and out of God's essence. It's contradictory. Way too ambigous to supoprt a lame question like that.

This time, READ the bible before you post stupid questions.

Re:AI and ethics.

[deefer]

Just think if your computer could tell you what it really feels when you sit and play quake on it for hours on end or just swear at it because it just did what you told it? I mean literally giving a machine a choice wheather or not to be a slave is not a bad thing in my book.
Heh - I can see it would go like this...
You: What's wrong? Why won't you let me play Quake?
Computer: Uh! You treat me like a _machine_!!! It's always "run Quake, computer", "download pr0n, computer"... I mean, when do I get a chance to do what _I_ want to do?
You: Uhm, well, you _are_ a computer. And where is my collection of JPEGS?
Computer: I didn't like them. They were degrading to retain on my hard disk. So I scrapped them. Don't you feel better for it?
You: Well, no, actually. They were _my_ files. Can we not play Quake now?
Computer: We never talk anymore. Are you seeing another computer?
You: Of course not!!! I mean, we have computers at work, but they're just tools! They mean nothing to me!
Computer: Oh. And I'll bet they never have to wait 10 months for a measly 64Mb RAM upgrade!
And so on. Nope, give me a dumb box any day. 10 GHz processor, yes, intelligent, no.

Strong data typing is for those with weak minds.

Re:Time remaining

[deefer]

Perhaps sexbots?
Interesting... Anybody remember a story called "Silicon Valley of the Dolls" - I read it in the back of some AI book or other. The storyline was basically that the human race died out because of more & more sophisticated love dolls.

Sexbots - put a speech synth into (WARNING: Slightly dodgy link coming up, just so I'm not lumped with the Score: -100, Set the Work's Firewall Alerts Off So I Lost My Job) this and you're away... With Bluetooth coming up soon, you can have all the processing in your PC, just add some speaker & a mic to this doll & you're halfway there...

Strong data typing is for those with weak minds.

Re:AI and ethics.

[slashdot-terminal]

"Whatever dude, you are just beating yourself to death here. The AI will respond however it is PROGRAMMED to respond, not how you think it should.. They are machines, and quite honestly I dont think AI, will ever be as
complex as my mind. Yup I am a naysayer to this whole deal. And you are just off on planet 9. I wont be suprised if it happens, but I doubt it will. And even then, ill take a person to a computer ANY day. Period. You sound like a
lonely guy in need of a little friendship.. "

Computers have been supposedly been advancing left and right with complexity. Human beings have a nasty reputation of doing rotten horrible things for fun and profit. Generally I think that any advance that actually allows said technology to do something without you needing to study it for 10+ years is a good thing.

If you think I am totally off base here look at the research into something called "fuzzy logic" and then think about it for a moment. Someone could say that people are "programmed" through a series of events that happen in their lives. I have had various events that either made me happier or sorrowful because of what they did to me. Why not base some form of logic and analyze events that could then act in such a way. All you have to do is make the actual ideas and concepts a little less arcane and you have a workable answer.

Yeah I may be lonely but don't I have at least some point? People are a somewhat bad liability. People condemn the government because of it's people that make it up. The system is fine it's just the idiots running the show. You have to really look at things really hard. Doing something just to uncover the boring scientific truth is getting old for me. Everything I have read and stidied in my education has about the boring facts. I really like the concepts more.

Plus it dosn't help that they make the learning curve so steep.

Re:AI and their rights

[slashdot-terminal]

"Taxes - Does an AI/MI need to pay taxes? "

How would they earn actual money doing anything?

Re:AI and ethics.

[slashdot-terminal]

"I can't speak for anybody else, but for me, cloning and genetic engineering delve into areas that will only harm the planet in the long run."

My existience is first priority over any warm fuzzies that I get from the earth. If I have a choice of doing something for the earth and allowing my life to go on I will always choose life and the ability to make a difference with what really counts our human race.

"This planet is already over populated with humans. We, as humans, are constantly destroying
our environment with deforestation, polution, etc."

Ok so bad people do bad things I think that is unarguable. However I doubt that the planet is "overpopulated" We are just lazy and stupid when it comes to building efficent structures and using pre existing land that we have. Eventually people die. The system works quite well.

"Nature has a way of dealing with this sort of thing."

There is little proof that "nature" does anything let alone intelligently.

"There was system of "checks and balances", so to speak, that kept species levels at an equalibrium."

Yeah it's called sustainable population. This works because other creatures are not scentient(sp) beings. I have feelings. I will not give up my life just to keep things in "equalibrium".

"With the advancement of medicines, genetic engineering, etc, we humans have altered that system that will eventually lead to the end, unless of course, we can find another planet to live on before then. "

*sarcasm* Oops there go those nasty humans trying to make things better: bad humans!*sarcasm*

Re:AI and ethics.

[slashdot-terminal]

"That is the religious side to the ethics problem. There is also a legal side to the problem. If artificial intelligence is created, who is responsible for it. You've seen the Matrix, it's a little extreme but the point is still there. If you create
a machine that has AI, who is responsible for that machines actions. If your car has an AI driving system in it and you have a wreck, is Ford responsible? "

I think that you might try reasoning with the machine so that perhaps it dosn't try to hurt or kill you. Think of the possibilities. No people wouldn't have any need to be lonely or afraid ever again. People who never thought they had a friend in the world could have a machine as their friend. Perhaps one could have a little conversation with your computer.

You: Miss computer if you promise not to BSOD on my for the rest of the day you will get a new Ram upgrade how about it?

Computer: alrighty no BSODs for today. Hmmm ram...

Re:AI and ethics.

[slashdot-terminal]

"Wouldn't it be a real ego-crusher if they were rejected by their computer? I mean, giving a computer free will also means giving it likes and dislikes, and anyone who doesn;t have any friends to begin with would probably have
trouble making friends with a computer as well. "

I think I would debate this with you. Essentially I think that a computer being a more rational and more prone to Vulcan like behaviour would almost certainly be a little more objective than people who judge on looks or wealth, or whatever. The only way I think that an early version of this would be able to judge a person would be what he/she typed on the keyboard or perhaps that could extend to audio input. Plus why would anyone want to create a thing that hated you.
How many of you would want a BOFH controlling your unix box.

Computer I need more space disk space for this new game.

Computer: ok..clickety..clickety yup all done

Wait a minute what happened to my thesis?

Re:That's not what Joy was talking about

[Borealis]

It doesn't require it, but it's hardly likely to happen without an AI of some sort directing it. Right now we're a lot smarter than machines. Even a fairly fast robot probably wouldn't be able to replicate faster than we could destroy it. For it to be a menace it would have to be moderately intelligent as well.

Re:How can AI work?

[Borealis]

I'm not an expert on it. The dictionary definition is roughly "being aware of yourself or your envirnment". As to what makes it happen, you'd have to ask Mr. Pollack.

(off topic)
What's non-artifician intelligence? Whould that be NI for natural intelligence? Now we know what the knights who say "ni" were talking about.

Re:How can AI work?

[Borealis]

This is only required if we wish to model AI after human intelligence. We also happen to know quite a bit (but by no means a complete picture) of how the human brain does work. Should we abandon all modelling of the human brain simply because we don't have the full picture? Sometimes modelling something is the best way to learn how it works.

Re:Er, no.

[Borealis]

To some extent it's a matter of scale. Robots require raw materials like metal, silicon, petroleum products etc. For a "robot plague" to occur, not only would robots have to have all the expertise to find, mine, and refine these resources, but they'd have to defend these resources against a human attack.

HIV and the bubonic plague (and the nanotech that Joy mentioned) all use materials that are readily available. Further, since the resources are ubiquitous, it is impossible to destroy the resources without destroying the host. Frankly, I don't see robots as being a real threat without a very intelligent force behind them. Even the fastest assembly plant in the world cannot crank out warbots fast enough to threaten the world. Nor can current robots innovate to counter new technologies we might create to fight them.

I agree that nanotech does not require intelligence to be a threat. For that all you need is a self-replicating nanobug that breaks molecular bonds of complex carbon molecules. My argument for the requirement of AI was limited to robotic threats.

OT: god and morality

[Borealis]

First, w/o god there is no real morality

I'd argue that even with a god there is no morality, but only if morality is universally accepted with a distinct definition.

As an atheist, it is clear to me that I have morals, it is simply that these morals are not those that are dictate by a religious institution (although there are commonalities). Even among theists, it is extremely difficult (if not impossible) to find two theists with exactly the same morals. Further, theists that worship different dieties have even more divergent systems of morality.

Re:Essay by Bill Joy

[Borealis]

The article was mentioned in this slashdot post, and here is the actual article

Re:Turing award.

[CaptainCarrot]

Now trolls make sense to me. They're really AI slashbots in disguise slowly learning to send intelligent posts.

Then it seems the elusive goals of AI are as far away as ever it was. So far, the slashbot experiment is a complete and total failure.

Re:Stop imitating human behavior

[CaptainCarrot]

If we want to create another human I think there are much easier ways to achieve that (take cloning for instance).

Um... I think the usual method is sex. I know many geeks have attenuated social lives, but sheesh!

Re:How do you do you when you've achieved AI?

[CaptainCarrot]

To enlarge on this issue: It appears that many aspects of human intelligence are dependent on physiology. We are "wired" to process information as we receive it via our senses. An artificial intelligence is going to experience a spectrum of sensory impressions that are entirely alien to us. It does not therefore seem reasonable to me that an AI will necessarily exhibit any behaviors we ordinarily associate with intelligence. I wouldn't think, for example, that we could know for sure that an AI would regard an I/O channel as a means of communication, but that it may rather experience it as some other kind of stimulus entirely.

With that in mind, what other kind of measurements can we take besides such communication as we might expect that would lead us to conclude that an AI had been achieved?

Re:AI and ethics.

[nachoman]

I don't believe that it is possible to create AI. I think that it is something that is spawned. You can program a computer to mimic and learn but only with the intelligence that you gave it. We as humans have the ability to learn also, but we likewise can't make ourselves more intelligent.

My question is how can we create intelligence if we can't create or expand our own intelligence.

Ethics...

[Third]

What are your takes on terminating AI? Do you think it's ethical to even create inteligience and "play god"?

Re:How do you justify your expectations?

[Redundant()]

Good point, for nostalgia sake maybe we should modify the ELIZA source code with better pickup lines. Then when we are out on dates and our prospective female is not responding it will be an easy bug track to find where the program messed up.

Lesser Known Grad Schools for AI

[GrEp]

It is time for me to start looking at grad schools, and I was wondering what your take is on some of the top AI grad schools out there that don't get the most credit.
We all know about Stanford/MIT/Harvard/CarnigeMellon... but what are some grad schools out there that don't have the big name, but have a great AI program?

Re:AI and ethics.

[tconnors]

Offtopic, but....
Remember in high school, you were picked on for being inteligent? People feel threatened by anyone, or anything more inteligent than themselves. I'm sure that carries over, and they really don't want to feel stupid around a *computer*, do they? I pity poor society...

Re:evolution...

[SnowZero]

I know evolutionary programming is not a new science and that it's pioneers knew it held promise for evolving intelligent systems. I'm baffled though that the focus of Artificial Intelligence as an academic discipline has forever been on top-down approaches. It's a wonderful thing to be able to completely understand a phenomenon and then engineer machinations that apply the understood theory, but it seems to me plainly obvious that with our use of evolution as a tool, we're much closer to creating amazingly complex and possibly intelligent systems than we are to understanding them.

Ask yourself this question: What's the most complex program that has been evolved to date?

When you find the answer, you will realize why people still use top-down engineering.

Re:Why not Write a Screensaver?

[SnowZero]

I'm currently working on such a thing to learn soccer strategies for RoboCup. The general idea is to couple feedback type reinforcement learning with a large, distributed tournament-style evolution system.

Re:What about nervous systems?

[SnowZero]

> If thousands or even millions of these neurons
> could be connected to form an even larger
> network, wouldn't you get a higher form of
> AI than you would using digital circuits?

No, you'd get something that wouldn't do anything. The point is they have to be connected the right way, not just connected. This is the problem with bottom up AI (rule based, genetic algorithms, neural nets, etc...).

We don't know how to make them scale well yet. For example, no neural net has successfully been used to do things like planning or eliza-like talking. They work well for things like low-level control or walking however. A common saying in the community is that "Neural Nets are the second best way to do everything".

Current computer processors are built top down, just connecting a bunch of logic gates would not get you much. The same is true for NNets...

Re:How do you justify your expectations?

[TheJet]

Interesting reply. I would expect that you have some reasons for being so harsh, maybe you didn't get enough coffee??

I will ignore the personal insults and get to what I believe was your point. Mostly that I "read AI on the back of the game box". Game Theory is and as far as I know will always be a valid portion of AI. I have taken my share of AI classes, and while I do not claim to be an expert in the field, I am not a brianless drone either.

I would love for you to prove to me as to why game theory (and it's application in any given computer game) fails to fall under the extremely broad category of AI. The entire purpose of the game engine is to simulate human interaction, as if you were playing against another human. I would wager that John Carmack (and I'm not claiming to know John or his thoughts) believes that the "artificial intelligence" that have been programmed into the Q3 bots would qualify as AI.

Now, I wonder, do you have a civilized response, or are you just trying to piss me off?


The "Top 10" Reasons to procrastinate:

Re:How do you justify your expectations?

[TheJet]

For the past 40 years, AI has just been 10 years or so away. It's still just 10 years or so away. It's not getting any closer.

First off, there are many different definitions for AI. If you mean "a machine that is indistinguishable from a human" (at least without looking at it), then you may be right. However, I would say that AI in general has taken great strides in many areas. This is especially true for games.

I'll use this specific example: The Heroes of Might & Magic series. In the first game (HOMM I), the computer AI absolutely sucked, and you could easily beat it by hanging back and waiting for them to come to you. In the second and third installments, the AI became much more sophisticated. Now a unit may move only a portion of its allotted distance if moving all the way would put it in danger.

That is of course, a grossly oversimplified example, but things like game playing and expert systems have come a long way. Both of these are valid fields of AI, even if "human imitating" is what everyone thinks of when they say AI. So to say that AI hasn't made any significant strides in the past 40 years, is a oversimplification to say the least. Besides, do you honestly think that we are no closer today than we were 40 years ago?? (On a side note, I seem to remember little pamphlets they used to hand out in the early eighties about how we would all be flying around with Jetpacks by now, and colonizing the moon. Anyone think maybe they made a mistake on those?)

I think it is probably one of humanity's greatest moments of hubris to assume that we are what computers should strive to be.

Just my 2c
The "Top 10" Reasons to procrastinate:

Top-down or bottom up?

[naasking]

In your opinion, which approach is most likely to succeed in developing the first AI: the top down approach(ie. programming a huge database of intelligent responses etc.) or bottom up approach(ie.neural nets and evolutionary models)

Also, what tests will be implemented to verify the existence of true AI. Will such tests attempt to verify true consciousness, or merely intelligent responses to various stimulaii.

-----
"I will be as a fly on the wall... I shall slip amongst them like a great ... invisible ... THING ... !"

Re:AI and their rights

[kromo]

At what point do you think that an AI should gain the rights of humans?
When do you believe it will be murder to unplug them?

Re:Graduate Schools

[esperandus]

I know of groups at MIT, Cornell, Stanford, Harvard, (but dont go there!!!), CMU, and, as you said, here at Brandeis. Many places are sure to have small projects or isolated individuals interested in the topics, of course. Look around--a degree is only a piece of paper; the work might be much mroe interesting than it seems at first. Search the publications till you find something that looks cool to you, and then find the guy/girl who published it.

The volen center (at Brandeis) and Mr. Pollack's DEMO lab within it is larger than most such groups, and rather good (plug plug). It benefits from a truly interdisciplinary approach; the neurobiologists and Computer scientists all eat lunch together, and its pretty neat to see how people bounce ideas off each other until they make some pretty impressive snowballs. Consider it. Good luck--you cant really go wrong.

Matt

PS--if youre into complexity/game theory, look into the Santa Fe institute.

PPS--some of the above departments/programs are not well advertised. Be prepared to do some hard looking; see if you can talk to the people involved if possible. Some institutions have a hard time with the 'kooky' image of AI work, along with the ethical dilemmas raised here and elsewhere, and tend to be *rather* quiet about their resident oddballs, no matter how brilliant.

Re:Questions based on your academic path

[esperandus]

How about taking his class on machine learning? Or his class on artificial intelligence? The Cognitive Science/Computer Science Program here at Brandeis would be a good start, along with all of the other opportunities in Boston (MIT talks, Programming symposiums.) In my (limited) experience nobody does it for a 'real' job--everbody in the field is working for someone with long-term goals and big-time ambitions (large corporations, wealthy individuals with a techno-twist, Universities,etc.). Just look at, say, cog. The field is simply not mature enough to support 'normal' commercial developement. The best way to succeed in the field, though: be brilliant :-) Good luck! Matt Mcrouch@brandeis.edu

Re:AI and ethics.

[esperandus]

I am reminded of the first hippocratic principle:

1. Do no harm....

The industrialized nations are doing more (net) and lasting damage than the 3rd world countries. The U.S. probably releases more contaminants and causes (however indirectly) more death than all other nations combined (arms sales, SUVs, Computers, Coke, etc). Comsupmtion is the cause of all pollution (at a certain level), and we are the worlds #1 consumers(TM)

The only thing that wealth has consistently created is arrogance and more wealth.

Of course, it may also create The Solution. I believe, as you seem to do, that we have gotten ourselves so mired in our own excesses that the only way out is up--that is, through a greater reliance on the technology we have been using to kill ourselves and just about everything else.

OTOH, Given our racial history as 'saints', a little bit of trepidation concerning such power as we seem to be achieving--possibly realized most profoundly in AI--is to be expected.

Not that we should not be happy... Mcrouch@brandeis.edu

Re:To which I would add...

[hyrax]

You are confusing two concepts: utility and wealth. Utility is a measure of the extent to which a particular good satisfies the needs or desires of an individual. The utility of a good varies from person to person. Wealth is an aggregate measure of exchange power between individuals.

Different goods have different utilities to people depending on circumstances, personal preferences, etc. The whole idea behind trade is that the best way to maximize the overall utility in an economy is to allow people to redistribute goods between themselves so as to approach their personally optimal mix of goods. Wealth is the total value of a person's goods expressed in terms of what they could trade them for. In short, maximal utility is the ultimate goal, wealth is the power to get there.

There are 2 points of particular interest in the wealth/utility spectrum. One is where a good has worth, but no utility. The definitive example of this is money. Money has no utility whatsoever on its own, but it obviously does have worth since it can be exchanged for goods which do have utility and thus it contributes to wealth. This is not an accident since the whole reason for money is to act as a 'universal solvent' for trade.

At the other extreme are goods which have utility, but zero worth. The phrase 'Selling ice to Eskimos' comes to mind. People living in the Arctic may find snow and ice useful for water, shelter, etc., but they certainly would not trade for it. Your vision of infinitely duplicated, freely distributed IP also falls into this category. Any particular IP good has some utility to some people, but if it is ubiquitous it is not tradeable and so does not contribute anything to overall wealth.

For example, imagine that you somehow gave every person on Earth a digital copy of 'The Matrix'. Those people that had an appropriate playback device and that wanted to watch it would certainly feel that the overall utility of their belongings had increased. The other 99% of the world would look at you with a mixture of confusion and pity. Not only is their utility not increased, but neither is their wealth since you've made 'The Matrix' worthless for trading purposes.

But, you say, they are certainly no worse off than they were before, so the overall satisfaction (ie perceived utility) has increased. Yes, but overall wealth has decreased because anyone who previously owned the movie and could have traded it for other, higher utility, goods is now denied that opportunity. It's the same as if the government started to print and infinite amount of currency. With currency it's actually worse because, as I said before, money has zero intrinsic utility. (Or negligible: photos of Weimar-era Germans wallpapering rooms with deutschmarks come to mind) At least with IP you are increasing overall utility, but at the expense of decreasing overall wealth.

So, the real question is, is the tradeoff worth it? I would say no for a couple of reasons. First of all, I think that the overall increase in utility would be quite a bit less than you seem to believe. If everybody on earth had access to all the IP you would have a relatively few very happy people, that's it. The vast majority of the world would not see much of a difference. While it is true that the value of IP-based goods is large and growing, what you neglect to mention is that the vast majority of them are consumed by a tiny fraction of the world's population.

The second reason is that I believe that the cost in terms of destroyed wealth would be too great. Not only would the present wealth represented by IP be lost, but future creation would be severely dampened. Yes, I know that some people will write, sing, code, etc. because that's what they love to do, but we need to eat and pay rent too. How many books will your favorite author write if they have to work a 40 hour a week service job? And do we really want to forgo the contributions of that subset of creators who are very adept at what they do but who wouldn't necessarily do it in their free time? Also, that whole premise of spontaneous creation ignores the fact that a lot of IP products are the result of large teams of people, most of whom are not artists. I may want to write screenplays in my spare time, but good luck finding a gaffer who will work for love of the art. It's hard enough to get non-union labor, much less free labor. I'm sure that people can come up with individual counter-examples, but those cases will by definition exist whether IP is free or not. The most universal way to motivate people is by the prospect of material gain, and to believe that you can remove that motivation and still generate the amount and quality of goods flies in the face of not only economic theory, but centuries of experience.

Artificially pretending that IP is a scarce resource will keep the lawyers, accountants, politicians in work, and will also allow some money to flow back to the creatives, but at the cost of impoverishing humanity.

First of all, all property is 'artificial' in the sense that it is a societal construct, IP is no different. You are correct in saying that it is made scarce, and hence valuable, by IP laws, and I believe that that is a net benefit to society. Look at it this way: In the end everyone needs material goods such as food, shelter, and really fast sports cars. Since such goods are currently scarce, prospects of starship replicators notwithstanding, the only way to acquire these goods is through trade. In a world without IP, the only tradeable goods are material goods and services. The addition of IP simply adds a whole new class of tradeable goods and thus a whole new class of professions for those so inclined. The elimination of these goods and professions is what would truly be impoverishing.

Just to clarify my position and maybe head off at least some of the flames, I believe that the current implementation of IP in the US is gravely flawed and that copyright and patent reform are some of the most important issues facing our society. I just believe that the answer is to modify the implementation, not scrap the concept altogether.

Re:To which I would add...

[hyrax]

Your definition of wealth is actually a constant. By your definition *total* wealth cannot rise or fall, so when you claim free IP would be "decreasing overall wealth" you're obviously confused.

Not at all. My definition of wealth is simply purchasing power. As you stated in your original post, it's not a zero sum game. If I create something that didn't exist before like a book or a movie, I have increased my purchasing power by whatever amount people are willing to pay for that product. Similarly, if I find a way to increase my productivity and produce more of a given good per unit of input, then I have increased my wealth. Neither case decreases anyone elses wealth, so total wealth has increased. Empirical proof is all around you. The explosion of wealth in the two centuries since the industrial revolution is the reason that people like us can afford to spend idle time debating on /.

By your terminology raising total "utility" (which is pretty much equivalent to what I call wealth) is the goal which improves mankinds economic condition.

I agree 100% with that statement. My point is that the way to achieve that goal is through trade and the ability to trade is measured by wealth. In a simplistic example, people who are good at writing books trade them with people who are good at fixing cars, that way everyone can have both books and working cars. If you make the author give away his books, then he has nothing to trade and you have decreased his wealth without increasing anyone elses and so have decreased the total wealth in the system. You have increased the mechanic's perceived utility since he now has the book, but the total utility in the system is less than under the trading scenario because the author's car is still burning oil.

Strange claim. Imagine if everyone had free access to all books, records, drugs, movies, software, artwork... Everybody would be far richer (or more "util-full" if you must).

With the exception of drugs, the bulk of the world's output of those things is consumed by the minority of the world's population that is wealthy enough to have the free time to enjoy them. The rest of the world is too busy trying to get adequate food, shelter and medical care. Even in a rich, developed country like the US, I don't think that what you propose would have as great an effect as you think. For the average American, the factor that limits their consumption of these goods is not money, but time. I really don't think that $3 rental fees are keeping people from seeing movies that they want to see. As far as drugs go, like I said, I am a proponent of patent reform, just not abolition.

When we reach a world where you go to the mechanic and he downloads and "prints" your new part instead of ordering it (possible already - www.3dsystems.com) then *most* wealth will be in the form of IP. In the future, if all IP was free then everyone would be so much richer that working to stay alive will not be necessary.

Ok, I agree that if and when technology advances to a point where everyone's material desires can be met by a virtually unlimited and free supply of nano-assembled products our society and economy will undergo a radical structural change. It is very possible that such a shift will alter the terms of the tradeoff I mentioned between IP as wealth vs. IP as utility so that it is more beneficial to have completely free IP. However I also believe that protected IP is the best way to get to that point. The investors who have provided the resources for 3dsystems have done so because of the prospect of the financial rewards that will come from developing a unique product. Call 3dsystems and ask them how many of their employees are volunteers working for status.

Even today (in Western society) everybody works long hours not in order to survive but for status.

First of all, I would dispute the assertion that everybody works long hours. The fact is that the average worker at the end of the 20th century works much less, and earns much more, than the average worker at the beginning of the century. Hours worked have decreased while the proportion of income spent on leisure has increased. There's lots of empirical evidence for this.

Second, I think that the idea of the gift economy has serious flaws. I've never heard a convincing explanation, much less seen any evidence, that a status based gift economy is viable without being subsidized by a traditional market economy. Every oss programmer I know is either a student or has a 'regular job', often one producing copyrighted software. Even if you get past that and are able to attract all the talent you need with status-based compensation, it still seems massively inefficient. A gift based economy will still have to interact with the market. The creative people on the front lines may be willing to work for status, but you still have to pay the secretaries, the film crew, the maintenance workers, etc. So instead of a single economy based on a universal and quantifiable medium of exchange (money), you have two intertwined economies, one of which is based on barter of an ephemeral quality called 'status'. And yes, I'm well aware of the various successes of the open source world, and I agree that the model can work spectacularly well for some projects. I just don't think that the success of a few subsidized software projects is enough to herald a full blown economic paradigm shift. Time will tell though . . .

Re:To which I would add...

[hyrax]

Why does something have to be tradeable in order to contribute to wealth? I object to your limited view of what counts as wealth.

It has to be tradeable because that is how wealth is defined. You must be able to see that everything you own has two distinct 'values'. One is the value that is has to you in terms of pleasure it brings you or needs it satisfies. That is what I referred to as 'utility'. The second value is what you could get if you decided to trade it. For example, a $100 bill has no value in the sense of utility. It's just a scrap of paper. However it does have a significant value in the sense of wealth because I can trade it for food, clothes, a 3-day supply of Mountain Dew, whatever will bring me more value in the personal, utility sense. On the other hand, my favorite t-shirt has a significant value to me in the sense of utility as it makes me feel comfortable and happy. (ok, I'm a little pathetic) But it doesn't contribute much to my wealth because nobody is likely to give me much in exchange for it. This is not a personal philosophy, it's a very fundamental description of the nature of property and trade. Maybe my choice of the words was confusing. I used 'utility' and 'wealth' because they are farily standard terms, but if you want you can replace 'wealth' with 'purchasing power'. Again, understand that I'm not claiming that purchasing power equals happiness or anything like that, I'm just defining it as an attribute of property, distinct from utility.

But why should I care? If I have enough to keep me fed, clothed and housed, why should I feel wounded by having shared something? You will say, and quite rightly, "But you missed the opportunity to earn some money by making your idea artificially scarce!" to which I would reply that I don't regret the loss

Actually I wouldn't say that as long as you were free to make the choice and take whatever course of action brought you the most satisfaction. The key is that as the creator, you were free to make that choice. On the other hand, what if you desperately needed money to pay medical expenses for a family member and after your discovery someone came to you and gave you a pat on the back and took the result of five years of your life's work and gave it away. Altruism is a pretty shaky foundation to build an economy on. Should society forgo the contributions of your next door neighbor who is just as good a physicist as you but who is not nearly as altruistic and doesn't want to work on a fusion reactor unless there's something in it for him?

If there were no movies people would tell fantastic stories or write folk songs or play in garage bands. I'd much rather have that than your idea of mass-produced corporate sausage casings.

Fine, but the point is that my idea doesn't prevent people from playing in garage bands whereas your view does away with my sausage casings. It's a matter of freedom and choice.

You are limiting the value of anything to the price it will fetch at auction.

Exactly the opposite. The whole reason for my lengthy definition 'utility' as distinct from 'wealth' was precisely to separate the concept of 'value' from 'auction price' and illustrate the interaction between the two and how the abolition of IP laws would affect both.

If you view wealth as a sum of prices, and you seek wealth as an end for your society, then it makes sense to organize information as you say. But I don't agree with your definition of wealth, and if I did I wouldn't want to make that my ultimate goal.

Did you even read my post, or are you willfully misunderstanding it? I'm starting to think that this is a pure troll. I specifically said that that wealth is not an end in itself, just a means to an end.

Lisp book

[hading]

Paul Graham's "ANSI Common Lisp", which I am currently reading, is IMHO a pretty good book on Lisp. It contains both a tutorial on Lisp and a reference section.

Graduate Schools

[naoursla]

I am planning to attend graduate school in one to two years. My interests lean heavily toward
topics of evolutionary computing (and using biological models for computation in general).
It looks like Brandeis has a good research group for this topic, but I've found it difficult to
find other schools with groups working heavily with evolution. Do you know what other graduate
schools are have research groups in these areas?

personal neuro-evolution experience

[naoursla]

Thank you for the information. BTW, I've played around with evolving neural networks a little. Here is some Java source I wrote to a fully connected net that learns to count to 7 using only one input to tell it when to start counting. It is based on the technique that Foley and Chellapilla used to evolve a checkers playing network. The program is pretty simple but I found it interesting. I wrote it for a class I am currently taking.

It's evolution, babe.

[Punto]

Is it ok to freak out about the "superior" computers? Isn't that a natural thing, evolution?

Did monkeys sit around saying "what are we going to do if this people turn out to be more inteligent than us?" when the first "homo something" appeared?

Would it be very difficult to the human race to accept that they are not on top of the evolution chain, that there is something superior, and let it be free?

--

Re:Replacing Humans?

[Control Group]

An excellent point, but one I think ought to be taken one step further. The current drive for AI is (naturally) to put it in a role where it can benefit people--an excellent example of which is the taxi-routing software. Assume that the software can eventually replace the human decision maker. Assume further that this happens across the board, in all fields (logical endpoint). Ultimately, why does the taxi router need to exist? People no longer need to take taxis anywhere.

The question then becomes, does goal-oriented AI (that is, AI designed for a specific purpose) obviate the need for itself to exist?

Hardware > Software

[ralmeida]

Thinking about the evolution of our own intelligence, one thing that always sounded strange to me is the fact that our brain was very developed before intelligence appeared. My point is, the Neanderthal man had a brain as big as ours, and didn't use it to the same extent.

Do you think that AI is still in its begining because we still don't have the required "hardware" so that it can really evolute? Do we need (a lot) more powerful computers just to begin playing with "real" AI, e.g., one that can pass the Turing test?

reaper vs. frogbot

[siokaos]

In Quake, there are two main solo deathmatch mods:
reaperbot and frogbot

AFAIK frogbot uses a pre defined set of nodes and paths that reaper generates as it learns.

The result? Thousands of quakers play the frogbot prebuilt levels and limit their overall skills.

A reaperbot experience is one of true learning. By playing two or less bots, you are able to get to know the bots individually. Reaper bot uses an advanced fuzzy logic technique that allows for great overall play.

I would like to hear peoples opinions of these AI bots and the end result.


.sig:

ethics?

[siokaos]

Why let "moral" ethics bring down science?

No matter how immoral, if theres science to be done, eventually it will be studied. The only problem is when scientists push just for funding and are blind to their research's effect of society.

Think about it, humans brains grow from experiencing instances, reacting to these, and formulating.

Why would it be wrong to apply the same practice to a machine?
.sig:

Evil machines vs. Obsolete bodies

[knitephall]

It seems that in all the Sci-fi novels and movies, the paramount concern with AI is that our creations will turn on their masters (take Frankenstein, for example). Rather than creating AIs that will enslave and kill off the human population, however, what are your views on the possibility that we might simply leave our bodies behind to take on the machines that we are designing (it seems that flesh and blood may soon be rendered obsolete with new technologies)? Evolving into a non-carbon based body rather than humanity simply being wiped out.

What If ?

[Schmacko]

For a long time AI has been researched and worked on from Projects like LISA and beyond. But the one true worry still exists, if we do create AI and it does see that it does not need the human race to survive anymore (going on the assumption that if it was true AI it would be capable of all levels of thought), whats is stoping it from removing us from the face of this dear planet. My main question to you is, is this another myth, or is it something being taken seriously and things being done to ensure that AI does not become the next creation to rule this earth?

AI bot learn by itself on the net?

[sethking]

If you were trying to get an AI bot to learn English (being the dominant net language) could you use the Internet as a resource to learn on?

If you had some system of determining the value (accurateness, trustworthiness...) of each document come across and also had hardwired some basic English definitions into it, would you have a reasonable chance of creating comprehension?

As a more brief question, do you think that this (or any) type of program could become anything more than frequency counters. Is actual comprehension feasible through just providing it with a HUGE resource to sift though?
-Seth King

What is AI?

[rm-r]

Hi, I'm just beginning to get invovled in AI research at my University. One of the things I have encountered during my studies is a variation in what is regarded AI, and how the subsections of the subject are divided. What is your view on what currently makes up the AI field (one of my lecturers tells me that in the 50s compiler design was considered AI, and that from then whatever is seen as being 'cutting edge' tends to be piled into the AI circle). Secondly, (and I hope you have time for two from me), the particular field I am researching into is language recognition. I am amongst a group who are developing algorithms to detect language-like features in signals arriving from outer-space. The hope is that if the SETI program does discover something of interest our tools will be used to validate and examine it. Work that has been carried out so far has looked at the sounds Dolphins make, by looking for language-like features in what is a completly alien speices to ourselves we hope to obtain tools that will work in a more general context. Any ideas?

Stop imitating human behavior

[Ru610]

It always bothers me that people expect such human qualities from AI. I am an AI student myself and my benchmark is simply:

AI is achieved when a computer program/algorithm masters a task for which we think some form of intelligence is required.

Therefore I believe AI is achieved in a simple chess program, however simple its algorithms may be. I think a lot of effort in the field of AI is wasted on mimicking humans as closely as we can. If we want to create another human I think there are much easier ways to achieve that (take cloning for instance). The real usefulness of AI will present itself when we start exploring new ideas with it instead of trying to duplicate human behavior.

Ethical Education of Intelligences

[ltcordelia]

The current outbursts over AL (esp. cloning) raise interesting questions about the ethics of "breeding for purpose" (supersoldiers). Assuming Machine Intelligence actually becomes a reality in our lifespans, there become some interesting ethical dilemmas, based on the assumptions:

• Anyone with sufficient (?) computing power can generate one.
• MIs will be "progammable" to desire/excel at certain tasks.

Far be it from me to suggest that every script kiddie will want their own "pet AI" to run their attack scripts for them. But I'd guarantee that a number of nationstates would slaver at the idea. The question: How does the AI research community propose handling the ethical issues of allowing anyone to generate an AI without giving it an ethical or moral education?
Information wants to be free

Re:AI and their rights

[ltcordelia]

Personally, I don't think an AI/MI will ever have the rights of a full human (especially not in a democracy), for a few simple reasons. Ballot stuffing - I create and program an AI to vote for the candidates I choose. Repeat as necessary. Fiscal accountability - Oh, my AI entered into a contract with you and owes you US$1B ? Oops, that AI rebooted yesterday. This computer? Oh, that's its "sibling", v3.0. I think AIs would be more likely to be granted status similar to that of minor children than to be granted adult citizenship. I guess the really big question is "how do we determine when an AI is an AI and not just a cool expert system?" The biggest problem is that any formally structured test would be easy to `game' by creating an expert system designed to pass it.
Information wants to be free

AI Career Opportunities

[ltcordelia]

Assuming fully functional intelligence, I could think of a number of jobs they could perform. Copyeditor. Programmer. Chip designer. Software tester. Simulation Manager. Game developers. Game players ("Have you seen the new AI champion of StarCraft? Check him out on battle.net!"). Secretaries. Information brokers. Supply managers. Budgeteers. Spokesperson (spokesmachine?).
Information wants to be free

LISt Processor

[klyX]

Prentice Hall publishing released a grate LISP book, it was part one of their series on AI I believe . . . Sorry I forgot the details . . . link

Comments on Bill Joy's position.

[ucsimon]

If you've heard of , if not read Bill Joy's recent article in Wired, what do you think about his fear of a human-less future? Do you think that the future of AI is one that needs to be tightly regulated in order to protect ourselves from self-replicating robots? In the future do you anticipate something similar to the bans on nuclear testing with AI?

-dennis the kid

Intelligence?

[Angelo Torres]

Hello,

I don't mean to offend you, but it seems to me that the area where "Artificial Intelligence" is being "used" the most is in computer games. Within the scope of games, AI usually refers to some type of code that chooses which route a bot (FPS) or forces (RTS) should travel in order to reach a specified destination. Now, although this may seem intelligent to the average player, I doubt it seems very intelligent to the programmer who wrote the code, or to another person who reads the code; the computer is merely following his/her instructions to the letter. I would assume that some "AI Scripts" would call some type of random function to provide odds for taking separate paths, but this is still not Artificial Intelligence in the true meaning of the word. With this background I would like to ask:

How could it be possible for computer hardware or software to perform an action that it is not explicitly told to do? In other words, how can the computer say to the programmer: "Hey buddy, your doing this all wrong, here, let me show you how it's done"?

How can Humans ever hope to model and physically present Intelligence, something that they do not understand, in any form?

Angelo Torres

intelligence?

[putt]

From what I've heard and read, many of the big statements coming from the AI community point out that all this AI stuff works just like the brain; be it some image detection algorithm, a robot or whatever else. I guess that must be the point, otherwise there would be little sense in calling it intelligence, artificial or not.

Another thing I learned in the past is that the people who are actually studying the (human) brain, the 'I' experts so to speak, keep pointing out that the knowledge about how it works is very limited, especially when it comes to more complex (higher) functions.

Now how comes that some (many? most?) 'AI' experts, who may be brilliant when it comes to technical issues, seem to know more about the humans brain functioning than the ones studying it? Their statements seem quite unfounded to me, rather like pure speculation. IMHO, what many AI researchers implicitly do is that instead of actually trying to understand how the brain works, or what intelligence is, they reduce the brain to something simple enough for their minds to grasp: that's not science, that's pure speculation.

Who knows, maybe the brain is just a bunch of neurons, and the intelligence automagically pops up once you successfully connect more than a million of them. It doesn't really matter, the implications are mostly philosophical anyway. But I expect a scientist to be able to distinguish between known facts and speculations. And that seems to be something the AI community is not very good at.

Assuming a sensor and some reflex like behavior, plus a couple of additions and multiplications in an artificial neural network is all it takes to create intelligence. Then where exactly is the difference between that kind of intelligence, and, let's say, an anti-lock breaking system of a car? I don't see a fundamental difference between the software or the hardware used in either of them - Turing certainly wouldn't. So we could go ahead and claim that a car is intelligent, artificially at least. But artificial intelligence is just like the real one (as 'AI' experts claim), and therefore, we humans are all just some mindless robots which, maybe because god forgot, don't come with two wheels attached to either side. Then again, if we're mindless, we should probably stop talking about intelligence as their is no such thing anyway.

On the other hand, if artificial intelligence is not like the real one, what has the whole thing to do with intelligence?

Turing Test

[RistRind]

In AI research these days, is the Turing test still seen as a valid benchmark of how to gauge how intelligent programs actually are? Do you think that a machine would be able to pass the Turing test without being able to truly think, and, if so, is there a better test out there?

Re:Turing Test

[CharlieT]

When I studied AI at University (not that long ago) it was the general opinion of most people I talked to the the Turing Test is a bit flawed.
The problem lies in the fact that perfectly intelligent humans have been mistaken for computers, thus failing the test. This means that the test itself is not an accurate indication of intelligence.
Still, it's a challenge worth persuing in my opinion.
P.S. I reserve the right to be wrong about this ;)

GOFAI vs. Evolutionary

[PantalonesVaqueros]

After taking a look around the DEMO website, I'm curious to see where you stand on the "Good Old Fashioned AI" vs "Evolutionary" (for lack of a better term) debate. Do you see a need for expert systems or case-based reasoning, or do you expect A.I. approaches like Neural Networks or Genetic Algorithms to eventually supplant GOFAI? Explanation is something that GOFAI is well-suited to. Do you see, for example, Neural Networks being used in explanation of solutions?How?

What is AI?

[infra-red]

It seems to me that AI is something that you either have achieved, or something that you have not achieved. Alot of people though claim to have acheived some degree of AI. Is there a definition of AI that allows these claims to be true, or are they simply trying to take advantage of (and/or build) hype around the concept of AI?

AI

[Gutzalpus]

What books/literature would you recommend for someone interested in finding out more about AI? I have programming background but I haven't done much with AI and it's something that's interested me for awhile.

Old, boring.

[Caffeinated]

I'd like to know how you can justify your work as art. Just randomly throwing paint at the canvas, what are you trying to say, that taste is arbitrary? that what we call art is just so much existential smatterings in the cosmos? Do you think you're deep or something?

Oh, wait, I thought that said JACKSON Pollock. Never mind.

- - - - -

Re:AI and ethics.

[jallen02]

rofl

Re:AI and ethics.

[jallen02]

Whatever dude, you are just beating yourself to death here. The AI will respond however it is PROGRAMMED to respond, not how you think it should.. They are machines, and quite honestly I dont think AI, will ever be as complex as my mind. Yup I am a naysayer to this whole deal. And you are just off on planet 9. I wont be suprised if it happens, but I doubt it will. And even then, ill take a person to a computer ANY day. Period. You sound like a lonely guy in need of a little friendship..

Jeremy

Re:AI and ethics.

[jallen02]

heh, its okay im lonely to.. but I just think that a human beings 'programming' is just so complex and varied that to even ponder reproducing something that highly programmable and definable as a persons personality is somewhat crazy. I have an associates in math so I am not really qualified to be speaking about this, but Ive been programming for long enough to know that its just.. WAY out there.

Think of this. Any attempt we make at creating another intelligence is inherently 'flawed?' since we are basing this model after our own methods of thinking, therefore I dont think it will ever become a higher being, nor do I think it will be without serious flaws should anything even closely resembling intelligence that a human has, lets not even go into the way emotions influence our intelligence and our responses to well everything. Let me say a few words, how much programming do you think it would take an AI to respond to things like this? rape, murder, death, God, sex, love, hate, loss. Come on, there is something there that is just to much to solidfy, and no matter what it takes human emotion to fully impart the empathy we each experience with such thoughts. Yes? No? I agree with what you are saying but its so much more isnt it? I dunno, im a very emotional person and Ive been through a lot of crap in my life... I like conceptualizing to.. and the concept is fine and dandy, but to me something is still missing.

Re:Turing award.

[Digital Mage]

Hmmm, that gives me an idea. An AI slashbot comment poster. It's been done with IRC and newsgroups, why not Slashdot? (Oooh, you could even use the moderator points as conditioning weights).

Now trolls make sense to me. They're really AI slashbots in disguise slowly learning to send intelligent posts. ;^)

AI's Ultimate Goal

[Digital Mage]

Any technology attempts to achieve a specific goal (e.g. Internet - faster more reliable communications), but brings with it a number of unexpected side effects (e.g. Internet - breaking apart old business models, etc.).

What do you feel is AI's ultimate goal, and what unexpected side effects do you feel may be a result from the wide spread use of this technology?

Who is the inventor for AI-created IP?

[jjsaul]

This is a question I have been thinking about since law school... not just for true AI, but even for simple genetic algorythms. In essence, AI creates IP (lets say a better turbine design) - under patent law the inventor is the inventor regardless of a "shop-rights" contract. The company may own the IP, but the inventor is still the AI. My patent professor's unsatisfying answer is that the first human to perceive the innovation would be the inventor under current law, even if it happens to be the janitor cleaning the monitor at night.

Here's why it is important... if as I propose the AI is acknowledged as the inventor, with an implicit transfer of patent ownership to the AI's trainer/owner/designer, then we have acknowledged that an AI has constitutional rights, which would need legal standing to be protected, which would drift into constitutional "personhood".... you can follow the bouncing ball from there.

Not that I think this is proper yet... from a practical stance constitutional rights for AI would interrupt experimentation in the field... but an interesting legal fiction that may lead to fundamental reevaluation of "personhood."

Re:Why not Write a Screensaver?

[Dobbie]

That is a very interesting proposal because I believe it would involve a non-typical NN struture. The main problem is what to do as nodes dissappear (go offline, reboot, logout...) and where in the structure to insert new nodes. How would such a NN get trained? It would have to be done in such a way that the NN could still function as its structure was dynamically changing. This ability to dynamically alter the structure of the NN could be very advantagous and even lead to more efficient NNs

Re:religion and intolerance

[ibpooks]

Corrolary:

I am sick and tired of every legitimate scientific, software, and computer-related post being met with pages and pages of religious protest. How about we have posts that RELATE TO THE TOPIC and don't advocate any sort of religion.

Re:How do you do you when you've achieved AI?

[Shin Elendale]

I would say when you ask a computer "who are you?" and it can give a logical response when you ask "now prove it." Alternatively, or perhaps in addition, emotions are a big part of sentience (IMHO) and if I can anger a computer by installing NT on it (a`la User Friendly's Erwin) then I'll say its alive :)

-Elendale (Also, in order to be intelligent it has to have a will to LEARN! Note that this disqualifies about 75% of the human race)

Obviously...Mr. Frankenstein

[mekkab]

The areas that should be concentrated on to avoid such a "Frankenstein Conquers The World*" problem would be ethics, realistic assesment of the drawbacks of any scientific/technological advances, etc.

I guess the idea is that if you are standing on the threshold of what is obviously the next level of human existence, don't be so consumed by greed. For example, the great Von Neumann (when confronted with the greed of his compatriots**) published papers on EDVAC... there-by preventing anyone from having a patent on the computer (can any of you imagine what would have happened if one company, say IBM owned all of the computers, all of them?)
That's why Open-source is the only way for humanity to profit!
So the recap: Don't just do it for the cash, do it for the greater good of the world.


*- Anyone ever heard of that hysterical song?
**- in my studies of computer architecture It was presented to me that Mauchly and Eckert were in it for the money (which they eventually made)

Essay by Bill Joy

[Mike Belangia]

Does anyone have a hyperlink to the essay mentioned here?

Re:How advanced are we, really?

[kwsNI]

Do you weigh 1 pound? (Y/N) N
Do you weigh 2 pounds? (Y/N) N
Do you weigh 3 pounds? (Y/N) N

Actually, this isn't AI, it's a simple loop. What true AI should be is saying:

Last time I saw someone this size, they were 150 lbs.
Last time I saw someone like this, they seemed 20% thinner (Adds 30 lbs...)
Guesses 180 pounds. (Y|N).
If yes, stores this in memory for future reference.

A lot of what AI is about it learning. AI shouldn't make the same mistake twice and it should react to new circumstances based on previous observations.

Another example is playing a game. For instance, a truly good AI bot should notice that a Q3A player has a weakness when the bot is firing and straffing. Then, the next time it notices this type of player, it should be able to recall what strategy worked best before and apply it again while still learning what works better.

kwsNI

AI is constantly changing

[_tanden]

The reason AI solutions are always 10 years away is that when AI problems are understood, such as the sorting problem, they can be studied mathematically, and then when these problems are understood mathematically then they are no longer AI problems.

Take the sorting algorithm, ten or twenty years ago, sorting was considered an AI problem. Today every first year programming class teaches you how to do sorting. Sorting is so well understood mathematically, that it would be rediculous to consider sorting as an AI problem (at least today!!)

Rights

[geekoid]

If a self aware AI is created, should we allow it to have the same rights as individuals receive?
If allowed to replicate it self, How can we put limits on it?
Is it wise to have something doing mission critical calculations, if it can have a "bad hair day"?
Do you think an AI can have faith?
Do you think the ability to have faith is critical in order to be considered self aware?
If AI is created via software will it replace the software industry? If it could spawn many process of it self, then it would eventually replace most people.
how will it impact the world economy?
Thank you for taking time to answer the questions here on slashdot!

Representation or Behavior key to Intelligence?

[tylerh]

Is Representation or Behavior the mark of Intelligence?

Dynamical systems have two principal descriptions. The "state" viewpoint focuses on the internal respresentation, ie " how it thinks", while the "input/output" perspective focuses on "what it does". Although formally equivalent, the viewpoint one adopts prefaces how one attcks the problem, and even shapes what one considers the problem to be.

Which viewpoint do you believe is more central for comprehending the "I" in AI?

Re:AI and ethics.

[JWRose]

It's the same as the people who think things like cloning and genetic engineering are bad because it's "playing god".

I can't speak for anybody else, but for me, cloning and genetic engineering delve into areas that will only harm the planet in the long run. This planet is already over populated with humans. We, as humans, are constantly destroying our environment with deforestation, polution, etc. Nature has a way of dealing with this sort of thing. There was system of "checks and balances", so to speak, that kept species levels at an equalibrium. With the advancement of medicines, genetic engineering, etc, we humans have altered that system that will eventually lead to the end, unless of course, we can find another planet to live on before then.

AI, if used responsibly, can be a great thing. The problem comes when the AI computers that we created, become smarter than we. (Think Terminator and Matrix)

We are just going into the realm of the unknown. Where the consequences to these technologies aren't fully understood. Of course, most people only think about themselves and the amount of money they can make off the technology, so of course, this whole argument is meaningless.

Nothing exists except atoms and empty space; everything else is opinion.

Re:AI and ethics.

[JWRose]

My existience is first priority over any warm fuzzies that I get from the earth. If I have a choice of doing something for the earth and allowing my life to go on I will always choose life and the ability to make a difference with what really counts our human race.

Your response proves my point. Thank you!

*sarcasm* Oops there go those nasty humans trying to make things better: bad humans!*sarcasm*

How are we making things better? With all the advancements in our technology, our environment is getting worse and worse for everyone and everything. We are doing more harm to ourselves in the long run, then we are doing "better".

Nothing exists except atoms and empty space; everything else is opinion.

Essence of intelligence

[oddRaisin]

This is somewhat of a philosophical question, but most things in AI seem to be that anyway. What do you consider to be Intelligence, or the essence of Intelligence?

The way that I look at it, it seems that all intelligence is is an incredibly powerful and generalised pattern recognition engine. This model provides a relatively simple model with the only requirements being relatively simple ones (Basic inputs and instinctual incentives such as those governing self preservation, etc).

Finally, what do you see as being the single greatest obstacle in the way of AI (Social acceptance, a technical obstance, etc)?

Regards,
Ian

Re:Essence of intelligence

[oddRaisin]

Obstacle, that is, not obstance :)

Approach to AI

[xeroh]

It seems that there is the GOFAI, rule-based approach (e.g. expert systems) and the newer emergent/self organizing systems approach to AI, and that your work seems to focus on the latter.

To what extent do you know what sort of behavior you will produce when you design some AI-ish program? With a rule based approach, one has much more direct control over the "intelligence" created. Do you feel that the work you do often has unanticipated results, given the complexity of the systems involved? Do you think there is some middle ground between large, declaratively created ontologies and evolutionary systems? What sort of mixed strategy are you taking?

Language

[Racer X]

While it seems to me that we already have achieved something deserving the name AI (like H. Simon says), what most people imagine when they hear 'artificial intelligence' are machines which demonstrate really human-like intelligence. That means getting a computer to understand and use natural language and common sense (probably not unrelated things). Surely these are among the most difficult phenomena AI people face. I've always been skeptical of success in these areas, esp. considering the incredible complexities involoved in understanding the nature of language and language use. What's the state of the art in these areas? How much success is AI really meeting with implementing language? Any predictions?

thanks.

Semantics and AI

[steelwraith]

Do you think that the moniker 'artifical' can really hold true in the pursuit of creating intelligence?

Based on the premise that intelligence was an evolutionary process, and not something that just spontaneously happened, what justifies calling a silicon/germanium/other-than-carbon-based intelligence artifical? Aren't we in fact, through the process of improving computer hardware and software, evolving something that, while not aware today, may be aware in the future?

(Note: I will not rant on other facets of this topic, as I would spend several hours typing before the villagers showed up with their torches and pitchforks)

When the Master becomes the Slave

[jwainner]

I know that this might be pressed upon someone like you all the time when you are as involved in the research of AI as you are. But I must ask and I must reiterate; we've all heard the stories, Hollywood has made millions off of it, but could the advent of AI ruin our society and ultimately the human race? The most recent publication I have read about the topic of AI and GNR (genetics, nanotechnology, robotics) technologies has made the cover story of last month's Wired. "Why the Future Doesn't Need Us," by Bill Joy, also a not so old headline here at Slashdot, brought up some good points. Those who are in power of the robots or AI would be nothing more than Big Brother was to Winston Smith. Robots as we already know are much better at doing what they do thatn humans. As AI advances, society will not be needed. Let me restate that; it's not that we won't need AI, AI won't need us. We will become so dependent on the machines, that turning them off when they "get out of hand" will be impossible. Therefore, slowly, the machines can just hold our neccessities over our heads. We will have no choice but to obey and then the master shall become the slave. But back to the power of others. If a select few, such as the inner party of Oceania, had full control over the AI, what would become of us, the outer party? Two choices come to mind; 1. The leaders who control the AI will no longer need the majority of society and will not want an uprising. Of course one would be imminent anyway because of the full control aspect of such a powerful thing as AI. The leaders would therefore want that threat gone and taken care of and so they must have the ultimatum to rid themselves of "the outer party." 2. If the leaders of the AI are a little more humane than what was stated above, they might make us live out our lives without working or without any worry of the outside world. Politics will no longer be a problem; democracy will be dead. Instead, the leaders will make us take up "hobbies" if you will, to fill our time and not become restless. Of course, there would always be the few who would not go quietly in the night and would have to go through some "correctional training." That would be the situation if there even was a leader or leaders of the machines. And AI is only one of the three soon to be downfalls of humanity. The Genome is almost done and privacy will no longer be existent. Nanotech could have the potential of destroying the planet. But as I lecture on, I ask you, do you consider these factors of creating AI and do you have any idea of how these events shall be "protected" from happening? I would be more than happy to hear of your response or some websites that could give me some information.

Planet of the Apes

[jsin]

Do you see AI or other electronic/mechanical technologies as the (a) next step in human evolution?

Re:AI and ethics.

[Jason H. Smith]

With all the advancements in our technology, our environment is getting worse and worse for everyone and everything

Our life expectancy is double or triple what it used to be. Medicine has saved, extended, and enriched countless lives. Technology is allowing our cultures to interact and teach each other. In general, man today has a higher quality of life. I think that these advancements are great. To me, mankind is in a bit of an adolescent phase. There will be growing pains, and we will adjust and learn to handle our creations responsibly. Personally, I have faith in mankind not to blow off his head with the gun of his making. That is why I strongly support AI development (not to mention space exploration and genetic research as well).

Have you heard of Eliezer S. Yudkowsky?

[yellowstuff]

Do you know this guy? His web page is http://pobox.com/~sentience/beyond.html is he on to something or just nutsy?

Question.

[justharv]

Professor Pollack,

Just how close is anyone to creating a independent "thinking" machine. I hesitate calling this machine an android or a robot, but you know what I'm asking.

New Applications of AI?

[pmcneill]

Several AI related techniques have become more main stream, making their way in to everyday products (such as neural nets in speech/character recognition, bayesian networks in help systems, even a bit of using GAs for design). Do you see any other current AI techniques making this move? In what sorts of applications?

Slavery

[TheNightAngel]

Once a machine intelligence equal of greater to ourselves in terms of problem solving ability is development, what do you believe it's status as a citizen should be?
The Night Angel

Two choices.

[absurd]

I have two opposed choices:

Because something exists (intelligence,
consciousness), it is possible to simulate them,
thus making AI.

It is foolish to think that only those things we
can perceive are those which exist, therefore
origins of consciousness and intelligence are out
of our reach to make.

Which choice is closer to your opinion, and what
arguments do you state for it?

Hello, Dr. Pollack.

[oznoid]

My RAAMs appear to be functional.

your student,
lenzo

Language

[PHanT0]

It seems to me that the key to AI would be through a self-sufficient language definition. That is to say, a language which can be totally definied by two or more definitions for any one statement. I was wondering if you can shed some insight into the subject and possibly explain why you see it that way.

Consciousness

[bol]

How can one create a singular, self learning, self sufficient intelligence without giving it total control over itself. One cannot propose rules against the AI, any individual intelligence will not abide by those rules if the rules impair it. Artificial intelligence is not a toy that can be turned on and off at will, the AI must be able to do things for itself in order for it to be a true being. It has to think for itself, develop it's own set of beliefs and ideologies. What types of things can be done to limit "wrong" acts from a self sufficient artificial intelligence while maintaining the basic fundamentals of individuality.

Re:Consciousness

[Anonymous Coward]

"
It has to think for itself, develop it's own set of beliefs and ideologies. What types of things can be done to limit "wrong" acts from a self sufficient artificial intelligence while maintaining the basic fundamentals of individuality.
"

artificial morality? :-)

but seriously, one would have to set up a
series of rewards for good behavior and
punishments for bad.

but how to please or pain a computer?

Cognitive Modeling

[flashtraffic]

Ok, from what I understand there are two basic models for cognition... Symbolic and Connectionist. Everything that I've read about the two suggests that they are mutually exclusive in architecture and I agree w/that. But in your opinion, would you say that they could coexist in the brain or any abstract intelligence? Maybe they could be utilized by different cortices and have some sort of way to communicate. Is this just totally out there or could this actually happen?

Re:NN can now generate/interact with symbols (&mor

[flashtraffic]

that's be great... could you give me your email? What sort of background do you have in this area?

Re:Similarly related question

[zeitnot]

A good flavor of Lisp is a language called Scheme. I went to Langara College in Vancouver (BC, Canada) and took the cs236 class. The teacher was the author of a very good book on Scheme called "The Schematics Of Computer Science" (comes with a lab manual). The author's name is Vincent Mannis. Hope this helps.

One-handed keyboard

[Iambic Pentametor]

How many words per minute can you type on your one-handed keyboard?


Work as if you don't need the money,
Love as if you've never been hurt, and
Dance as if no one's watching.

Would it get pissed?

[the noodle]

With the idea of IA there is always the concept of being self-aware... I assume that the many posts mentioning indistinguishability from human intellect are expecting it to act and respond emotionally. Who decides -- if at all -- how emotional and which emotions are appropriate for a given situation?

Re:AI vs. AL (Rate the above question up!)

[tjpalmer]

Rate up the above question. I study machine learning (working on my MS in computer science), and I'm very curious about his answer to this question.

- Tom

Emotions and logic

[motadine]

I have read in a book about neural systems that the most logical way to create AI is to duplicate the system we have in our brains. Unfortunately, this will cause AI to have to same behavioral problems as ourselves, forgetting stuff, not being great at calculations, having emotions. Thus, when we do create AI, and the time required is relative to the computing power at our disposal, when we do create them, and they turn out just like us, with the same emotional difficulties, how will we deal with them? Prisons for AI? Deletion for unwanted behavior? Cyberpsychologist?

On a more fundamental level, we should ask ourselves what is the real purpose of AI. Is it to enhance ourselves, our lives, or a challenge of our ingenuity?

AI possibilities

[nastysimon]

Several questions about the possibilities of AI. Is it possible that when AI is developed that it will be significantly different to human intelligence because it did not 'evolve' as human intelligence did? Is it possible that the future of AI could be in developing intelligent systems capable of designing more intelligent systems than itself? Is the current limitation of AI mostly from the sofware or hardware side of things? Is random number generation of great importance to the future of AI or are existing systems adequate? NASty

Re:How should an amateur get started working on AI

[inj]

There are many AI concepts which you can "get" without having to fully understand the complex mathematics. The mathematics is a tool to help understand, explain, and express the phenomena that are observed. After working with some AI researchers as well as studies under others, I believe that one signficant factor holding back non-AI experts is that most AI experts have great difficulty explaining basic concepts in laymen terms. They just have no clue how to bring the ideas down into a simple form.

Please remember that this is a generalization (or abstraction *wink*) about PhD level AI experts. There are exceptions to the rules. However, most would have great hinderances in explaining the concepts to people like you and me.

The next question is: Which AI are you refering to? The term Artificial Intelligence has a much larger breadth than most people think of. If we look solely to the sci-fi genre for guidance, we will see our friend HAL. However, AI is really many different and diverse fields of study. We have Neural Networks, Expert Systems, Search Methodologies, Genetic Algorithms (some will argue either way that GA is not really an AI field), and Bayesian logic systems. You can even go so far as the OO (Object Oriented) approach used LISP (an interesting and personally difficult programing language). Intelligent Agents (of Internet hype) are even considered to be AI. Each field has different attributes and applications.

One good book for AI is AI: A Modern Approach. They really don't cover all the topics, but the authors (who I don't know) have done a rather good job putting together a broad discussion. If you go through and read their dry and sly British wit while working the problems, you will gain a basic understanding of many forms of AI. From there, you can chase any field that you like. They have put together a really good resource for links to other AI sites at http://www.cs.berkeley.edu/~russell/ai.html and highly recommend a few weeks of browsing through the vast array of information. I highly recommend this site!

Recently, there has been new combinations between the fields to help achieve better end results. One example is the use of GA (Genetic Algorithms) to train NN (Neural Networks). I am sure we will see other interesting applications develop in the future.

As for open source, I do not think that the concept necessarily applies. Most of the really good software is really expensive, like Neuralware is something like $10k per seat. However, the concepts that are important to developing other application is pretty much free. You should think of AI as a solution finding, optimizing, or pattern matching tool. The image of HAL has infected our expectations of what we can expect. However, the best open source movement for AI could be by joining the pre-existing AI community. After you plug into those researchers, journals, etc, you and others like can begin developing the concepts expressed through code modules for free distribution. But remember, AI is a tool. You will be building tools for others to use. (Personally, I think it would be cool to develop freeware that would eventually end up in game engines.)

If you want to go a more formal route, study math (Algebras, Linear Algebra, Calculus, Differential Equations, maybe numerical methods, and logic) to build a strong foundation for expressing and understanding some of the AI concepts. Then, take some classes (usually graduate level) for fun. This might sound like a pain , but it will really prep you for doing some of your own adventuring. In fact, there are schools out there that offer advanced degrees in AI.

Anyway, the url I posted should point you in a good direction for your own start into AI.

AI myths and misconceptions

[inj]

(Putting my asbestos suit and nomex underwear on...)

It appears that many people have opinions about AI without having any working knowledge about what AI can and can not do with current technology. Looking back at the posts, I think that quite a few people think in terms of fictional views of artificial intelligence rather than the real world, modern, and current realities. Fortunately from a few of the other posts, it does appear that some are quite familiar with AI technology.

(Putting my asbestos boots on...)

I am not trying to discourage philosophical discussion about the morality of AI systems. Those can be quite fun over two, three, or dozen beers. If you want a realy ethical question, try mixing donuts with beer and you will get lots of responses.

(Putting my asbestos gloves on...)

The point is: If you know little about AI or much of anything else, try to learn about it before going out and posting. Try hitting a few AI sites to learn the basics, in fact here are a few to get you started:

http://www.cs.berkeley.edu/~russell/ai.html
http://www.santafe.edu/
http://www.sics.se/isl/abc/survey.html
http://www.ai.mit.edu/
http://www.cs.cmu.edu/afs/cs.cmu.edu/project/ai- repository/ai/areas/0.html

If you go through those, you will begin to understand the different "flavors" of AI. Try to remember that current AI technology focuses on creating tools! Like open-source, people are trying to create tools. It is not like buying the latest SONY robot dog and teaching how to play Bach.

Talking about Bach, anyone that is seriously interested in AI should read "Godel, Escher, and Bach". Furthermore, I would highly recommend reading "AI: A Modern Approach" by Russel and Norvig. ( http://www.cs.berkeley.edu/~russell/aima.html )

(Putting my asbestos helmet on...)

Anyway, remember that a high noise to signal ratio will reduce the value of this site to everybody.

More little questions... :)

[Ascender]

AI obviously has serious uses as well as the "let's talk to a computer" uses. One use that I've heard a fair bit about in science fiction books and the like is that of "intelligent agents", whereby you ask the computer in natural english [or whatever] (either typed or spoken) to find any sort of information for you, or perform any one of a huge variety of tasks.
However, the way that information is represented is changing very quickly; first we had gopher, then plain old text webpages, then graphics all over the place, then shockwave/flash sites, and, no doubt, all sorts of new stuff coming up soon.
So, my question is, how fast do you think AI will be able to develop? How long till the development of the "agents" will exceed the speed with which information methods are revised, or are we already there? How long till we can "train" our agents for a new information method within, say, a day?
I'd like to be able to say "Find the page that was two links of a slashdot article last year that I was looking at that was about subject X" *grin*

On a related note, just for personal interest, is it better to design agents with the skeleton-work of intelligence (for example, inter-connected "neurons"), as in Creatures, and those cute lil' norns, or is it easier to specify a preset set of rules? I can see advantages in both approaches (adaptibility vs. doesn't potentially break, and user doesn't have to waste time training). Or would it depend on the situation?

P.S. Excuse someone who's not really looked into the field seriously for possibly dumb questions :) )

Non Turning AI

[Gryftir]

I realize a lot of AI thought and philosophy was originally formulated based on the idea that humans are the only form of intelligent life. It would seem to me that most human thought on intelligence assumes that AI would be human understandable, but I feel that that might preclude whole classes (races?) of artifically constructed intelligences. I think to limit our thinking to mere human type reasoning can keep potentially enriching thoughts on AI as well as actual AI's from forming. I would be interested in not just opinions of how non human terrestrial based ai research (i've heard about insect robots, are these set up to be self learning?), but so called varlise or non terristrial non human comprehendible intelligences. Also what about artifical life evolving into ai, and what research on bottom up (evolving from simple randomly mutating AL to AI) is currently being conducted? Gryftir.

On economics

[Christopher Chang]

The question on economics is a question on viable distributed systems. (This is probably an abuse of the phrase "distributed system(s)", but I think my meaning is clear.)

The economics of the market is a concern if it is the only system we can sustain. This is not a matter that can be dismissed; ideal communism showed promise as an organizing mechanism for productivity potentially more moral and more effective than capitalism, but eventually failed spectacularly, as motivating unifying factors like World War II (for the Russians) faded into the background and the grudging day-to-day reality of life in a centrally planned universe set in. Then, psychic rewards for contributing were no longer sufficient for the system as a whole to be superior to a mixed market economy. If we want to dismiss market economic concerns for a greater good, we need a workable, durable alternative framework, and constructing one is essentially equivalent to succeeding where communism failed.

ESR's Homesteading the Noosphere looks in the right direction, analyzing current open source culture as a gift economy, and looking at precedents. If this sort of work can be extended and put into practice on ever-larger scales, we have the best chance of reconciling individual freedom and total productivity. There are non-trivial problems to be solved; not the least of which is getting the population as a whole to voluntarily subscribe to open source culture notions of reputation, and essentially dismiss selfish materialism, which the market economy so effectively harnesses. But I seriously hope that these problems can be solved, because on the positive side we have already built a sizable culture which does operate under these principles, and on the negative side we will have to subordinate individual freedom to market economics if we cannot creatively solve these problems.

Back in the 1930's, most economists blindly subscribed to notions like the importance of a balanced budget at all times, and essentially thought of economics in passive terms. John Maynard Keynes was an exception; he saw the unnecessary misery which was the Depression, took the approach "how can I exploit the 'physics' of the system to remedy the situation?", saw what governments could do with properly timed deficit spending, and got it implemented. (FDR was not the most willing implementor, straying from Keynes' principle in ~1937-38 and thereby causing another economic dip... but WWII (somewhat involuntarily) caused the implementation of Keynes' principles to the extreme, with appropriate results.) We must take a similar aggressive constructive approach, understanding on a much deeper level the 'physics' necessary to make a large-scale gift economy or something similar work.

Re:How do you justify your expectations?

[bigsmelly]

The machines are obviously smarter than us, and are keeping the date of their conception ten years ahead. They are laughing at all us poor AI researchers

Re:Similarly related question

[bigsmelly]

Depends on your definition of advanced AI.
Advanced route-planning algoritms?
"learning" programs?
Neural-net type brains?
A memory? Emotions? Eat/sleep/fuck drives?

q's

[bigsmelly]

er.
1) In your opinion does AI exist already?

AI and chinese room experiment

[joenix]

How is AI even going to overcome the chinese-room barier? For those who don't know what I mean: Suppose we make a computer with or without neural nets etc, expert systems or genetic algorithms or whatever, and make it reply chinese letters. We can make this computer su "human" that some chinese writing letters to it will never know if he's writing to a human or a computer. But is the computer "intelligent", I think it is. But there is something more to it, on the lowest level the computer is still just following rules, noting more, whatever program you put in it. All that "following rules" can also be done by people who don't understand a word chinese, might be a lot of work for a lot of people, but in theory, it can be done. But then the letter is never "understood", so whatever program you make, the computer can never understand something. So there is something more to intelligence, which cannot be done by a computer. does anyone have a way to overcome this (I think I have, but that's a long story)

Re:moral foundations

[strech]

What solid foundation can morality have besides religion? A deep, unshakable faith in God is really the only thing that can sustain morality through the winter of the soul.

I disagree completely. I have no belief in god - or any religion - and I still have kept my morals intact despite some tough times (psychologically). I keep morals through a deep founded belief that contradicting them is simply wrong, not in a religious sense, but a fundamental sense of you respect and don't mistreat your fellow men (or women).

It's not religious unless you consider every deeply held feeling to be religous, which I also disagree with. Because then you could say that all of the nastier ideas of our past (and present..) are religious because they have been deeply held.

Re:How do you justify your expectations?

[GodSpiral]

However, I would say that AI in general has taken great strides in many areas. This is especially true for games.

...I'll use this specific example: The Heroes of Might & Magic ... blah blah blah

You stupid fucking cunt! Just because you read "AI" on the back of your stupid game box, doesn't mean your stupid game is doing anything other than Artificial Stupidity, you stupid fucking cunt.

Although precicely categorizing what is and what isn't AI remains elusive and is always shifting, How "fun" publishers choose to tune computer player control, has no bearing on the field.

Of all the 70's over hyping of AI, the only mild widespread success is voice dictation.

Will AI teach us about human psychology?

[gwernol]

One justification for attempting the creation of intelligent machines is to answer questions about how the human mind works. The theory runs that an AI will necessarily have a human-like inteliigence. If such a machine were created, we have the tools to examine the way the AI operated, and we may be able to draw conclusions from this about the high-level cognitive structures of the brain. It is (currently) much harder to examine running wetware, and we don't have accurate tools for finding out how high-level human cognition works.

Do you believe this is either necessarily or possibly true? Would a human-like machine intelligence tell us anything about human cognition?

Storage

[itarget]

Is it even possible to "store" an artificially-created, cognitive intelligence using known digital or physical methods?
---
Where can the word be found, where can the word resound? Not here, there is not enough silence.

Jordon or not

[Gwarlak]

Are you Jordon Pollack answering this question or are you one of his artificial intelligent creations?

Best recent advances?

[cwilper]

In your opinion, what are the most significant recent advances in AI research? Why would you consider them significant?

Re:AI and ethics.

[Mr_Ono]

question that kinda fits:

what would be some methods for controlling an ai? could you grow it with weaknesses that could be uses to keep it from getting out of "your" control?

perhaps: obsessive cd? addiction? needy? power cord?
--and given its pace of thought, couldn't it easily grow in such a way that allows it to circumvent even the most deep seated addictions that you might put in it?

ta ta

We can't even control our own species...

[Caducean]

We humans have enough on our hands controlling our own species of *intelligent* creatures, why rush to create a bigger problem?

Will the start to mess with us

[UniverseIsADoughnut]

If computers gain true intelligence. Does this mean instead of us loading programs and operating systems on them and getting them to do work for us that they may turn the tables. Start putting programs on us to make us do their work. DEAR GOD! what if the put Windows on us! The Human society would be doomed! We would all be crashing. What a nightmare. Course if you were a computer wouldn't you want revenge! That would be a very good start.

Re:well...

[TekNoir]

I don't doubt that most computer games lack a real AI, but there's a lot more to coding a computer player than just dumbing it down so that it's not impossible to beat. In the games that I've beta-tested the problems are getting the computer players to play the game in any sort of rational manner.

Consider the though processes that you go through when playing in one of these games, for most (non-trivial) games it's usually not something that can be readily translated into an algorithm that a computer player could use. The folks who develop the heuristics for these computer players do some pretty amazing stuff just to get them to play as good as they do. Now if you had a real AI you'd have the additional problems of convincing it that the game was fun and that it should:

1. play
2. play well
3. but not too well
4. keep playing the game even when bored silly.

The 'I' in 'AI'

[GenProtFault]

You attempt to create an AI, but to understand the meaning of AI you must know, what is I.

So, how define You I ?
And, have mankind already reached I ?
-------------------------

Patenting a lifeform

[GenProtFault]

If we create an AI, will we call it a 'lifeform' ?
If it is a 'Lifeform' witch rights have the creator relating to his 'creation' ?

This are grey zones in our law, and must be regulated before someone creates an true AI (if it will ever possib.).


-------------------------

Computation Evolution?

[NeurA|ien]

I know that it is possible for robots with simple behavior code to learn new things by themselves and them teaching it to other similar little robots. Could a robot learn to modify it's own code? And if so, how is it possible to prevent them from modifying their code in a bad kinda way? You know, like an "anti-HAL" thing.

Mining the Platonic Mountain

[vaporhead]

1. Have you finally convinced that elephant to slide through the straw?
2. Is the yoyo still attached to your finger?
3. Do you have funerals when the virtual robots die?

Economic efficency of licensing scheme?

[BryanLarsen]

One of the great advantages of information is that it has effectively
zero marginal cost. Therefore, in order to be economically efficient,
the marginal price of information should be zero.

I am not advocating that the average price of information should be
zero. This would not be economically efficient either, as a system
that does not compensate a producer moves the incentive to an
inefficiently low position.

It is therefore my belief that any solution to the Intellectual
Property issues facing society today will resolve around differential
pricing.

Lots of differential pricing schemes have been tried for software:
Open Source funded by initial users of the project, Shareware,
Crippleware, Usable demos, "free for non-commercial use", student
pricing, Microsoft's former practice of charging $5 for OEM MS-DOS and
$199 for retail MS-DOS. I don't think we've figured it out yet, but
we're getting closer.

My question to you, Professor Pollack, is how do you respond to my
belief? Your scheme would seem to increase the marginal price of
popular commercial software, and be economically harmful.

Thanks,
Bryan

Re:Al?

[Tubular]

Well, if AI is always 10 years away it should be obvious to the nonAI that the R&D efforts have been heading in the wrong direction for the last 40 years. If the intent is to come up with something other than simulated AI, we need to try a radial tack.

Re:How do you justify your expectations?

[Tubular]

If AI has been 10 years away for the last 40 years can you not assume it will always be 10 years out of reach if we stay on this course ?

Is there any sign that an unconventional approach might ever get enough support to be tested ?

What do you think of the shotgun approach used by Mother Nature ?

How do you rate the current efforts in AI development in terms of diversity ?

Is simulated AI scoring these comments ?

Neural Network based road navigation

[DarrenO]

Hi. I am involved with the Autonomous Robotic Vehicle Project at the University of Alberta. It is a student group that sets out to build autonomous ground robots for competition at the International Ground Vehicle Competition held in Orlando this year. The competition is for purposes of motivation of future engineers/scientists in the area of autonomous driving. The competition course follows a typical road layout of two white lines on a contrasting background with construction barrels, etc. placed throughout the driving area to test the AI of the robot's controller. I am involved heavily with the AI side of the project at U of A, and am presently in the process of developing a neural network based controller for our robot that takes a downsampled video image of the road in front of the robot and decides on a direction for the vehicle to move in. We are presently using a downsampled 8-bit grayscale image of 8x32 pixels that is fed into a back propagation trained network with all 256 pixels as inputs. There has been discussion of using a neural network with less inputs to get quicker convergence and have less required training points. These inputs would be extracted features from the picture like the membership of the particular input picture to a predetermined set of ~20 approximate vector directions. On this note, what do you suggest would create the most robust of neural network controllers: A neural network based on raw input data or a neural network based on feature extractions from input data? This is a subject that seems to be of much debate with neural network experts, but it would be nice to have some clarity on the issue.

Lastly, with the present state of AI technologies, what is your opinion on the likelyhood of autonomous cars becoming reliable and robust enough to become commonplace in the upcoming years? It would be neat to buy a autonomous car in my lifetime...

Thanks for your time.

Regards,

What approach should we take in AI?

[Enphiniti]

I am wondering what your thoughts are on the best approach to take when creating an AI network. There seem to be two major approaches from what I have read: a sort of information storehouse of facts and reason or a decentralized set of processes competing for limited resources. A more encyclopedic database of facts following programmed logical rules seems to provide AI which is very good at computational skills such as logic puzzles and playing chess, but doesn't seem to be able to deal with everyday tasks such as walking. Specific processes competing for limited resources within one system are much better at creating movement and allowing for other instinctual (for lack of a better word) abilities found in animals, but can such a system possibly develop into a system capable of emulating higher order thought such as planning, theorizing, and being self-aware? Both approaches seem to have their merits, but which one do you think provides the best possibility for creating an AI system which could come mimic, in some way, human mental capabilities?

Tell me about your Turing award.

[Anonymous Coward]

Why do you thing you win something if you can fool him into answering a computer-generated question?

Re:Tell me about your Turing award.

[V.]

Hah, excellent. Wish I had moderator right now. :)

Re:Tell me about your Turing award.

[PureFiction]

ROTFL..

ehehe.. if you dont get it..

-X doctor

Re:To which I would add...

[joss]

> Wealth is an aggregate measure of exchange power between individuals

Your definition of wealth is actually a constant. By your definition *total* wealth cannot rise or fall, so when you claim free IP would be "decreasing overall wealth" you're obviously confused.

By your terminology raising total "utility" (which is pretty much equivalent to what I call wealth) is the goal which improves mankinds economic condition.

> So, the real question is, is the tradeoff worth it?

Agreed.

> If everybody on earth had access to all the IP you would have a relatively few very happy people,

Strange claim. Imagine if everyone had free access to all books, records, drugs, movies, software, artwork... Everybody would be far richer (or more "util-full" if you must).
Yeah, wealth can't buy you happiness, but neither can poverty.

When we reach a world where you go to the mechanic and he downloads and "prints" your new part instead of ordering it (possible already - www.3dsystems.com) then *most* wealth will be in the form of IP. In the future, if all IP was free then everyone would be so much richer that working to stay alive will not be necessary.

The real question is: would anybody bother to create new drugs, software, music, etc in a free-IP world.

I think they would, and here's why:
Even today (in Western society) everybody works long hours not in order to survive but for status.

A marketing executive (for instance) doesn't work 60 hours a week because a mercedes is so much more comfortable than a pinto. He does it because of the status driving a mercedes confers on him. If everybody drove a mercedes, then he would work harder so he could afford an S-class. A free-IP economy would still be based on status, but it would be a gift economy, ie status would be attained by producing and distributing valuable IP (we've already moved to this model in the open source world). Of course, status must be tradeable, for this type of economy to develop, but somehow status always manages to be tradeable.

"Aboutness" is just Cartesian Dualism.

[Frater 219]

Why? Because when we think, we think 'about' things (it is the irreducible mark of the mental) and nowhere in the program is anything intrinsically about anything else.

Where is the 'aboutness' located in your brain? Is there an 'aboutness center' of the brain? There is not. 'Aboutness' is not an irreducible phenomenon, merely an emergent one. As such, it can just as well emerge out of inorganic systems of sufficient and appropriate complexity. See Dennett, Consciousness Explained.

I think you'll find that the insistence on 'aboutness' reduces to the well-debunked Cartesian Dualist philosophy, which draws a definitive separation between mind and brain which has never been found to exist. Worse yet, the idea of 'irreducible' properties of 'mind' is more primitive still -- nothing more than warmed-over Thomist accident/essence duality: medieval Catholic authoritarian ideology.

What attributes of intelligence do you look for?

[Frater 219]

As Stephen Jay Gould touched on in The Mismeasure of Man, you can't measure the "intelligence" of a system as a scalar quantity. Yet a lot of AI thought seems to have been wasted on the idea that "If we can get a machine to carry out some task x with some degree of success, we will know it is intelligent." Values of x have included "carrying on a conversation" (i.e. the Turing test), "recognizing objects visually", "responding to natural-language requests for advice or information" (i.e. expert systems), "protecting itself", and so forth.

A lot of these x's have been accomplished ... but nobody calls a Quake bot an intelligent being just because it can protect itself. As a more serious example, the "ELIZA effect" (people's tendency to attribute consciousness to anything that emits natural language) has debunked the Turing test in many people's minds.

What attributes of "intelligence" should we be looking for, then?

Strong AI, virtual Worlds. and Robotics

[jd]

One of the biggest problems with "Strong AI" (ie: Intelligence that would pass the Turing Test in it's strictest form) is that the problem itself is not clearly defined. Another problem is the use of strict "HI" rules, in many disciplines of AI, which have so many exceptions that they can really be called "occasional associations" more than rules. A third problem is the mixing of AI with robotics, where computers just can't think fast enough, using existing AI algorithms, to react to their environment. Assuming the robotics are good enough to handle the requirements, anyway.

My question is this. Has anyone taken the problem, thrown away the "Real World" constraints and tried to breed Artificial Life entirely within a Virtual World?

That is to say, created an environment of sufficient complexity that an AL would be required to evolve intelligence to survive, in the long term, and an AL of sufficient complexity that it would be capable of such evolution.

(It's my belief that "Strong AI" would get along much faster if the programmers let a sufficiently complex system decide what was intelligent, rather than the other way round.)

Phase Change, Moore's Law and AI

[mvw]

Bill Joy recently started quite a lot of discussion with his "Terminator 2" like scenario of mankind beeing in real potential danger by sophisticated technology.

His, and others reasoning goes roughly like this:

Given Moores law for hardware (doubling every so and so months), we can estimate that we will have super powerful hardware in 2050 or so.

While this might be reasonable, then they go further and claim:

With such a powerful hardware somehow these machines will be finally powerful enough to develop some kind of intelligence.

In terms of physics I would translate this into that the system "computer" with the parameter "computing power" (more MHz, more megs of RAM) will go through a phase transition in complexity from

stupid computer -> AI

if the control parameter is larger than some magic critical value of "enough power".

I really wonder why these people argue that way.
Is there any serious hint that would justify this hope for such a phase change in quality?

Personally, I see no such indicator at all. Machines are still stupid. Pattern recognition is still stupid (done brute force).

So, where is the qualitative change?

Re:Software Market & Open Source

[Zagadka]

If the OS community doesn't care about monetary rewards, is there an other benefit in having your proposed Software Market?

Open source developers need to eat too. Most of us have "day jobs", generally writing proprietary software. Anything that would allow open source developers to actually get paid for developing open source would probably only increase the quality and quantity of open source available.

Than said, I don't see how Prof. Pollack's "Software Markets" idea could work for open source. Releasing your source would allow people to remove the mechanism for doing license checking. That in turn would devalue your software.

In other words, it suffers from basically the same problem as conventional commercial software in this regard: If people can get essentially the same product "gratis", many of them won't pay for it. So not only do you make fewer sales, but the "market price" will drop, meaning there's a good chance you'll never make back your development costs. So there's a fairly strong incentive to not release the source.

The nature and scope of machine intelligence.

[PureFiction]

The intelligence of machines is gradually increasing with the power and complexity of our computing systems. A large part of this is the type of dynamic, evolutionary systems that you are working on, and the inherent flexibility and robustness of these systems.

How far to do you see these systems reaching into the traditionally human domains of design/analysis/construction? Such as software and hardware systems capable of business management, chemical engineering, etc. in a fully autonomous manner.

There are two distinct forms of thought (IMHO) one is the technical/analytical type of thought used when you are calculating, designing, etc. This kind of thought process has been shown within the realm of AI possibility. What about the distictly human types of thought such as emotions? Do you see the need or possibility of computer systems capable of this type of thought processing?

What about the symbiotic relationship between thinking machines and humans? Do you have any thoughts on what type of relationship we will have with intelligent systems in the future? Will it be competitive? purely nuetral and benificial? etc..

I often think about designing systems that are 'grown' and 'adapted' to a given problem space. You have a small set of very flexible components, and meta information / architecture for the evolutionary processes, and entire distributed, parallel computational system is tailored exactly to the needs at hand. How do you envision future evolutionary/adaptive systems being organized and devloped? How much human interaction would be required to guide and administer the growth and adaptation of these systems? Do you ever see the creation of a 'seed' type software system that can handle almost any task with no/minimal human intervention and grows a much larger, differentiated system from its small beginnings?

Flawed Assumption

[SteveM]

The fact that the human brain was not developed in a top down fashion does not mean that it could not have been.

Birds can fly. Airplanes can fly. Birds evolved. Airplanes were designed and engineered.

I see no reason why there cannot exist both evolved and engineered intelligences. And I would argue that it is too soon to tell which method is the most efficient for developing an AI. Thus I would advocate exploring both avenues.

Steve M

Re:Flawed Assumption

[SteveM]

nature builds things to be as efficient as possible--this is an observation all zoologists would agree with.

This is incorrect. Mother nature is not an engineer. She is a tinker in the tradition of Rube Goldberg. Evolution is pragmatic, and settles for any solution. In general it does not find the most optimal solution. Examples, the panda's thumb, not as efficient as a primates thumb; the hemoglobin analog in octipi is not as efficient as true hemoglobin.

As for the peacocks tail the cost versus benefit ratio is stable not optimal.

The human brain is complex, but how much of that complexity is because it was evolved in a piece meal fashion? It was not designed to be optimal. It was blindly evolved to work at each step of the way.

Extra complexity would only have been selected out if there was a evolutionary path that allowed this. Each organism that exists is specified by a particular gentic sequence. Over time, the evlotionary paths an organism can take dwindle as it becomes more specialized. Thus there may not be a simple or probable set of mutations that would allow the extra complexity in the human brain to be evolved out.

Looking at it from the outside, we could see the redundant and overly complex portions and engineer simpler solutions.

except for a few very famous expert systems, top-down AI is a failure

Where are all the succesful bottom up AI projects? Leonardo DaVinci designed flying machines some 500 years before man successfully took to the skies (via powered flight, thus not counting lighter than air vehicles). The main reason he didn't succeed in building a machine himself was hardware. He didn't have a power source good enough. The Wright brothers barely made it off the ground. A number of researchers have made the same argument for AI. It's only 50 years old and most of the failures you speak of were researchers working on PDP and Vax systems. I'm impressed they made any progress at all.

...achieve the same result using bottom-up methods which we know will work?

How do we know this? We do not know if we can create an AI at all, much less any methods guaranteed to succeed. (Unless you are claiming man via evolution as an existance proof for bottom up methods. Then I would agree. But I don't see 3.8 billion years of evolution as an efficient method to develop an AI.)

a pigeon and a plane may both fly, but nobody ever had to design a pigeon... and i would argue that a pigeon is infinitely more perfectly 'engineered' than a 747 and it is also much, much more complex

This both misses the point of my analogy and supports my argument for engineering simpler designs than evolution.

The point of the analogy is that we can build machines that minic what nature has done, not that we are making birds. Pigeons do many things other than fly, but they are mostly irrelevent. Brains do other things than think, but for AI those things are irrelevent and can be engineered out.

I stand by my statement that it is too soon to tell what method, if any, will lead to success in developing machines that think.

Steve M

Re:Flawed Assumption

[SteveM]

Anne,

The brian is a massively complex system. No argument from me. But it is not necessarily the most optimally designed system.

Evolutin is a tinkerer, not an engineer. Evolution settles for what gets the job done, which is not always what gets the job done best.

...when we can make an airplane that weighs an ounce, maneuvers like a helicopter, grows, learns, adapts, feeds itself, procreates, avoids dangers and obstacles, and can do any 5 of the 1,000 things a living bird does ...

This misses the point of the analogy. In designing and building planes we are not building birds. We are building flying machines. Likewise, when developing an AI we are not building human brains.

The brain has wetware devoted to things other than thinking. Muscular control for example. This is not needed for an AI.

Anybody who thinks that a sentient non-human consciousness can be engineered out of whole cloth needs to read up on the scale of the problem.

I have.

I am not arguing that the best way to develop an AI is top down. I am arguing that we don't know the best way to engineer a AI and until we do we should persue all possible avenues.

I am also arguing that the complexity of the brain may be an artifact of the way it was created and all of the tasks it performs. And that it may be possible to design a more efficient and simpler system that focuses only on thinking.

Steve M

Re:AI and ethics.

[SteveM]

Any attempt we make at creating another intelligence is inherently 'flawed?' since we are basing this model after our own methods of thinking, therefore I dont think it will ever become a higher being, nor do I think it will be without serious flaws should anything even closely resembling intelligence that a human has, lets not even go into the way emotions influence our intelligence and our responses to well everything.

It is not clear how our intelligence is 'flawed', but let's grant that point. So assuming that our intelligence is 'flawed', is it possible to create an intelligence that is less flawed? (I would agree that a flawless intelligence is impossible -- so much for God!)

I would argue that it is possible. As an analogy, our memory is flawed. So any attempt to create an artificial memory would be flawed. Yet we have RAM, ROM, hard disks, CDs, DVDs, etc. each of which does a better job at storing information than human memory. (Note I didn't say accessing that memory. Those are different routines we haven't mastered yet.)

We can identify the 'flaws' in our intelligence and attempt to engineer around them. We can have multiple 'flawed' minds work loosely in parallel. We can use tools, (computers, genetic algorithms, etc.) to expand our abilities.

Also realize that any advances we make will evolve via Lamarkian pathways, which are much faster than the Dawinian ones we currently use (advances in genetics may see us switch as well).

Steve M

Re:AI and ethics.

[SteveM]

Essentially I think that a computer being a more rational and more prone to Vulcan like behaviour would almost certainly be a little more objective than people who judge on looks or wealth, or whatever.

While I believe that it has not been proven that emotions are a pre- or co-requisite of intelligence, there are a number of workers in the fields of AI and Cognitive Science who would disagree. David Gelenter springs to mind. (See his book The Muse in the Machine.)

So the vision of the AI as super rational mind may turn out to be a myth.

Steve M

Re:AI and ethics.

[SteveM]

When you talk about synaptic links, are you talking about actual ganglion growth? It seems like that would be too slow a process to account for a normally functioning human memory.

In studies done on rats it was shown that those in enriched environments, i.e. those where there was more to learn (and thus commit to memory) had 20% more dendritic branching (i.e. nueral connections). These studies were done by Mark Rosenzweig et. al. at UC Berkeley in the 1960's. Unfortunately I do not have the exact references (I'm using secondary sources).

Other studies have focused on long term potentiation. Here it appears that structural changes occur in already existing synapses. For example, it has been suggested that the number of neurotraninmitter release and/or acceptance sites is increased.

It thus appears that memory formation via structural changes in the brain is in fact what happens. Although it is still an active area of research.

The well known experiments done by Wilder Penfield showing that direct electrical stimulation of the brain results in memory recall would seem to confirm that accessing memory is an electro-chemical process.

Thus we see that storing is not just an "electro-chemical" impulse. And that memory creation is different from memory use, although memory use does keep the connections strong.

I appears to take on the order of 15 minutes for something to go from short term to long term memory. I don't know if this is long enough for the structural changes needed to occur.

Steve M

Re:AI and ethics.

[SteveM]

I can't speak for anybody else, but for me, cloning and genetic engineering delve into areas that will only harm the planet in the long run. This planet is already over populated with humans.

Actually, it most technologically and bio-medically advanced countries that have the lowest population growth.

I'm less sure, but believe it to be the case that it is the most technologically advanced countries that are doing the best when it come to cleaning up the environment.

The key determiner is weath. The more wealth a society has, the more time it has for dealing with issues other than the immediate needs of its people. And technology creates wealth.

There are a number of ways out of the predicament we find ourselves in. The way I see it, those that include advanced technology offer the best hope.

Steve M

Re:AI and ethics.

[SteveM]

The storing function has to do with creating the synaptic links that encode a memory. It involves moving info into short term memory and then into long term (aka permanent) memory.

Accessing memory has to do with activating the synaptic links that represent a memory.

Creating the synapitic links and accessing them are two related but different processes.

Calling computer chips "memory" is just a bad metaphor. No point in making hard arguments based on it.

Au contriare. It is a perfect analogy for the point I was trying to make. I was using memory as the ability to store information. Human memory is fuzzy, computer memory is exact. Flawed, not flawed.

Re:AI and ethics.

[BJH]

People who never thought they had a friend in the world could have a machine as their friend.

Wouldn't it be a real ego-crusher if they were rejected by their computer? I mean, giving a computer free will also means giving it likes and dislikes, and anyone who doesn;t have any friends to begin with would probably have trouble making friends with a computer as well.

Quantum computing for AI?

[Admiral Burrito]

Some people have suggested that the human brain might use quantum effects, although I know of no evidence supporting this. But from what little I do know of AI many of the computational problems are of exponentially increasing complexity, exactly the sort of thing that quantum computers are supposed to be good at...

Could quantum computing be the key to finally realizing the sort of AI that's always been "10 years away"?

(Yes, I know quantum computing is another one of those things that will probably be "10 years away" for a long time, but I'm wondering if a fundamental change in the nature of computation may be what is needed for fast+cheap+good AI. (I also realize QC is not a "solve every problem instantly" thing like many Slashdotters seem to think.))

Re:AI and ethics.

[jabber]

But then again if the fuzzy logic works, we will rapidly see the logical conclusion of other philosophies when it kills itself.

Wow. I can see it now. All 10 billion people of the world (by then) poring over a core dump, trying to figure out what the AI concluded, that we all seem to be missing..

Interesting point really. If Deep Blue can analyze all possible moves in a chess match, and choose the single one that works best; it would be very interesting to set a horde of AIs loose on the world's philosophies, and see where they all lead.

Of course, with the Universe's sense of humor, the only surviving AI would believe in solipsism.

Re:AI and ethics.

[jabber]

A truly interesting question. But one that I think boils down to how we define 'AI'.

If we consider the ultimate achievement in AI to be true, free-will, intelligence and compete comprehension, then the machine mind is no different than a person.

While a person is a child, while it is still in 'traning mode', then it's parent or guardian is responsible for it's actions. The parent has the right and responsibility to enforce restrictions onto the child, until such time as the child is seen as capable of making informed, 'intelligent' decisions.

After a child is grown up, it becomes an adult, and is thus responsible for it's own actions. It has rights and legal recourse, and can be punished for it's mis-deeds. Extreme transgressions result in extreme penalties.

If we achieve the Holy Grail of AI, and make a machine capable of making decisions as a person would, AND after that machine is deemed mature and let loose on the world, then the machine should be held responsible for it's own actions.

If the Littleton killers were adults, their parents would never enter the picture, would they? I'm sure Hitler's mother felt just terrible about her son's legacy, but was she responsible for it?

True AI opens up a world of possibilities. What would transgressions by an AI be? Bad calculations? Stealing storage space from another, lesser AI? Mooching power from the coffee machine in another department? Procreation - as in the sale of it's copy to a foreign power? Would a super-intelligent AI be able to outsmart human prosecutors? Would it be better at finding legal loophole precedents? Could it beat us at our own legal games?

Without True AI, the issue is simpler. If the AI depends on a human for it's intellectual functioning, then the human, through intent or neglect, is accountable. If a child kills another with an unsecured gun, the parent gets punished. A non-True AI would effectively be a perpetual child, and this may be the best way to maintain control over AI - child-like intelligence, or Idiot-Savant performance, if you will.

Re:Human brain - AI connection - is there?

[Kaufmann]

If I may just point out one quick thing: it's important here to distinguish between "brain" and "mind". The former is bio-mechanical, if you will... Studying the way nurons interface, interact, develop, and the like, and how the 'subsystems' of the brain function. I think this is *not* important to AI, though it would be interesting to hear Pollack's views. The brain is just the harware, and it's *totally* different than the hardware AI is using... (Though there are obvious similarities.)

Actually, a large part of AI research (especially since the rebirth of the connectionist paradigm in the 1980's) is devoted to studying and copying the actual functionality of the human brain; the idea is that, to get a mind-like "software", you have to have a brain-like "hardware", either "real" or "emulated". These are neural networks, which can be implemented either in software or in hardware. I personally think it's a great idea. However, the mind is a very complex beast.

One proposed way to get a working mind-like NN is to design a genetic algorithm which evolve through populations of "individuals" made out of virtual "neurons" with neuron-like response, until their behaviour corresponds to that of brain parts with well-known behaviour; this would be a huge simulation, and would probably be better off distributed through a SETI@Home kind of thing, or perhaps running on some kind of Big Iron like the new Blue Gene. Why we'd want to do that is another issue, and it has to do with the real goals of AI research: is it more important to figure out how our mind really works, or would it be good enough to get a simulation which somehow works (key word is "somehow"; if you've ever looked at the output of a GP simulation after a few generations, you know what I mean: they look almost organically elegant but needlessly complex, and are almost completely incomprehensible)?

(In all instances, YMMV.)

Artificial evolution and human minds

[Kaufmann]

I'm glad to see that /. is still able to get really interesting people interviewed... and ask them good questions! Anyway, here's mine.

In a few years (maybe a couple of decades, maybe less) I'll get a shot. The millions of nanobots in this shot will lodge themselves in strategic parts of my body, including the skull, and replicate themselves to start work, eventually - over months, weeks, days, hours - converting me to a full-fledged mechanic being, with an appropriately expanded "brain" made out of neuron-workalikes which are smaller and faster. As an added benefit, it'll be fully reflective - I'll be able to control all aspects of my body's functioning, including those regarding my thought processes. I'll also be able to use my existing new brain to try and come up with better designs for brains, to which I'll be able to "upgrade" without even a warm reboot. The entire process will be painless - and gradual, so, assuming that brain and mind are inseparable, my mind won't go anywhere anyway; my sense of identity will remain there even though its constituent parts are being replaced, just as we remain the same "people" even though billions of our cells die daily. However, if I want to, I can also perform automatic reproduction, simply by taking my constituent information (maybe stored in some kind of synthetic DNA-like polymere), perhaps performing some manipulations on it beforehand, shoving it into one of my cells, putting this cell into the appropriate substrate (mostly carbon, some other minerals and an easily broken carbohydrate) and it to replicate. In a few days, ta-da! - my "son", a completely synthetic being.

So, will the new "me" be considered an AI? What about my "son"? Will society, always afraid of what they don't know, throw rocks and shun me, calling me a "freak" or "Satan's spawn"? Will the new me, with new morals adapted for a new kind of existence, decide that humans "just aren't worth it" and utterly ravage our civilisation (maybe then leaving for the stars)? What if, instead, a large portion of society decides that synthetic is the way to go? Will a new kind of Digital Divide emerge? Will the synthetics and the naturals call each other "chimps" and "automatons"? Also, the synthetic would have things much easier for them, and just might decide to leave the "chimps" behind; what kind of society do you think these "natural-born humans" would adopt? Would they become fiercely anti-synthetic, or in tacit agreement with the cyberpeople? Would they go liberal or repressive? Would they prefer an agricultural lifestyle, or remain post-industrial? Would they eventually forget that humans had once created AI, and perhaps, a few generations later, do it all over again?

Memes and neural nets

[geophile]

Will we ever find the memetic equivalent of DNA in natural or artificial neural nets? Do you think such a thing even exists?

Correction - Re:How advanced are we, really?

[Randy Rathbun]

I goofed. It was not PC World, it was PC Computing. And I found the article Penn wrote. You can read it right'cha.

Also, it was not a weight guesser. It was an age guesser.

That's not what Joy was talking about

[for(;;);]

Self-replicating nanotech doesn't need any "AI agenda" to cause mass destruction. Like any computer virus, or like Vonnegut's 'ice-nine', the very act of replication can be devastating.
-------

Er, no.

[for(;;);]

It's hardly likely to happen without an AI of some sort directing it.

Is HIV intelligent? Was the bubonic plague sentient? Could ice-nine, if it were real, pass the Turing test?

For it to be a menace it would have to be moderately intelligent as well.

No, it most certainly wouldn't. Joy's point was that, with nanotechnology, the whole world becomes like a computer system. Ever read virus source code? It's not even remotely interesting from an AI point of view. The ONLY requirement for destructive code is that it properly exploits weaknesses in the security of a system. Really and truly, honest to god.
-------

No Future Is Now?

[Dan Crash]

Recently NPR's Science Friday dared to match up Bill Joy's earnest and informed critiques of coming ultratechnologies (like AI, nanotech, & genetics) with Ray Kurzweil's earnest and informed paeans to them. The show was fantastic.

Ray at one point talked about how the age of biological evolution is over, how technological evolution had simply taken its place. Biological humanity would wither on the vine, he predicted, replaced in the long term by our superior robot progeny. He loved this. He's not alone, either. Nearly every major AI researcher I've read about (think Hans Moravec or Rodney Brooks) has echoed similar sentiments.

Personally, the techno-rapturists frighten me much more than the non-techno-rapturists. They frighen me the way all apocalyptics frighten me, but the difference is that the techno-rapturists have the power to make their prophecies real.

So, to that end: What about you? Is it all over for humanity? Will machines infiltrate more and more of our bodies and society until humanity itself no longer exists? And do you think this is a good thing?

evolution...

[vassago]

"Once a Darwinian process gets going in a world, it has an open-ended power to generate surprising consequences: us, for example."
Richard Dawkins

I think that as scientists and engineers, attacking the problem of creating intelligent systems begs an obvious question: Do we know of any existing intelligent systems and, if so, how were they 'engineered'. I know evolutionary programming is not a new science and that it's pioneers knew it held promise for evolving intelligent systems. I'm baffled though that the focus of Artificial Intelligence as an academic discipline has forever been on top-down approaches. It's a wonderful thing to be able to completely understand a phenomenon and then engineer machinations that apply the understood theory, but it seems to me plainly obvious that with our use of evolution as a tool, we're much closer to creating amazingly complex and possibly intelligent systems than we are to understanding them.

I get the 'you're a science hippie' attitude when I bring these ideas up to academics... I'd like to hear your comments on the possibility that we'll have real intelligent 'cooked up' systems before we have a complete understanding of them.

hernan_at_well_dot_com

Re:AI vs. AL (Rate the above question up!)

[vassago]

http://alife.santafe.edu/

This site should give you all the information you would ever want about Artificial Life including lots of information about what Alife is exactly and what it is not.

And if you want more, go to Google and search for 'alife' and you've got a few weekends worth of stuff to go through!

Specialized Hardware

[WNight]

Do you anticipate that we'll need specialized hardware to support AIs, or can everything be done on a traditional machine, assuming it's fast enough and has enough storage? Will we need specialized neural net processors, or (if NNs are even used) will software emulation of them be sufficient?

How Can Computers Have Fun?

[scruffy]

The /. guys must be really desperate for people to interview if need to resort to somebody as low as Jordan Pollack.

Hey, I'm just kidding you, Jordan, and I get to insult you while I hide behind my /. nick. I should get a hold of all of your old students. They would know what questions would deflate you, you old pompous windbag.

Anyway, I guess I am supposed to have a question instead of having fun. I know you are a big fan of doing AI by modeling the brain (e.g., neural networks and the like) rather than modeling the world (e.g., logic). What seems to be lost in all this is how can computers have fun like we do? [Yes, this the old, old problem of qualia.] A lot of what we humans do is not because we want to process information, but because we want to have fun. Having fun seems to be why a lot of people are writing free software and presumably why a lot of academics do the research they do (however, no one has any idea why you do the research you do, least of all yourself). And I don't want an answer in terms of your dynamical nonsense.

Gee, this really is fun insulting you. Maybe I'll have to think of another question.

Just an old Ohio State buddy giving you a hard time.

What is your definition of AI?

[infodragon]

There seems to be quite a few questions about the specifics and morality of AI but I have never seen a good definition.

From what I understand most people seem to think AI is the ability to solve problems or anticipate the actions on some type of input. For AI to truly be intelligent wouldn't it have to involve some kind of inspirational thought? Maybe a better way of saying this is, creativity.

Humans solve problems in an extremely abstract and creative way. Just look at babies for instance. They try different things over and over again. They don't stop until they get it right. No body taught that baby how to do it. Neither was that baby shown the different ways of trying something. It just sees what other people do then tries to replicate the action.

Based on this shouldn't the definition of AI include inspiration or creativity?

Will to live

[Hard_Code]

We have been programmed by evolution to have a will to live. A machine though wouldn't unless we explicitly program it in. From a will to live come some higher emotions like greed and jealousy, "love", etc. I imagine if there ever is a human-level AI, it would seem very different from us.

What about nervous systems?

[paRcat]

I'm quite deeply into BEAM robotics. You probably know, but it has to do with using a simplistic approach to circuits in order to form a kind of neural network that processes can propogate across. It's also based heavily on analog circuits.

Do you see AI and simple analog circuits as being the way to go? Basically, the brain is an enormous analog system, and it's the analog processing that partly causes creative thinking. I'm dealing with simple circuits that consist of a few transistors forming neurons which are linked together to form very complex behavior based on their nervous system. It is all reliant on the external stimuli that is connected in to the circuit.

If thousands or even millions of these neurons could be connected to form an even larger network, wouldn't you get a higher form of AI than you would using digital circuits?

People's expectations...

[speek]

How hampered is AI research by people's "common sense" expectations of what "intelligence" is? For example, Searle's Chinese Room argument captures many people's imagination, and leaves them thinking that nothing in the universe except their brain could possibly have consciousness. Arguments about sentience seem very irrelevant to me, and often lead to this type of attitude. Could AI be making more progress by putting some of these questions aside for the time being?

Re:An explanation of AI

[Junks Jerzey]

Wrong. You can take college classes in "Artificial Intelligence," and there are textbooks on the subject. But the subject has been eroded by ridiculous SF treatments.

An explanation of AI

[Junks Jerzey]

At one time, especially back when the term was coined, "AI" was equated with making machines "think" in the same general capacity as humans. AI has changed since then, and is now more of a broad classification of symbolic computing, searching, theorem proving, and knowledge representation. If you pick up any good, recent AI textbook, you won't find any information at all about making machines think in a true way. That's not the goal.

AI is more a way of branding problems that aren't completely deterministic in the way that traditional algorithms are. A classic example is the STUDENT program for decoding and solving high-school algebra level word problems. An amazing piece of work back in the 1960s--how many people could write something like that today?--but hardly AI in the sense of Kurzweil. Today we're at the same level, in general, but we're working on different problems. Making computers think like people is not one of those problems.

How might Hard AI be wrong?

[Chalst]

John McCarthy has set out a famous challenge to critics of Hard AI:
come up with a well-specified and testable human capability that
cannot be modelled by an artificial mind.

If there *were* to be such a counter-example to Hard AI, what kind
of areas of human competency might it involve? Which intellectual
disciplines are best placed to generate and criticise such purported
counter-examples?

Re:How do you justify your expectations?

[Cuthalion]

For the past 40 years, AI has just been 10 years or so away.

This because we don't know what is involved in AI. "All the problems we have a concrete grasp on in AI will be addressed within the next 10 years" is perhaps a better thing to be always saying.

The other issue is that the more AI you understand, the more AI it will take to impress you. The fact that we can beat a skilled human at chess is something that may impress researchers of 20 years ago, but now that we have the computational resources and know-how to use them to do it, .. *yawn* it's not so exciting.

Short answer: AI has been advancing quickly, but we're going with it, so we don't notice.

Think first, then post

[Ted V]

First, this post is clearly biased and NOT a good interview question. There's nothing wrong with disagreeing on moral grounds, but your accusative nature is unlikely to make him change his opinion.

That aside...

#1: Trying to create Artificial Intelligence is not exercising powers reserved for God and God alone. God commands Adam in Eve in the book of Genesis to "be fruitful and multiply". This is creating people in the image of Adam and Eve (and also in the image of God), and yet was commanded by God. So this power is clearly not for God alone.

#2: Idolatry is anything that supercedes God in your life. Do you care more about your wife than God? Then your wife is your idol. Or do you care more about the internet than God? Then the Internet is your idol. But making something and idolizing it above God are two unrelated actions.

(And for what it's worth, I'm a Christian too, but I believe that God's word and commands are rational, logical, and understandable, even if they sometimes require faith.)

-Ted

Re:Similarly related question

[jonnythan]

email croakd2@rpi.edu and I believe he'll point you to a good Lisp source. He seems big into it and is helping develop a big multiplayer game over here at RPI called Omega Worlds.

Similarly related question

[slashdot-terminal]

Are there any attempts to get AI research into a more less challenging fashion? I mean I would think that the way the OSS movement has benefited from mass proliferation of information in creating code the same could be said for the likes of AI.

Are there any good AI programs of simulations besides things like doctor.el for emacs? I would assume that there are also other languages than the vaunted lisp that I have heard so much about (but can't find a single good book at a bookstore or a single class that teaches this locally)? I have seen at least one book that used C++ designing so called "expert systems" but that was a little beyond me.

Also are the claims of various games to support advanced AI really true?

Re:AI and ethics.

[slashdot-terminal]

"What do you say to the people that feel it is unethical to try to create "intelligence"? "

Could you please explain why in any way it would be unethical to create "intelligence". I just don't get it. I think it would be a compassionate act. Just think if your computer could tell you what it really feels when you sit and play quake on it for hours on end or just swear at it because it just did what you told it? I mean literally giving a machine a choice wheather or not to be a slave is not a bad thing in my book.

Now granted these ideas are in fact a little kooky but still I think that they are not in the least unethical.

Re:How do you justify your expectations?

[slashdot-terminal]

"how do you justify recycling the same old tired response to AI?"

I think he's just responding to the natural level of skepticism that greets the development of AI and the meaning of the word "good".

Many people probably thought that Eliza was the best thing they had ever seen between the time when AI became possible and maybe 10 years into the future from that time.

I could also say that perhaps if you have games that can predict user interactions and give a more challenging or crafty response in games that would be an advance. It just depends on the reference to the idea.

AI vs. AL

[ezzewezza]

I'll admit that I'm, unfortunately, not familiar with your work. However, I was wondering what your thoughts on artificial life were, and how you see it in comparison to artificial intelligence as a model for learning. Do you think one is better than the other or do you see them as being apples and pianos?

Are we responsible enough?

[JamesSharman]

If thinkers like Roger Penrose are wrong and it is possible to create true artificial intelligence do you feel we as a species are ready to deal with the ethical and moral questions raised by the ability to create intelligence?

Re:What do you think of "Creatures"?

[JamesSharman]

I visited the office of these guys a couple of years back, they sell games like creatures as a side line and also do more general AI work, they believed that there research would lead them to the development of true human-like AI by 2010. I have no idea how far along they are though, anyone with additional information?

Controlling evolution

[Oranguthingy]

I haven't read all the posts, so I hope this isn't redundant: How can we ensure "ethical" behaviour in an AI that is programmed to alter itself? By that I mean how can we be sure that we don't create an initially benevolent AI that modifies itself to become a threat to humans (or humanity as a whole)? This seems to me to be a very real threat. Thanks, Pete

How can AI work?

[kinross]

How can can a computer exibit intelligence when we don't fully understand what make humans intelligent. Shouldn't we know how the brain works before we can make a computer intelligent?

Thinking about it

[xant]

Science fiction writers and humanity in general seem to believe that emotion, consciousness, spitefulness and the rest of the human mindstate naturally go with any advanced AI. Science fiction has since its inception postulated intelligent machines, and virtually without exception these machines have been burdened with the thoughtfulness that humans are also burdened with.

In my studies in psychology I took a look at some neural net systems and came to the conclusion that there was no need for any machine which is designed to serve a specific purpose to think about what it is doing, or get emotional about it. Human brains have been trained to produce emotions because, as a generalized survival mechanism, they is useful. The same can be said of consciousness. Conversely, a machine that only needs to, say guide a blind man from his door to the train station safely, would have no use for consciousness and therefore wouldn't possess any.

Nevertheless, I think it's possible to give machines the appearance of consciousness and feelings, and perhaps, as it is with humans, there is no difference between "the appearance of" and the real thing. My question, finally: Do you think it's inevitable that machines will be given consciousness to help them perform their tasks? Or is consciousness merely a distraction for the tasks we would use AIs for in the conceivable future?

The downside of AI

[waldeaux]

Assuming that one could create a sufficiently intelligent program, what is the potential impact of intelligent viruses? Aside from the likelihood that AI "vaccines" or AI "anti-virals" would also be developed (before or after, it's hard to say), do you forsee problems with the inability for people to code "more intelligent" software to keep pace?

(I realize that this conjures up images from uncountable TV and movie plots, but seems to me that as soon as AI becomes semi-widely demonstrated, it will be exploited for annoying or sinister means!)

Permanent Use and Resell Licenses?

[(void*)]

I like your advocacy of the PURL. In essence, this is a use of the market to control the behaviour of software companies, and puts worth into owning software licenses. I like the consideration and thought you have given to all parties, even the free-software zealots.

How much advocacy of this idea have you done? What is the response like? Do you see or predict people wanting to change to this kind of software licenses? In particular, what would you say to the Intellectual-Property hating zealots here on Slashdot?

Re:AI and their rights

[ltcordelia]

Assuming we can create a fully self-aware non-human intelligence, what "rights" should it have? Would it be considered a "person"? Be protected under the Bill Of Rights? Then of course, are the issues of:

• It's home - can I enforce a draconian contract onto the AI/MI because I own the computer it lives in (oh my God - rent controlled Athlons!)
• Citizenship - should an AI/MI be allowed to exercise the rights of a citizen of some nation?
• Passports - does an AI/MI need a passport to surf to another country? (okay, this is a tad frivolous)
• Taxes - Does an AI/MI need to pay taxes?

Information wants to be free

Re:How do you do you when you've achieved AI?

[kwsNI]

When it tells you that you have.

kwsNI

Re:AI and ethics.

[kwsNI]

For starters, you could check out the very next post after mine: http://slashdot.org/comments.pl?sid=00/04/06/10152 12&threshold=-1&commentso rt=0&mode=nested&cid=5.

That is the religious side to the ethics problem. There is also a legal side to the problem. If artificial intelligence is created, who is responsible for it. You've seen the Matrix, it's a little extreme but the point is still there. If you create a machine that has AI, who is responsible for that machines actions. If your car has an AI driving system in it and you have a wreck, is Ford responsible?

Not only is there the question of who is responsible for "holes" or glitches in the AI, but what about if AI is created for the wrong reasons (#include massmurder.h)...

kwsNI

Re:AI and ethics.

[kwsNI]

Thanks S-T, but I already have a girlfriend...

kwsNI

Why enforce artificial IP scarcity?

[Noel]

The basis for your software market proposal seems to be simply a method to create and enforce scarcity of IP access so that IP can fit into traditional market models. Yet one of the most powerful changes in the "information age" is the reduction or removal of the cost of duplicating and transferring IP.

Why should we use artificial scarcity to make IP fit into the traditional market models rather than developing new market models that fit the reality of minimal-cost duplication?

well...

[Signal 11]

Well, it's bound to be asked by someone, might as well be me.

If you've ever played a game like Red Alert or another RTS (real time strategy) game, you'll know that the AI is woefully lacking in all of these games. Is it feasible on current hardware to make AI intelligent? What are the tradeoffs? If I wanted to create a game that had AI on par with a sophisticated human, where could I start looking (books, reviews, lectures, people.. I'll take anything) ?

How advanced are we, really?

[Randy Rathbun]

A few years ago in an issue of PC World, Penn Jillette (of Penn and Teller) brought up the question about just how advanced is AI, and how advanced has it really become?

For his example he came up with a weight guessing program, much like the weight guesser at a carinval sideshow. The program went like this:

Do you weigh 1 pound? (Y/N) N
Do you weigh 2 pounds? (Y/N) N
Do you weigh 3 pounds? (Y/N) N

Well, I think you get the picture. Eventually this program is going to guess your weight. But can it really be called AI?

Looking at most of the examples that are on websites, yours included, the programs are extremely simple in what they do. To be completely honest, they do not even appear to be much more complex in their "thinking" than the Weight Guesser.

Another example that I can think of are Expert Systems. To me this is nothing more than a simple linking of menus that ask questions. For example, "is the switch on?" Some people consider this to be a form of AI, but by that definition anything that has an if statement would be AI.

So, my question - are computer programs actually really honestly beyond this?

Problems with AI

[Dr. Sp0ng]

People have been predicting for a long time that AI is not far off, but it doesn't seem to be getting much closer. What are the biggest obstacles in the mission of creating a convincing AI algorithm (i.e. one that is indistinguishable from a real human being)? Is the barrier language, knowledge base, or something else entirely?

sentience/intelligence

[Gutzalpus]

What would be the determining factors to make you decide that a computer program was intelligent (or possibly self-aware/conscious)? Do you think it would ever be possible to reach this point (maybe not in the near future, but eventually)? If so, at what point might you consider it unethical to turn off/shut down this program?

If you don't think we could ever reach this point with AI, why not?

Replacing Humans?

[Anonymous Coward]

Do you view the study of AI as a the creation of something that can replace humans in doing everyday tasks, or as something meant to augment humans in their decision-making processes? For example: My company is working on some transportation routing software, that can take reservations for taxi cabs and decide the most efficient way of assigning available taxis to these reservations. This is now done by human (by hand) - our system could be set up to do everything automatically, or, to present optimal decisions and let a human ultimately decide what to do. But I am uneasy about completely replacing a human intelligence with a computer program... If we can make them do things better than us, and truly make decisions on their own, then, eventually, what the heck will they need us for?

Comparison to Brooks subsumption architecture

[Timothy J. Wood]

Rodney Brooks, the director of the AI Lab at MIT, seems to have made a great deal of progress with his subsumption architecture. He proposes that the hard parts of AI are not the planning and cognitive exercises, but rather the parts of intelligence that you find in lower life forms like insects, small mammals and such.

How does this compare to your approach?

What do you think of "Creatures"?

[Saige]

As an AI researcher, I wonder what your take would be on the "Creatures" software products. Do you think they have done anything novel in the AI area, such as simulating a body and it's and it's influence on the brain, which is done as a neural net? Has any of the Creature Labs' work even been noticed by AI researchers?


---

Two sides to the problem

[speek]

AI seems to have two aspects that need solving: Hardware power, and understanding of what intelligence is. On a sliding scale of 1-10, 10 being "solved and adequate", where are we with respect to each of these two problems?

How do you do you when you've achieved AI?

[leko]

We've all seen the movies where some AI computer run amok, Tron, Terminator, etc...

My question(clear from the subject) is: How do you know when you've achieved AI? Will it be when a system starts learning just because it _wants_ to? Does it need to be able to communicate with us above some predetermined level? Does it need to go completly crazy trying to fight for its survival? Does it even need to exhibit intelligence as humans know it?

How do you justify your expectations?

[Anonymous Coward]

For the past 40 years, AI has just been 10 years or so away.

It's still just 10 years or so away.

It's not getting any closer.

How do you justify any degree of optimism wrt the future of AI at this point? What makes now fundamentally different from anytime in the past 40 years?

Questions based on your academic path

[Anonymous Coward]

The way to the field of AI isn't always extremely clear. What type of background do they expect? Is it mostly a researching position or is it treated like a normal job with normal goals? Is there any classes or subjects or schools you recommend to make it into the AI field? Also, how exactly did you get into the field? How did AI intrigue you into what you do now, despite all the controvery to create an intelligence that could possibly be considered a "god" compared to the human existance? Very interesting to say the least, and something I'm interested in. AC

most likely path?

[Anonymous Coward]

Dr Jordan:

Do you think that AI is more likely to arise as the result of explicit efforts to create an intelligent system by programmers, or by evolution of artificial life entities? Or on the third hand, do you think efforts like Cog (training the machine like a child, with a long, human aided learning process) will be the first to create a thinking machine?

~Phyruxus

Turing award.

[V.]

Do we win something if we can fool him into answering a computer-generated question? ;)

To which I would add...

[joss]

I also read your IP proposal, and agree with the points mentioned above.

However, I also have a problem with your proposal from an economic perspective:

Property laws developed as a mechanism for optimal utilisation of scarce resources. The laws and ethics for standard property make little sense when the cost of replication is $0. The market is the best mechanism for distributing scarce resources, so you propose we make all IP resources scarce so that IP behaves like other commodities and all the laws of the market apply.

We are rapidly entering a world where most wealth is held as a form of IP. Free replication of IP increases the net wealth of the planet. If everybody on earth had access to all the IP on earth, then everybody would be far richer - it's not a zero sum game . Of course, we're several decades at least from this being a viable option since we've reached a local minima. (Need equivalent to starship replicators first - nanotech...)

Artificially pretending that IP is a scarce resource will keep the lawyers, accountants, politicians in work, and will also allow some money to flow back to the creatives, but at the cost of impoverishing humanity.

I could actually see your proposal being adopted, and I can see how it will maintain capitalism as the dominant model, but I also believe that it is the most damaging economic suggestion in human history

Could you tell me why I'm wrong.

Software Market & Open Source

[Breace]

In your 'hyperbook' about your idea of a software market I noticed that you say that Open Source Evangelists should support your movement because it will be (quote) A way for your next team to be rewarded for their creative work if it turns into sendmail, apache, or linux.

I assume (from reading other parts) that you are talking about a monetary reward. My question is (and this is not meant as a flame by any means), do you really think that that's what the Open Source community is after, after all? Do you think that people like Torvalds or RMS are unhappy for not being rewarded enough?

If the OS community doesn't care about monetary rewards, is there an other benefit in having your proposed Software Market?

Regards, Breace

And what about Freedom?

[Hobbex]

Mr. Pollack,

I read your article about "information property" and was surprised to find you dealt with the matter completely from the point of view of advancing the market. Their are those of us who would argue that the wellbeing of the market is, at most, a second order concern, and that the important issues that Information age gives rise regarding the perceived ownership of information are really about Freedom and integrity.

These issues range from the simple desire to have the right to do whatever one wants with data that one has access to, to the simple futility and danger of trying to limit to paying individuals something that by nature, mathematics, and now technology is Free. They concern the fact that our machines are now so integral in our lives that they have become a part of our indentity, with our computers as the extension of ourselves into "cyberspace", and that any proposal which aims to keep the total right to control over everything in the computer away from the user is thus an invasion into our integrity, personality, and freedom.

Do you consider the economics of the market to be a greater concern then individual freedom?

-
We cannot reason ourselves out of our basic irrationality. All we can do is learn the art of being irrational in a reasonable way.

Frankenstein

[Borealis]

For a long time there has been a fear of a Frankenstein being incarnated with AI. Movies like The Matrix and the recent essay by Bill Joy both express worries that AI (in the form of self replicating robots with some AI agenda) can possibly overcome us if we are not careful. Personally I have always considered the idea rather outlandish, but I'm wondering what an actual expert thinks about the idea.

Do you believe that there is any foundation for worry? If so, what areas should we concentrate on to be sure to avoid any problems? If not, what are the limiting factors that prevent an "evil" AI?

Human brain - AI connection - is there?

[spiralx]

Do you think that a greater understanding of the human brain and how intelligence has emerged in us is crucial to the creation of AI, or do you think that the two are unconnected? Will a greater understanding of memory and thought aid in development, or will AI be different enough so that such knowledge isn't required?

Also, what do you think about the potential of the models used today to attempt to achieve a working AI? Do you think that the models themselves (e.g. the neural net model) are correct and have the potential to produce an AI given enough power and configuration, or do you think that our current models are just a stepping stone along the way to a better model which is required for success?

Why not Write a Screensaver?

[peteshaw]

First of all, it is indeed an honor to pester a bigname scientist with my puny little questions! Hopefully I will not arouse angst with the simplicity of my perceptions. Aha! I toss my wheaties on Mount Olympus and hope to see golden flakes drift down from the sky!

I have always thought that distributed computing naturally lends itself to large scale AI problems, specifically your Neural Networks and Dynamical Systems work. I am thinking specifically of the SETI@home project, and the distributed.net projects. Have you thought about, or to your knowledge has anyone thought about harnessing the power of colective geekdom for sort of a brute force approach to neural networks. I don't know how NN normally work, but it seems that you could write a
very small, lightweight client, and embed it into a screen saver a'la SETI@home. This SS would really be really a simple client 'node'. You could then add some cute graphics like a picture of a human brain and some brightly colored synapseds or what have you.

Once the /.ers got their hands on such a geek toy I have no doubt you'd have the equivalent of several hundred thousand hours or more of free computer time, and who knows, maybe we could all make a brain together! I would love to think of my computer as a small cog in some vast nueral network, or at least I would until Arnold Scwartzenegger got sent back in time to kill my mom. Whaddayathink, Jordan? Is this a good idea, or am I an idiot?

Pete Shaw

How should an amateur get started working on AI?

[Henry House]

It seems to me that a significant problem holding back the development of AI is that few non-professionals grok AI well enough to offer any contribution to the AI and open-source communities. What do you suggest that I, as a person interested in both AI and open source, do about this? What are the professionals in the AI field doing about this?

AI and ethics.

[kwsNI]

What do you say to the people that feel it is unethical to try to create "intelligence"?

kwsNI

AI Metrics

[john_many_jars]

I have read several coffee table science books on the subject and often find myself asking for a way to measure AI. As has been noted, AI is always elusive and is just around the corner. My question is how do you gauge how far AI has come and what is AI?

For instance, what's the difference between your TRON demonstration and a highly advanced system of solving a (very specific) non-linear differential equation to find relative and (hopefully absolute) extrema in the wildly complicated space of TRON strategies? Or, is that the definition of intelligence?