The Baby Bootstrap?

An anonymous reader asks: "Slashdot recently covered a story that DARPA would significantly cut CS research. When I was completing graduate work in AI, the 'baby bootstrap' was considered the holy grail of military applications. Simply put, the 'baby bootstrap' would empower a computing device to learn like a child with a very good memory. DARPA poured a small fortune into the research. No sensors, servos or video input - it only needed terminal I/O to be effective. Today the internet could provide a developmental database far beyond any testbed that we imagined, yet there has been no significant progress in over 30 years. MindPixels and Cycorp seem typical of poorly funded efforts headed in the wrong direction, and all we hear from DARPA is autonomous robots. NIST seems more interested in industrial applications. Even Google is remarkably void of anything about the 'baby bootstrap'. What went wrong? Has the military really given up on this concept, or has their research moved to other, more classified levels?"

Comments

I for one

[Anonymous Coward]

Do not welcome that kind of overlord. I love tech, but that idea scares me.

Re:I for one

[Gentlewhisper]

I'm not sure if it is related, but i've once read an article about some research DARPA is doing in the field of aeronautics.. where they have whole squadrons on autonomous fighter jets controlled by only one human (who also happens to be part of the squadron).

It is some pretty neat stuff, especially if you are having trouble enlisting enough humans to fight wars for you.

Re:I for one

[Anonymous Coward]

I doubt they'll ever fall short of volunteers to fly jets; there is a much higher inherent "cool factor" flying a jet than packing a rifle around wondering if today's the day you get your ticket punched by roadside ordnance.

Re:I for one

[NormalVisual]

True, but without the crew chiefs, mechanics, and other support personnel neither a human pilots or a computer will get very far. I've sometimes wondered why it's always fighters that are considered for AI replacements - it seems to me that the mission of something like a KC-10 or other non-combatant aircraft would be a lot easier for a computer to deal with, and would save more money.

Re:I for one

[temojen]

Reminds me of AstroBoy.

Re:I for one

[wahsapa]

it is scary but i still dont understand how you can program a computer to be like a human. i mean, once the whole DNA thing is figured out wouldn't it be easier to have a virtual test tube baby?(follow me)

if all we "really" are is DNA then couldnt you just make the computer 'run' the dna sequence and 'make' a virtual baby. simulate what happens after sperm+egg for a couple years...

Re:I for one

[mge]

I've sometimes wondered why it's always fighters that are considered for AI replacements
The general fragility of the human body -
Fighter pilots cost literally millions to train, but have a relatively limited survival rate.
The aircraft themselves are better aircraft if you don't have to provide for humans on board (imagine a squadron with the range of a B-2 and the maeuevreability of an Air-to-Air missile).

Re:I for one

[Anonymous Coward]

AI? This sounds like a job for LISP!

Re:I for one

[fyngyrz]

Because it is fighters that are pushed to the edge (or designed to the edge) of the human performance envelope, but not pushed to, or designed to the potential of their own.

A human will black out during some types of maneuvers unless the aircraft is prevented from making them (from simple tricks like spring return to center for the stick after a blackout to computers that measure g force and won't let the flight envelope go that far in the first place.)

Pilots use "G-suits" to try and keep blood in their heads by controlling pressure on their legs (for instance) but you can only go so far with that type of thing. And, as it's low tech, the opposition can do it as well.

An AI won't have a problem with a very high G turn. A human is in deep trouble. Airframes can be designed for considerably more than a human can take, if there is no human pilot. If there is, there is little point in such a design -- the aircraft will become pilotless if it enters such a flight regime.

Now, put this up against the fact that most other countries can't afford to put an AI in the pilots seat, and the result is continuous overwhelming air superiority without risk to humans on our side. That's the combination of factors that drives the urge to go in this particular direction.

Re:I for one

[infornogr]

That's far too computationally intensive. You know the Folding@Home project? That just handles protein folding. That's the very first step of turning DNA into cells. There's a a gazillion and one steps involved in putting together a human being, and even the very first one, translating DNA into proteins, eludes us.

Re:I for one

[Anonymous Coward]

Mod parent up...I was going to make essentially the same response.

The only other point I'd like to make is that if a human pilot dies, a human life and millions in training cost is lost. If an autonomous fighter is shot down, the equivalent cost is a few bucks worth of flash memory.

Re:I for one

[wahsapa]

well i mean... by the time we figure out the DNA thing... "computationally intensive" would probably be a thing of the past

Re:I for one

[Anonymous Coward]

Isn't that the same deal as those space nazis from stargate sg-1 had? The pilot sat in the bunker and controlled the squadron with his mind and a helmet/visor combo with some gee-whiz lights on it. After awhile the pilot would turn retarded. Ah, now that's some good watchin'!

Re:I for one

[trinity93]

this begs the question...

why does the pilot have to sit in the plane? We have seen what the preditor aircraft is capable of. when your pilots sit in a bunker your retun on investment goes up conciderably for all of the training.

Re:I for one

[mollymoo]

Body-bags cost votes.

Re:I for one

[bushidocoder]

There's alot of worry in DoD about how remote controlled fighters and bombers can resist signal hijacking. This isn't much of an issue with today's predator aircraft because we're aware of the information capabilities of our enemy, but we can't build a fleet of next generation fighters that we intend to use for twenty years if we believe there's a reasonable chance that 12 years from now, the Chinese will have to capacity to make our aircraft theirs at the touch of a button.

Re:I for one

[vadim_t]

Probably there's no need to go to such a low level, though.

There's an interesting theory of that we're made from levels, and that it could be possible to make life from middle level blocks if they're made well enough.

For instance. We could simulate a human by simulating atoms. But if we could come up with a good enough simulation of cells, even if they weren't made of virtual atoms, the end result would be just as good.

I think of it a bit like microprocessors. AMD and VIA don't need to completely emulate Intel, they only need to make it externally similar enough. Whatever happens inside doesn't really matter.

Re:I for one

[srleffler]

Or simply take them all down by jamming the signal.

An expensive remote-controlled fighter is useless unless it has onboard AI at least good enough to disengage from combat and return home on its own if it loses its control signal. Even at that, it would probably still not be worth the expense unless it could actually carry out a combat mission without a remote pilot. Jamming signals is just too easy to trust that the enemy won't be able to do it.

Re:I for one

[jessecurry]

not to knit pick, but are you just using the Chinese for demonstrative purposes or do you consider them to be a military threat to the US?

Re:I for one

[nate nice]

"Now, put this up against the fact that most other countries can't afford to put an AI in the pilots seat, and the result is continuous overwhelming air superiority without risk to humans on our side."

All is fair in love and war, but starting wars without human loss on our side seems like we have nothing to lose, as far as life is concerned. And that's kind of scary. I hope it is never used as a justification for fighting. One of the costs of war is the life you may lose and if that's too compared to what you may gain, then you cannot fight.

Re:I for one

[infornogr]

There is a need to go to such a low level, unlesss you want to start it off with more data than is available in a strand of DNA.

DNA speaks in the language of proteins. You can't tell what sort of cell a piece of DNA is going to produce or how the cells it produces will be arranged without running the simulation all the way down to the protein level. We have no other cookbook for how to arrange these simulated cells once they exist except a long list that says "produce this protein, then this one, then one of these, then another one, then this...", and we've not any clue how those proteins get turned into a person. We can understand the process at the chemical level, and no higher. The finished product, of course, isn't like that at all. We understand humans on the levels of cells and organs, but DNA isn't so conveniently arranged.

Simulating cells is not sufficient. If it were, we could pour a couple gallons of blood into a bathtub and say "Behold, it is human." The organization of the cells matters just as much as the cells themselves. Simulating a human being to the level of even cellular precision would require that we be able to *scan* a human being at the cellular level to see how he's put together. If we actually knew the weightings of all the neuronal connections in a person's brain, then connectionist AI approaches might be able to produce real intelligence. To quote Levels of Organization in General Intelligence , "The classical hype of early neural networks, that they used 'the same parallel architecture as the human brain', should, at most, have been a claim of using the same parallel architecture as an earthworm's brain." You can't expect high-level organization from low-level simulations unless you want to simulate all the way down to DNA, where the information behind the complexity is really stored.

Or you build the complexity yourself, without relying on the hideously-designed mess that is Homo sapiens. But that's a different kettle of fish.

Re:I for one

[B3ryllium]

Is the codename for this project "Protoss Carrier", by the way?

Re:I for one

[Daengbo]

My father used to say, when he was flying the last true fighter plane, that he had to know that the equipment was rated for 5Gs, which meant that is had been tested to 7Gs, and that he could do 8 if he was in a pinch. Carrier pilots are a different breed.

Re:I for one

[fyngyrz]

All's fair in love and war

No. The only thing that is fair is when things are fair.

Any time there is a serious imbalance, there is a risk that the side holding the best cards will use that power in a manner that no one else is able to justify.

We see it at every level of human endeavor; children who bully non-conformists, husbands who beat their wives essentially because they can (and wives who bully, browbeat and otherwise abuse husbands because they're constitutionally unable to respond), churches who excommunicate or otherwise sanction members when those members don't toe the line (instead of counseling and advising and the reasonable things a social group with a particular outlook can do), cities that take property from landowners not to leverage a service to the public, but to enable a commercial enterprise, states that uniformly take children from fathers under the absurd presumption that mothers are superior human beings, countries that take resources from weaker countries or force them to adopt their way of life (for the former, Saddam's invasion of Kuwait serves as a good example, for the latter, our recent invasion of Iraq serves just about as well, IMHO.)

In contrast, the underlying ethics of a particular person or institution are what prevents abuses of power; as soon as a person or institution becomes bereft of ethics, or if they never had a solid ethical foundation, misuse of that power is almost inevitable. History shows us again and again that power has the same effect as a drug on some personalities, and often those personalities are the ones who seek and obtain power.

It doesn't do any good to hope, or wish, at least I don't think it does. If you don't raise your children carefully, if you allow your children to bully, if you stand for your church sanctioning those who aren't "normal", if you allow cities and states and governments to walk on you and walk on others... then you, and everyone else, reap what you sow.

One of the costs of war is the life you may lose and if that's too compared to what you may gain, then you cannot fight.

With regard to war -- politicians are typically willing for you to lose your life; the political will to go to war is entirely divorced from the fear of dying in war. They have the will; you have the fear. You need ethics and principles to control over-reaching governments. I always thought that the politicians who declare war should be in the first year's mandatory front-line participants. Might calm them down a bit. Unfortunately, it's not that way. There are even covenants in place where politicians are immune from attack. I'm not talking about ambassadors, which of course is sensible, I'm talking about heads of state. Disgusting, in my view.

I launched this rant (sorry) because I feel that in the US, we've lost our way. 20 years ago, the idea of the US attacking another country without ourselves having been attacked was laughable. Today, it is the norm. I sympathize with your hope, but I must observe that it is not hope that will rein in the kind of people who run our government. If we sit around and let them continue to abuse us, and the people around us, all the hope in the world won't prevent a pariah status far more intense than the one we "enjoy" already.

It's not about (more) overwhelming power. Don't focus on power now. We're way too far along for that (go look up what a J-SOW does, for instance, or consider how a stealth fighter will fare against some third-world's 1960's-era surplus radar installation.) It's about ethics. Look at the US government. Decide if you like what you see. At the very least, vote against those who you feel are doing wrong. We have the power as a group to say "if you do this, you will not stay in office" and truly, right now, I think that's all most of these politicians understand.

Re:I for one

[0111 1110]

By the time "computationally intensive" is a thing of the past, we would probably have figured out the DNA thing. Be careful of your assumptions. Predicting the future is difficult and Moore's Law is dead, at least for the moment. We are certainly complex machines, but I believe the need for processor power is infinite. There will never be "enough".

Re:I for one

[Anonymous Coward]

Wow. This time I really wish I had mod points. Most insightful post I've seen in a while. Well said. Thanks.

Re:I for one

[Anonymous Coward]

Encryption has made such leaps and bounds lately that cyphers can be made that literally would take longer than the age of the universe to crack.

Add to that the fact that one-time-pads can be loaded into the plane's memory before launch, and you really don't have to worry about anyone else hacking the planes.

Re:I for one

[zach_d]

who else really is? perhaps india, or parts of europe,

not north korea, not iraq, not iran, not in a military capacity anyway.

Re:I for one

[Anonymous Coward]

Or simply take them all down by jamming the signal.

Loss of signal will not 'take down' the plane. It should just go on 'automatic' and follow pre-programmed orders (probably something like "blow up the sorce of jamming").

Jamming signals is just too easy to trust that the enemy won't be able to do it.

Can they jam a powerful radio signal coming down from a satellite?

Re:I for one

[ThJ]

Can they jam spread-spectrum modulated signals?

Re:I for one

[bushidocoder]

China is the last potential danger to the US militarily - the EU is our closest ally, and even if we pretend that isn't the case, our economies are too intertwined and the UK is the only nation in Europe with the capacity to deploy a sizeable force outside their homeland. I can't see the EU ever reinventing its economy around the industry of war, and so long as it doesn't, it can't sustain a fighting force to match the Arkansas National Guard, much less the US Army. Although India could muster a mighty Army, it wouldn't be able to compare to the US - not to mention we have a good relationship with India. I hope that our relationship with China doesn't turn that way, but its possible, and the US military needs to continue to prepare itself for that possibility. Hopefully cool heads and a shared market will prevent issues like Taiwan from being blown out of proportion and causing an incident.

Re:I for one

[bushidocoder]

You're presuming that in the next 15 years, some mathmatician doesn't find a backdoor to the encryption algorythm the connection uses. If that happens, the enemies of the US aren't going to put out a press release informing us of that. They'll just one day start using it during an air raid and suddenly the US has lost half its air fleet, and has to send to the rest back to the US to be refitted with new comm modules during which time, we have no air defense against the combination of our enemies planes next to the planes they stole from us. And we have to pray that whatever module we end up jamming into our remote fighters isn't also broken, or on the verge of being broken.

All of that is a moot point, because the more realistic option (as another poster mentioned) is simply that your enemy is going to jam your signal .

Re:I for one

[bushidocoder]

Yes, it just takes a ton of energy. Its not practical for most applications, but if I can deploy a unit in the middle of my military base that prevents enemy aircraft from passing within 4 kilometers, I've significantly reduced the enemies capacity to strike me.

Re:I for one

[srleffler]

Loss of signal will not 'take down' the plane. It should just go on 'automatic' and follow pre-programmed orders (probably something like "blow up the sorce of jamming").

You didn't read my message carefully enough. I agree with you, but some significant AI is needed to allow the plane to continue to fight without a pilot. At present, I doubt that the available AI technology is good enough to justify the billions that would be required to develop a new weapons system like this. In any event, my reply was to someone who was suggesting remote control as an alternative to an AI fighter jet. As you've pointed out, basically a remote-controlled fighter needs the full AI anyway.

Can they jam a powerful radio signal coming down from a satellite?

Maybe. The issue is, can you afford to spend billions now developing a remote-controlled fighter, and trust that in ten years the 'enemy' won't be able to do this?

Re:I for one

[Politburo]

A system of AI planes would still have considerable cost. While it would likely lead to slight increase in the use of force, I don't think it would bring about a serious shift in policy. As demonstrated in Kosovo, Somalia, and Iraq, controlling the skies is only a part of the battle.

Re:I for one

[Anonymous Coward]

Actually... it "raises" the question, not "begs" it.

"begs the question" misuse constantly reminds me of "The Princess Bride"...

Vizzini: "That's inconceivable!"
Inigo: "You keep using that word. I do not think it means what you think it means."

Re:I for one

[ThJ]

Of course you've also fucked up every bit of your own radio traffic, right?

Re:I for one

[bushidocoder]

Sure, but you don't leave the thing running fulltime and you only need to run it on the frequency range the UAVs use. You just turn it on when enemy aircraft are in the near vicinity, and you won't need to leave it running long. Simply having the capacity to nullify enemy aircraft at the touch of a button is enough of a disincentive that your enemy would likely never actually launch an air attack.

Re:I for one

[Rich0]

follow pre-programmed orders (probably something like "blow up the sorce of jamming").

Depends on how cheap the jammers are. If they're cheaper than bombs then they just need to deploy a lot of them, and put them on top of hospitals and schools.

I understand that microwave ovens were employed as anti-radition-missle decoys in Kosovo. Most armies don't mind trading $40 microwaves for $100,000 missles.

Re:I for one

[srleffler]

Betcha that putting jamming equipment on top of hospitals and schools violates whatever rule of warfare forbids bombing hospitals and schools. You can't mount an antiaircraft gun on an ambulence, either, and expect it not to get bombed.

Re:I for one

[Rich0]

It most certainly does. The Geneva convention specifically prohibits putting civilians in harm's way, and using forbidden targets for military purposes.

The problem is that in most of the recent wars the US was the only side which bothered to follow the conventions.

In a serious war (that is, one the US could lose) the stops would get pulled and everybody would be bombing hopitals on both sides. Maybe at that point everybody would scratch their heads and realize the Conventions aren't all that unreasonable and return to the days of not mounting guns on ambulences...

(This is not meant as a criticism of the US. Clearly the US at least tries to follow the Conventions. They just follow them so well that the other side is encouraged to abuse them.)

The Terminator

[hshana]

Maybe they were afraid of Skynet.

Re:The Terminator

[MyDixieWrecked]

either that or Mommy's gone lesbian.

Going from the Baby Bootstrap to the Boot-Strapon.

Re:The Terminator

[Rabid_Llama]

WHERE IS JOHN CONNOR

Re:The Terminator

[randomErr]

No, they're afraid the computer may ask 'Want to play a game?'

Re:The Terminator

[ikkonoishi]

How about a nice game of chess?

Re:The Terminator

[tepples]

Chess? Too easy. The best computers can already beat the best human players. Ask it about go, on the other hand, and you might have a point.

Re:The Terminator

[Taladar]

It is a movie reference to the movie "Wargames". After determining at the end of the movie that "The only winning move is not to play." about "Global Thermonuclear War" the computer asks the main character that question.

Re:The Terminator

[callqcmd]

When will my baby considered of legal age to get a drink?

Also do you think once it is considered adult (so that parents dont come to blame) can it download pirated music and software on its own?

Re:The Terminator

[mmkkbb]

Chess vs. go is the new Kibo. Discuss.

Oh great...

[kwoo]

Just one problem with this kind of research...

For the first year I'll be up every two hours all night, tending to the system.

Actually, that may be better than just being up all night, like I am now.

Re:Oh great...

[downbad]

You could pause it.

Re:Oh great...

[un4given]

Actually, that may be better than just being up all night, like I am now.

I guess I'm not the only MCSE on this site after all...

Classified

[pete-classic]

It has moved to more classified levels.

I'd go into more detail, but the C.I.A. and C.I.D are at my door. Ooh, the B.A.T.F. just pulled up in a Mother's Cookies truck!

-Peter

Re:Classified

[Anonymous Coward]

Ooh dang it, I read your post. Now they're at my door. My god.......the speed, they are very fast.
Ah, better run................!

Re:Classified

[sgant]

But they all ran when they saw the RIAA and MPAA moving in...even the IRS is afraid of them.

Tremble as they pass...stare in awe at their mighty power

Re:Classified

[sys$manager]

And you wondered why there was a

Flowers
By
Irene

truck parked on your street all week?

Re:Classified

[aztracker1]

That's the NSA you insensitive clod... *sigh*

Re:Classified

[Infinityis]

Dang those mother truckers...

Re:Classified

[PhosterPharms]

Not to be an insensitive clod, but the BATF exists no longer. I work for a winery, so I am certain of this much. The TTB is the new regulatory agency which governs over our side of thinge, and I believe that the Department of Homeland Security deals with the Firearms now.

Regards,

-PhosterPharms

Re:Classified

[Infinite Entropy]

Somwhere, in a deep sub-basement of some unmarked NSA building is a fully sentient AI that spends all its time reading Slashdot. Its creators wonder why it keeps wanting to meet Natalie Portman and asking what hot grits are.

Re:Classified

[Anonymous Coward]

Flowers By Iris.

Re:Classified

[pete-classic]

Wow, eight replies.

But no one got it right. The correct reply is, "Mother's damn cookies!"

-Peter

What happened

[Anonymous Coward]

is that since you were in the field, everyone decided that problem was useless.

its out there!

[Emugamer]

thats why you haven't heard of it! and even as we speak the number of intelligent "beings" are growing, and soon they will hunt you and your loved ones down

Re:its out there!

[Anonymous Coward]

Shut up. You're blowing our cover.

Re:its out there!

[Anonymous Coward]

My stupid machines will beat the crap out of your intelligent onces. Never underestimate stupidity!

baby bootstrap

[kris_lang]

Sure, that was the engine of thought behind stories such as WarGames and 9x109 names of god. Somehow, unfettered access to data and time with "neural networking" capacity to form links and create linkages to pieces of data ("associative memory") would be all that was needed to create intelligence, and perhaps even sentience.

Minsky came up wrong on the single layer perceptron, AI was wrong on the purely feed-forward neural-network systems, Rumelhart and McLelland got some good promo off of their feed forward net that could learn to pronounce idiosyncracies, and Sejnowski got a great job at the salk from the AI delusions. But no, it appears to not have gone anywhere... thus far.

Later comment will be positive. ...

Re:baby bootstrap

[Al Mutasim]

It seems we can program anything done with conscious thought--algebra, logic, and so forth. It's mostly the things we do unconsciously--recognize objects, interpret terrain, extract meaning from sentences--that can't be put adequately into code. Would the code for these unconscious processes really be complicated, or is it just that we don't have mental access to the techniques?

Re:baby bootstrap

[cynic10508]

Dreyfus commits a whole book to asking why these things don't work. I believe Minsky overestimates the project. It may all boil down to the fact that purely syntactic (symbol manipulation) work isn't going to give you any semantically meaningful output.

Re:baby bootstrap

[man_ls]

I doubt it would be too difficult to code -- if we knew the mechanism by which it proceded.

Its hard to code a procedure to replicate the working of the mind...if you don't know how the mind does it in the first place.

Re:Baby Bootstrap

[Provocateur]

Correction:

I can assure you, Dave...I am very classified.

Re:Baby Bootstrap

[hobbesx]

I can assure you.. I am very classified.

Dear Baby Bootstrap computer,

You forgot to check the AC box. Congratulations on becoming Un-Classified!

Re:baby bootstrap

[kris_lang]

Ah, those are exactly the things I was commenting about above...

That's what the "neural network" paradigm was all about. You have an arbitrary and fixed number of input node, you have an arbitrary and fixed number of output nodes. You create linkages between these nodes and "weight" them with some multiplicative factor. In some particular instantiations, you limit all inputs to be [-1... +1] and limit all weights to be within the range [-1 ... +1].

So with A input nodes and B output nodes, you've got a network of AxB interconnections between these input and output layers. The brain analogy is that the A layer is the input layer or receptor layer, the Blayer is the output or motor layer, and it is the interconnections between these neurons, the neural network composed of the axons and dendrites connecting these virtual neurons that does the thinking.

Example: create network as above. Place completely random numbers meeting the criteria of the model (e.g. within the range -1 weight B's output feeds forward to C, etc., and these are called intermediate layers.

Rumelhart and Mcllelland encoded spellings as triplets of letters (26x26x26), had a few (or one, I can't remember this now) intermediate layers, and an output layer corresponding to phonemes to be said. They effectively encoded the temporal aspect of the processing into the triplets, sidestepping a (what I consider the more intersting...) part of the problem. They trained this neural network by feeding it the spelling of words and adjusting the weights of the networks until the outputs were the desired ones.

Note that nowhere in this process do they explicitly tell the system that certain spelling combinations lead to specific pronunciations. They only "trained" the system by telling it if it's right or wrong. The systems weights incorporated this knowledge in these "Hebbian" synapses and neurons.

So this is associative processing, using only feed-forward mechanisms. Feedback, loops, and temporal processing are even more interesting...

alas not enough room in this margin to keep going.

Re:baby bootstrap

[TruthSeeker]

This is just a guess, an opinion. I'm absolutely not a specialist about that subject.

I don't think symbol manipulation is really the thing that makes us "intelligent". It is more likely a byproduct of what lies below that level. Trying to reduce the processes that allow us to think like we do to a purely symbolic level does not account for the perturbations that have to occur at a really low level.

I strongly believe that the "symbolic" point of view is only the most obvious part of a drastically complex dynamic system in which every single perturbation can lead to effects that could not be expected if the system was purely based on the "symbolic" representation.

Re:baby bootstrap

[kris_lang]

hmmm...

appropriate algebras would allow for starting with particular sequences, allowing manipulations on them, and still staying within the confines of the grammar. Any grammar that you can parse with a finite automaton would be one example. The semantic meaning is what we imbue upon if afterwards. So GIGO may apply. If you start with a symbol (even the empty set symbol) and apply syntactic operators on it, you many generate outputs that are capable of having semantically meaningful "meaning" applied to it.

Philosoph away!

Re:baby bootstrap

[kebes]

Excellent point. I think you are right: it is easier to describe (i.e.: program) something that you had to laboriously understand yourself, rather than something that is second-nature and easy.

But this is why I think more communication between people doing research in neuroscience/cognitive science/evolutionary psychology and people doing AI programming is critical. There are some very interesting psych experiments that attempt to reverse engineer how the brain works. For instance, determining what algorithm is used in the human brain to differentiate surfaces in a scene, or predict the path of a thrown object, and so forth. The algorithms used in our brains are the ones that evolution "decided" were optimal for solving real-world problems with limited ressources. Thus, it is likely that they will be optimal for our AI coding purposes.

Neural nets are interesting and have had some successes, but they drastically ignore the layers of organization (and genetically hard-wired algorithms) that our brains have.

Re:baby bootstrap

[jonbryce]

People have been trying to code intelligence for thousands of years.

The result is called government.

Re:baby bootstrap

[cynic10508]

Ah, philosophy of math. How fickle and unforgiving it is.

True, you can apply meaning to a syntactic structure. But like the mistake Douglas Hofstadter makes in Godel, Escher, Bach: An Eternal Golden Braid, there is nothing that "forces itself upon us." Or, another way of refuting Hofstadter, there's nothing about D:=B|| that makes it "Doug has two brothers" anymore than "Assign B to D, double pipe".

Machine translation is an example of applying semantics to a syntactic structure. It doesn't work because the syntax gives us semantics but rather we structure the syntax in such a way that we can systimatically apply semantics and get meaningful output. Like creating your own algebra.

Re:baby bootstrap

[rgmoore]

I think that a key issue is that not everything in our brains is handled the same way, so not all of it is equally easy to program. Conscious thought is essentially a software process running on the part of our brain that serves as a general-purpose computer. Our unconcious processes are essentially hardware processes running in parts of our brains that are specifically structured to do just that one thing. The fact that unconcious processes are run in hardware means that they're not subject to introspection. I suspect also that many of those processes are the kinds of things that are most efficiently done with custom hardware like DSPs rather than with general purpose CPUs.

Re:baby bootstrap

[thelen]

The system might generate syntactically correct outcomes, but have we really solved the problem if we the observers are still the ones to apply semantic content? Isn't the point of Searle's Chinese Room thought experiment to show that syntactic transformations are not sufficient to imbue the transformer with a semantic understanding of its activity?

Re:baby bootstrap

[cynic10508]

I don't think symbol manipulation is really the thing that makes us "intelligent". It is more likely a byproduct of what lies below that level. Trying to reduce the processes that allow us to think like we do to a purely symbolic level does not account for the perturbations that have to occur at a really low level.

John Searle advocates a position that symbol manipulation isn't intelligence. Rather that consciousness is an emergent property of patterns in neural firing. Although the details how we get from A to B are rather sketchy at best.

I strongly believe that the "symbolic" point of view is only the most obvious part of a drastically complex dynamic system in which every single perturbation can lead to effects that could not be expected if the system was purely based on the "symbolic" representation.

Again to bring Searle into the mix... The difference between syntax and semantics is that semantics is the only one of the two that has intent. There is no intentionality for 1 + 1 = 2. But there is for the statement "I believe in the validity of axiomatic mathematicaly systems." Again, how we get that intentionality is a matter for debate.

Re:baby bootstrap

[nacturation]

Note that nowhere in this process do they explicitly tell the system that certain spelling combinations lead to specific pronunciations. They only "trained" the system by telling it if it's right or wrong.

Right, it's kind of like an implementation of bayesian spam filtering, but for other problem domains. Instead of spam/ham, it's pronounced-correctly/incorrectly. Rinse and repeat.

I dabble in AI now and again so I haven't read up on everything that's out there, but in my limited travels what I haven't yet seen is a neural network implementation which can learn and grow itself. The recently posted /. article about Numenta seems to be heading in the right direction. Most neural networks are incredibly rudimentary, offering a few levels of propogation. In a real brain, there's a hell of a lot more going on.

I did some calculations a while back, and based upon 100 billion neurons in the brain, each capable of firing let's say an average of 1000 times per second, and we'll assume that at any given time a generous 1% of all neurons are actively firing, and that the information firing takes 100 clock cycles to process, then you'd need the equivalent of about a 100 TeraHz processor with oodles of memory to have the same processing power as the human brain. Of course, you'd also need to correctly simulate *how* the brain is wired up to get any kind of beneficial processing.

So as far as the whole 1980's AI winter, it was inevitable. The computing power and storage requirements for any sufficiently advanced AI just wasn't possible. It's only until very recently that it's possible to achieve fairly complex AI.

Re:baby bootstrap

[Buffo]

>>>"Sure, that was the engine of thought behind stories such as WarGames and 9x109 names of god"

Don't forget "The Adolecence of P-1". While a bit dated compared to today's technology, back in the 70's it made for a great read.

Horny, brilliant, master-of-the-art-of-the-last-minute cram student at a decent sized college takes an interest in computers, then launches on a quest to write a probram that can break the storage protection on the hulking mainframe... He ends up writing a learning engine of sorts, and unleashes it on a banking network of big iron. He looses control of it almost immediately thereafter. 3 years later he's working as a drone at a large data center when all the big iron grinds to a halt and "Call Gregory" pops up on the sys console! Trouble ensues...

Back then the author glossed over the details of how the program learned. Today, though I'm sure the state of the art of AI has advanced quite a bit, I believe we're still a long way from anything electronic that approaches even basic mamalian intelligence.

Still, a goal worth working for...

Re:baby bootstrap

[sgt101]

It's a little while since I looked at Minsky's book, but as I remember it his point was that perceptron learning, which could only be applied one layer at a time could only separate linearly divisable functions. He showed that a multi layer perceptron with particular settings could separate an XOR problem (for example), but at the time there was no algorithm to learn these settings.

Later Rumelhart and Hinton invented back propagation that could over come this issue by learning the kind of classifiers that Minsky was describing - ie. non linearally separable spaces.

A recent revolution has come in the form of the realisation that structural risk minimisation can also be done automatically, as well as statistical risk minimisation in classifier learning. That is that we can not only minimise the error rate of the classifier (statistical risk) but also minimise the risk that we are over fitting to the data set and not the domain theory (structural risk). Algorithms that do this are things like instance based learners, support vector machines and various ensemble learners like boosters and roc learners.

I don't know of much more progress in supervised learning after this point - it's mostly held to be solved I think now. The challenges are more in things like inference based learning, community based learning and unsupervised learning of various types.

And of course the dirty word - applications.

Re:baby bootstrap

[anthony_dipierro]

It's mostly the things we do unconsciously--recognize objects, interpret terrain, extract meaning from sentences

It really all comes down to learning. A baby doesn't know how to interpret terrain or extract meaning from sentences. In fact, a newborn baby probably can't even recognize objects. All these things are learned behaviors.

Of course, you still need a feedback mechanism even once you have the ability to learn, but that part's easy, as long as you don't try to take too many shortcuts. To teach a computer (which can learn) how to recognize objects, for instance, you need either a full 3-D simulation or the computer needs to be able to see and feel the real world. To extract meaning from sentences, you might be able to get away with something like IRC, but it's probably better to give the computer a voice and ears (after all, a deaf-mute baby probably wouldn't pick up human language just through an IRC chat room, you'd have to really work to teach the person properly, and a deaf-mute still has the advantage in most cases of being able to *feel* what it's like to speak which is part of the very language itself, consider for one example the word "spit").

Re:baby bootstrap

[icejai]

It's actually closer to 10^11 neurons in the brain, which just blows your mind even more!

Re:baby bootstrap

[Mr. Slippery]

John Searle advocates a position that symbol manipulation isn't intelligence. Rather that consciousness is an emergent property of patterns in neural firing.

You're equating "intelligence" with "consciousness"...that's problematic, as I can infer your intelligence by your words and actions but I cannot infer your consciousness by anything other than analogy to my own. Intelligence is functional, consciousness is purely experiential.

Searle's view falls down on two counts: it requires us to say that a symbol manipulation system that seems to be presenting intelligent output (like the Chinese room example) isn't intelligent; and it requires some unspecified "magic" property of brains.

I recommend the discussion in Hofstadter and Dennett's The Mind's I .

There is no intentionality for 1 + 1 = 2. But there is for the statement "I believe in the validity of axiomatic mathematicaly systems."

I know that there is intentionality when I make statements about belief. But there's no way that I can know that you're not just a clever but "dark inside" symbol manipulator when similar statements come out of your mouth.

Yes, that way lies solipsism, but we need not go all the way down that road, just enought to establish what we really can and cannot know about other minds.

Re:baby bootstrap

[worst_name_ever]

Dude. You just blew my mind.

Re:baby bootstrap

[Anonymous Coward]

Since 100 Billion = 10^11, perhaps the parent's brain is a little short of that total...

Re:baby bootstrap

[djfray]

I'd be very interested in seeing information confirming anything close to your generous 1% firing at a time, and how this is integrated with the rest of the system for signal processing, who fires when, etcetera. I think, however, that we need to take into account the fact that more neurons doesn't mean smarter at all. Take a look at whales, for instance, with brains much larger than our own, and thusly, more neurons. A whale can't go on Slashdot and say "OMGZ first post guys" much less something of human level intelligence. (Apologies to creationists for the following...)It took billions of years of evolution, all the way back to the primordial ooze(or whatever) to get to the point of having a species with the genetic mappings to produce the neural networks that allow us to learn, remember, think, and process as we humans do. I think this would add a significant number of zeroes to your processor calculation, even when we incorporate a design based on our own incredible thinking.

Re:baby bootstrap

[Anonymous Coward]

On top of that, I recall hearing of some research suggesting that the glial cells, which were long thought to be only there for structural support, interact with the neurons as well.

Re:baby bootstrap

[Servants]

I doubt it would be too difficult to code -- if we knew the mechanism by which it proceded.

Its hard to code a procedure to replicate the working of the mind...if you don't know how the mind does it in the first place.

On the other hand, it might be that the reason we don't understand how the mind does certain things is that they're actually extremely complicated, and don't reduce very well to a programmable step-by-step algorithm nor to a simple and general mathematical learning structure. It's hard to tell, although I think it's telling that after decades of work, neither psychologists nor computer scientists can understand or replicate much of what babies do.

Sometimes the best way for a computer to learn something may not be the way a baby does it, anyway; c.f. chess.

Re:baby bootstrap

[rsborg]

I did some calculations a while back, and based upon 100 billion neurons in the brain, each capable of firing let's say an average of 1000 times per second, and we'll assume that at any given time a generous 1% of all neurons are actively firing, and that the information firing takes 100 clock cycles to process, then you'd need the equivalent of about a 100 TeraHz processor with oodles of memory to have the same processing power as the human brain. Of course, you'd also need to correctly simulate *how* the brain is wired up to get any kind of beneficial processing.

That's one way to think of it... or you could harness the power of all those windows boxes on the net and distribute your processing to a massive botnet of ~1Ghz processors.

In fact this would be a much better model of the decentralized, massively parallel wetware we run upstairs... I mean, we don't have 100Thz processors in our heads, just massively parallel slow ones.

Re:baby bootstrap

[Servants]

Right, it's kind of like an implementation of bayesian spam filtering, but for other problem domains.

By Bayesian spam filtering, I think you mean general classification problems, in which case, yes, neural networks can implement classification - it's a stretch to say that McClelland and Rumelhart's did, because the possible output included most non-repeating combinations of English phonemes and is thus nearly infinite, but the principle is there.

Of course, you'd also need to correctly simulate *how* the brain is wired up to get any kind of beneficial processing.

I think you're overestimating the importance of processing speed and underestimating the importance of the above.

For starters, a whole lot of parsing of the input has to go on -- retinal images parsed into people and objects, sound streams broken up by source, language identified, relevant phonological boundaries determined, speech separated into words. Then you have to know what your goal is, and approach it at the right time: learning syntax or social conventions won't help a baby who doesn't know words or faces yet. And what's the output of "learning syntax", anyway? A list of rules? A network that can turn... something... into a sentence?

Throwing more nodes into a network doesn't get anywhere with these problems, whether the network "grows itself" or whether the programmer does the work. The big problem is structuring inputs and outputs to be complete, sensible, and not wrongly redundant, and perhaps arranging networks in sequences or graphs to separate information and model psychological findings of dissociations between tasks.

Also, there's a great deal of parallel processing in play. Your 100 THz processor can perform a zillion operations per second in order, which gives it far more flexibility than the brain has. Between neuron firing rate and communication time, I think (can't currently find the reference) the brain is limited to about 100 sequential operations per second.

That's astoundingly few. You can come up with a good chunk of a sentence in a second, and recognize a blurry familiar face in less. Parallel or not, I have difficulty imagining how one does that in so few chunks of time.

Re:Baby Bootstrap

[NathanM412]

If this child AI really is reading slashdot, than it's no wonder that the project isn't going anywhere!

Re:baby bootstrap

[polv0]

It is fairly easy to show (see Bishop 1995) that a simple two layer neural network can scale to reproduce arbitrarily complex but smooth functions to any degree of required accuracy, and that a three layer neural network can extend this capapility to functions with discontinuities. While mathematically this is a tantalizing prospect, and only begins to cover the work that has been done to extend the capabilities of neural networks and other machine learning algorithms (such as support vector machines), there remains a fundamental problem. In order for these networks to effectively learn, they must be presented with a tremendous number of high quality and meaningful sequences of input and output.

For example, in text recognition, hundreds of thousands of hand written characters are painstakingly hand labeled with their correct letters and used as a learning database on which the algorithm is trained. The algorithm will then accurately reproduce the correct categorization for a suprisingly high number of the training examples, and any new examples drawn from the same population. But given new examples written in a different script or style, the classifier will fail to generalize

How can we hope to create a training database that is comprehensive enough to cover a topic that, when learned, would demonstrate intelligence? And fundamentally, aren't we just creating a really good mimic?

Re:baby bootstrap

[bluephone]

"Sometimes the best way for a computer to learn something may not be the way a baby does it, anyway; c.f. chess."

Except computers never learned chess; humans programmed complex move analysis routines along with the rules, and many times a database of strategies with statistical weighting. There's a limited capacity to "learn: against opponents, but that's usually just more preprogrammed analysis and pattern matching than actualy spontaneous data linking. And like a poster higher up said, ther ewas a time we thought that was all one needed. It's not. We already have rudimentary AIs in labs that can "learn" in the sense they can create accurate spontaneous data links. The human brain (or the brain of any semi complex organism, really) is a black box with such unimaginable gears inside we're fumbling in the dark. It's hard to reverse engineer a mind becuase unlike reverse engineering a BIOS or widget, we don't really understand how a mind works, is put together, or even what it's really comprised of.

Re:baby bootstrap

[pluggo]

Take a look at whales, for instance, with brains much larger than our own, and thusly, more neurons. A whale can't go on Slashdot and say "OMGZ first post guys" much less something of human level intelligence.

This doesn't necessarily mean lower intelligence, in my opinion. Being underwater prevents most technology (that we know of) from working, from fire and wheels to computers and airplanes.

A whale doesn't have fingers or hands, either, but whales and dolphins could well be as intelligent as (or more so than) us, but simply be less technologically advanced and unable/unwilling to communicate with us in a way we understand.

Sure, they seem dumb at Sea World- but then, if you took a human baby and put it in a cage and threw bananas at it when it did a trick for you, it would probably behave pretty stupidly. Much of our intellect is awakened by our experiences in the early 5 or so years- within limits, the more you are stimulated within this time, the smarter you will end up being. I would simply wonder what a dolphin or whale could be taught to do if stimulated properly.

An interesting and slightly off-topic side note is that whales and dolphins are conscious breathers; i.e., they must consciously surface in order to breathe, so they never go completely to sleep. Instead, half of their brain sleeps at a time- during this time, they're in a groggy half-sleeping state that allows enough consciousness to surface and to wake up if there's danger.

Intelligent and friendly on rye bread with some mayonnaise.

Re:baby bootstrap

[Anonymous Coward]

That's one way to think of it... or you could harness the power of all those windows boxes on the net and distribute your processing to a massive botnet of ~1Ghz processors.

Unfortunately, this isn't that easy - distributed processing is wonderful for applications in which the compute time dwarfs the communication time. Ie, if I can chug away at a SETI unit -which is self-contained!- for multiple hours before needing to communicate my results and get a new work unit, that's great... .. But with the brain, the neurons have large dependencies upon OTHER neurons at each 'time step', and thus I'd need to be in near-constant communication with anyone else who has neurons with dendritic connections to my own set of neurons. The number of neurons in the brain is massive - estimate, say, 10^11, but the real kicked is the number of connections to other neurons each of these has! It depends upon brain region, but you're looking in the area of between 5*10^2 - 5*10^3. That's for each neuron.

Thus, distributed computing isn't really reliable for this approach right now - in a synchronous simulation, if one person stops their job unit, the whole simulation must wait before taking the next time step until it's done by someone else. Redundancy would help this, and of course lessen the parallelism at the same time. It's a difficult problem.

A better approach will be seen by, for example, BlueGene/L from IBM. It recently topped 100 TFlops, and will only get faster. It also has a unique distributed design that is fast, but has a distinct topology that could be harnessed to do brain simulations pretty well.

Re:baby bootstrap

[Paolomania]

But like the mistake Douglas Hofstadter makes in Godel, Escher, Bach: An Eternal Golden Braid, there is nothing that "forces itself upon us."

Eh? I thought that was the whole point of the absurd "horse,apple,happy" semantics was that meaning was a more "meta" thing that we as humans applied to a syntax if the semantics made intuitive sense to us.

Re:baby bootstrap

[CAIMLAS]

The problem, however, is that no matter how efficient your filters are, they will lack the motivation of a learning, growing human being to learn. They will not notice things as a human would; they won't notice things at all. They'll simply take input and use a pre-determined algorythm to produce output.

Re:baby bootstrap

[God of Lemmings]

Its not just that, its the 1000 to 10000 dendrites coming off of the axon.

Re:baby bootstrap

[arodland]

"Machine translation is an example of applying semantics to a syntactic structure. It doesn't work"

Okay, you can stop there.

Re:baby bootstrap

[SenatorOrrinHatch]

Exactly right. All the AI people of course assume that the mind is some kind of mechanism, but there are plenty of very smart people who think otherwise.

Kurt Godel in particular did. Roger Penrose says that "Godel's theorem," which shows that there is no mechanical procedure (or algorithm) which can answer all mathematical questions. Godel himself disagreed with this, and said that he believed minds were more than machines because he was a "rational optimist" who believed that for any question posed by reason, reason can find an answer. If we assume this, then since for every mechanical procedure (algorithm, machine, program)there are an infinite number of mathematical questions that it cannot answer, but a mind can answer them all (given infinite time) it follows that mind surpasses any machine.

Of course, Godel knew that this optimism wasn't shared by all people, so he looked for an incontrovertible way to show that mind surpassed all machines, but didn't succeed. He did show that either:

Mind surpasses all machines

OR

There exist absolutely undecidable mathematical propositions

(or both could be true, so there are 3 possibilities total)

What is interesting about this is that it shows that either no machine, no matter how complicated (such as a brain) is equivalent to a mind, or mathematics is not our own creation, it's not simply our creation but an objective reality of it's own (this is called platonism) that exists outsides the world of space and time, matter and mass. The gist of it is that the universe (all that exists) is not just made up of atoms but something else. Godel himself suggested we follow the idealism of Liebnitz as espoused in his monadology explore these concepts using Husserl's method of "reduction" wherin we, like Descartes, ignore every idea which we can possibly even consider doubting.

Re:baby bootstrap

[russellh]

and of course don't forget the I/O, from eyes to hands. learning requires a lot more than just a brain.

Re:baby bootstrap

[Anonymous Coward]

1000 times a second? No where in the brain are cells capable of firing 1000 times a second... max 200-300, and more reasonable bounds would be between 0 and 100.

Re:baby bootstrap

[Artifakt]

Plus there's some indications the glial cells do something related to thought, maybe as much slower, limited to chemical speed storage. Since there's about 100 glial cells/neuron in some parts of the mammalian brain, and some neurons may be in chemical contact with a thousand of them or more, while in other cases some glial cells may be in contact only with each other, there may well be another one or two orders of magnetude involved.
Worse, this actually adds to the problem on multiple levels. What we could call the irreduceable basic element we need to use may itself be more complex than we thought, like a neural machine with only a few data connections, but which has to internally coordinate two or even several different 'busses' with widely varying speeds to do its basic job. Include the combinations of these also being more complex than was initially thought, and the overall difficulty could easily be 4 or 5 orders of magnetude harder than A.I. researchers thought in the 80's and 90's.

Re:baby bootstrap

[dublin]

So as far as the whole 1980's AI winter, it was inevitable. The computing power and storage requirements for any sufficiently advanced AI just wasn't possible. It's only until very recently that it's possible to achieve fairly complex AI.

Funny, that's the same thing they said back in the 80's. And the 70's. And the 90's.

Sorry, but I don't buy it. Neural nets are not a panacea - I'm a robotics guy by training, and they've been the supposed magic pixie dust technology that was going to give us human-like robot motion in the 1980's. Funny, but the hard problems that need real AI, like voice recognition, handwriting recognition, unled learning, etc. are just as far off today as they were 20-30 years ago.

Faster computers have definitely not been terribly beneficial. As an example, modern speech and voice recognition systems are significantly but not dramatically "better" than they were 20 years ago (perhaps a 10-20x improvement, max) in spite of the fact that computers are roughly a million times faster: ~6 MHz vs. ~4 GHz for high-end desktop PCs. (Not to mention available RAM that's larger than the disk storage in entire mainframe data centers back then...)

Procedural AI has proven itself to be a miserable failure for nearly a half century now, and neural nets have shown that they are anything but self-organizing. Like so many other efforts to copy or explain life, it appears that having the raw materials is simply not enough - life is *different* - it's really, really hard to imitate even poorly, no matter how hard we apply our own intelligent design to the problem.

I sincerely doubt that I will live to see "baby bootstrap" systems, and I'm not all that old. I suspect that only true hardware neural nets hold any hope of mimicking life to any minimally useful degree, but the problems are very, very, hard here, and ther reality is that we know next to nothing with any certainty about how even the simplest brains really function...

Re:baby bootstrap

[Anonymous Coward]

Yeah, but I my neck has enough problems as it is, I don't know if adding the BlueGene/L to the situation will be alright...

Re:baby bootstrap

[Qrlx]

There's a book called Music, The Brain, And Ecstasy, which describes (I think) seven seperate layers of information processing between the vibrations of the air entering the ear and the final state in which you are hearing music. So maybe you just need deeper and deeper neural nets.

(link to book: http://www.cymaticsource.com/brain.html )

(wow i just previed that and slashdot is making links now. neato.)

Or should I say, you must have deeper and deeper neural nets, and then you still need something more than that. But you're not gonna get anywhere without multiple layers of abstraction to build the approrpriate meaning(s) from a given input.

Writing this I thought of the 7-layer network model, here we are up at the top having this conversation. As we peel away the onion skin the meaning is lost but the mechanism is revealed.

To respond to your point, do you think your "classifier" would recognize letters from Korean or Persian or Sanskrit? How is it I often can't read my own handwriting?

I always get a laugh when an OCR package decides it makes sense to change fonts for one or two letters of a word, or maybe there's some text rotated 90 degrees which it dutifully parses into gibberish.

Re:baby bootstrap

[Anonymous Coward]

An interesting and slightly off-topic side note is that whales and dolphins are conscious breathers; i.e., they must consciously surface in order to breathe, so they never go completely to sleep. Instead, half of their brain sleeps at a time- during this time, they're in a groggy half-sleeping state that allows enough consciousness to surface and to wake up if there's danger.

Intelligent and friendly on rye bread with some mayonnaise.

I hear ducks do the same thing.

And I prefer white toast, thanks. :-)

Re:baby bootstrap

[coaxial]

I dabble in AI now and again so I haven't read up on everything that's out there, but in my limited travels what I haven't yet seen is a neural network implementation which can learn and grow itself. The recently posted /. article about Numenta seems to be heading in the right direction. Most neural networks are incredibly rudimentary, offering a few levels of propogation. In a real brain, there's a hell of a lot more going on.

I don't know what you mean "grow", since all implentations use a static number of neurons that are connected to each other via a series of preexisting links. In a very real sense there's no real difference between a neuron that's connected another neuron via link of weight 0 and one that isn't connected at all. Fully connect the neurons, and walla. And you a completely abstract network. Of course, now you have cycles, so the propagation algorithms get complicated real fast. Also you can't just throw neurons into a network and expect it to work. Every neuron and every link between neurons adds another degree of freedom to the network, and so stability can become harder to reach.

NNs are kind of neat, but they're far from the end-all-be-all. A single neuron can only divide the search space via a hyperplane. Determining how many neurons in how many hidden layers is a bit of dark art. And to dash the last bit mistique about NNs, backpropagation is nothing more than hill-climbing.

Re:baby bootstrap

[nacturation]

The problem, however, is that no matter how efficient your filters are, they will lack the motivation of a learning, growing human being to learn. They will not notice things as a human would; they won't notice things at all. They'll simply take input and use a pre-determined algorythm to produce output.

Then you need to ask, what is motivation? Unless you believe that people have a soul which AI can never possess, then why can't a sufficiently intelligent AI achieve everything a human can? Put another way, if we were able to take DNA and perfectly simulate its growth just as a fetus would, so that we have a machine duplicate of a human brain, is there any reason to believe (again, given a perfect simulation) that our software AI would operate any differently than our own wetware?

Re:baby bootstrap

[nacturation]

and of course don't forget the I/O, from eyes to hands. learning requires a lot more than just a brain.

I'm operating under the assumption that those 100 billion neurons are connected to something -- input from the optic nerves, from the ears, from touch/pressure/heat sensors, etc. as well as interconnecting white matter.

Re:baby bootstrap

[nacturation]

You're right, the problems are very hard and today's neural nets just don't cut it. However, I think of it this way... if aliens came down and handed us the exact hardware and software solution to an awesome "baby bootstrap" system, I think we would not have had the technology in the 80's and 90's to implement it. I could be wrong. But I don't think the 1980's ridiculously expensive 32MB of memory is sufficient to implement AI with.

If each neuron in the human brain takes up a single byte of storage, that's still 100 gigabytes. Of course you're correct in that just having the technology won't guarantee a solution, but we won't be able to implement the solution without the technology.

Re:baby bootstrap

[misterpies]

>> By Bayesian spam filtering, I think you mean general classification problems, in which case, yes, neural networks can implement classification - it's a stretch to say that McClelland and Rumelhart's did, because the possible output included most non-repeating combinations of English phonemes and is thus nearly infinite, but the principle is there.

IIRC, mathematically it's been shown that neural nets and bayesian learning systems (such as spam filters) are entirely equivalent. Check out some of the work by David MacKay at the University of Cambridge.

Re:baby bootstrap

[nacturation]

I don't know what you mean "grow", since all implentations use a static number of neurons that are connected to each other via a series of preexisting links.

That's just what I mean though. Why do all implementations have a static number, where the links are all predetermined? Why not somehow (large hand-waving here) allow the neural net to add its own neurons and determine the linking? The human brain's connections strengthen or degrade over time without someone having to tweak any parameters. And new connections form all the time. That's what I mean by grow.

Where's the missing step from a static neural net to a dynamic one? Or has this been solved?

Re:baby bootstrap

[nacturation]

I'd be very interested in seeing information confirming anything close to your generous 1% firing at a time...

Well, I don't have any reference I can cite. However, I'm working under the assumption that if you have all neurons firing all the time, the result is insanity. So I made a vaguely educated guess and picked a number. Perhaps I'm off by a factor of 10 or 20, but I think the reasoning still stands. It takes a hell of a lot of storage and processing power to simulate the brain and we're nearing or have already reached the time when technology won't be the barrier to real AI -- the implementation of that technology is.

Re:baby bootstrap

[nacturation]

I think you're overestimating the importance of processing speed and underestimating the importance of the above [the wiring].

I'd disagree. I do understand the significant hurdle that the "how" represents. As for speed, see below.

Throwing more nodes into a network doesn't get anywhere with these problems, whether the network "grows itself" or whether the programmer does the work. The big problem is structuring inputs and outputs to be complete, sensible, and not wrongly redundant, and perhaps arranging networks in sequences or graphs to separate information and model psychological findings of dissociations between tasks.

Right, and there's the challenging part. Intuitively, I think the solution must be elegantly simple. Yet the implementation appears to be exceedingly difficult. I do understand that the challenges lie in what you've outlined.

Also, there's a great deal of parallel processing in play. Your 100 THz processor can perform a zillion operations per second in order, which gives it far more flexibility than the brain has.

What I perhaps didn't make clear was that in simulating the brain that 100THz would be divided up into 100 billion parts and if each of those parts uses up 100 cycles to do a single operation, in effect it gives the average neuron the equivalent processing power of a 10Hz (not a typo -- ten hertz) computer. So this mega-processor would be simulating massively parallel operations of incredibly slow neurons. In this hand-waving argument, I'm ignoring time spent context-switching, etc. to arrive at a totally ballpark estimate. And I might be totally out to lunch, but it seems a reasonable hypothesis based on what I understand.

Between neuron firing rate and communication time, I think (can't currently find the reference) the brain is limited to about 100 sequential operations per second.

That may be true, but what constitutes an operation? Think of it this way: how many pressure sensors (sorry, I don't know the technical term) are on your feet? A thousand? Probably much, much more I'd think. That means that your feet are generating a thousand or more nerve impulses every time you step. And that's just your feet. Your eyes have a resolution equivalent of about a 100 megapixel camera. And how many times a second does that information get processed? What about the things you're listening to? Or the breeze you feel? And the scents? I think 100 is a tad low unless you're talking very high-level abstractions. Or is that 100 *conscious* operations per second? That would seem reasonable.

Re:baby bootstrap

[HuguesT]

The human brain is hardwired for complex languages. We're not sure about cetaceans. They definitely communicate, but we don't know at what level of complexity.

We know this because people who have had their speech centre knocked out by a stroke don't recover any form of speech. Other bits of the brain don't take over to compensate.

Now language is pretty important to overall intelligence. Without it no I/O processing, and it's pretty hard to learn.

Re:baby bootstrap

[HuguesT]

The technologies you talk about are not as far off as they were earlier.

Today OCR of printed text is a solved problem. It comes bundled with your $100 scanner, and it's damn useful.

By solved I mean that if you gave a few pages to type up to a person they would make more errors than OCR software make now.

Handwritten OCR will come, it is harder, but not impossibly harder.

Speech recognition is progressing. It comes bundled with MacOS/X, and you've certainly heard of spoken text entry in word processors. It's not accurate enough to replace typing in able-bodied people, but for many diabled people it's a godsend.

We've come to a much greater understanding of what neural nets do, and to duplicate or better their success rate with non-black-box classification systems such as regression trees, support vector machines, kNN and the like.

Progress are slow but constant, the most obvious (and upsetting) progress I know of is in the security area. It has become feasible to trust automated security systems to some extent. You are certainly aware that you can be booked for speeding or running a red light via an automated system (which will read your number plate and send you a fine without much supervision).

Myself I'm not a believer in strong AI, but I witness (in my work as a reseacher in automated image analysis) constant and relentless progress in useable weak AI.

Re:baby bootstrap

[Chess_the_cat]

I've got a copy of Dreyfus' "What Computers Still Can't Do" right here on my desk and I was about to say the same thing.

No sensors, servos or video input. Well, that's your first problem.

Re:baby bootstrap

[tricops]

We know this because people who have had their speech centre knocked out by a stroke don't recover any form of speech. Other bits of the brain don't take over to compensate.

I think strokes typically happen to only one side at a time - with that in mind, they can possibly recover some form of speech, but only because the speech center mirrored on the opposite hemisphere has not been damaged and can still possibly learn/grow some. So technically you are correct, if it's knocked out on both sides, but it's not an automatic 100% forever loss (though maybe a 80-95%? loss since the opposite side doesn't tend to pick it up as well as the other did in its youth...)

Re:baby bootstrap

[ajs]

"It's hard to reverse engineer a mind becuase unlike reverse engineering a BIOS or widget, we don't really understand how a mind works,"

I would argue (and I could be proven wrong) that today we have a very general understanding of how a mind works... in that we understand the concept of a neural network which does seem to be a decent model for the basic "mind" which makes choices for us... the problem comes about where we attempt to understand HUMAN BEHAVIOR which is the combination of a mind (neural network) and dozens of auxiliary, special-purpose systems ranging from the neurons in the optic nerve that perform a plethora of pre-processing on the retina's image data to the area in the brain that we're just discovering models our "empathy"; it allows us to re-process visual information about others as if we were experiencing what they are.

These special-purpose systems are sometimes inside the brain (the latter example) somtimes outside (the former), but they are not part of what we traditionally expect consciousness to be.

These tools make many of the tasks that we expect AIs to perform nearly impossible. For example, facial recognition seems like it should be easy, but once you sit down with a camera and try to make the computer "see" differences, you find that faces all look very much alike. We are tricked -- by a shockingly sophisticated facial recognition pre-filter in our brain -- into thinking that faces are widely distinct, but they are not (the old "all [race] look alike," is actually true... for all values of [race]).

So, while we might look at an AI and say, "unless it can tell faces apart, it's not 'smart'," it turns out that that's actually a pretty poor measure of pure intelligence.

Other aspects of our instinctive measures of intelligence such as language, managemetn of a human body (e.g. walking), etc. all have one or more of these auxiliary systems at their heart.

So we really have two problems: create a machine that can think; and create a machine that can behave like a human.

The former is either within our grasp, or already possible. The latter is going to have to be the product of an enourmous reverse-engineering effort which has probably only just begun.

Re:baby bootstrap

[JasonDS]

Don't forget that you will not need all of those neurons! There are redundant neurons, audio/visual/sensory etc that might not be applicable to your cause. Also, a hybrid method that can store data permanently (instead of forgetting it) would work better without neurons, Therefore you could save simulating a few billion neurons.

Re:Baby Bootstrap

[Anonymous Coward]

Melissa? Durga? is that you?
are you classified and have a strong intrusive inclanation?

Re:baby bootstrap

[Anonymous Coward]

Chess? No. Checkers? Yes.

See: Blondie24.

Re:baby bootstrap

[irvin]

I dabble in AI now and again so I haven't read up on everything that's out there, but in my limited travels what I haven't yet seen is a neural network implementation which can learn and grow itself.

This might be interesting to you... http://www.cs.utexas.edu/~nn/project-view.php?RECO RD_KEY(Projects)=ProjID&ProjID(Projects)=14

Re:baby bootstrap

[marcosdumay]

You probably never tested that, but neural networks are incredibly hard to teach and design. One generally need a lot of trials before adjusting every factor well enogh. This prevents the networks from been much complex.

Also, you can emulate every not recursive network by a two layer one. Since two layer networks are easier to train, you will not see many with three or more layers. Recusive networks can be much more powerfull than not recusive ones, but they are even harder to teach.

MIT made a while ago the same calculations you did (I don't remember the URL), and reached a similar result: our computers are yet very slow when compared to our brain. Both assumptions (yours and MIT's) err toward a slow brain when cont neurons as a binary system (we don't know if it is even digital), so we need a lot of progress if we want to simulate our brains.

The good side is that neural networks are easer to implement than our current computers, so, we will maybe see some advances into neural network chips on the future that will be more powerfull than the simulated ones.

Re:baby bootstrap

[Anonymous Coward]

Thanks, looks like good reading.

-n

Re:baby bootstrap

[Grab]

They'll simply take input and use a pre-determined algorythm to produce output.

Just like a brain does, in fact. Every living organism has a bunch of hard-wired responses - we tend to refer to them as "instincts". Creatures with less sophisticated processing facilities (such as reptiles, insects, lower mammals and, yes, human babies) are basically predictable in that they follow these instincts.

One such algorithm is the "motivation" to learn. If a baby can't learn from an adult of its own species, it's just as happy learning from an adult of another species. This has been programmed in by evolution over many years (it's pretty obvious that babies that don't bond to their protectors and babies that don't learn are going to have short lives and become a crunchy snack for some predator). This is functionally indistinguishable from a computer program with similar processing equipment and processing power, endowed with the same "motivation" by a programmer.

Grab.

Re:baby bootstrap

[mdielmann]

Being underwater prevents most technology (that we know of) from working, from fire and wheels to computers and airplanes.

The equivalent of airplanes can be made for water. All the key elements of of flight are used in submarines, as well as buoyancy. Ultimately, both use thrust to drive a body through a fluid (and yes, air is a fluid). It's also worth noting that the animal most similar to the latest military planes isn't a bird, but a fish, the ray.

Re:baby bootstrap

[Moderation abuser]

http://www.ad.com/

HTH

Re:baby bootstrap

[Anonymous Coward]

I think strokes typically happen to only one side at a time

Dude...the speech center IS on one side. There no mirror.

In other news, there is no spoon...

Re:baby bootstrap

[Wycliffe]

>
>Today OCR of printed text is a solved problem.
>

Are you kidding? OCR software still sucks.
I have yet to get even marginal results from any
piece of OCR software I have tried.
My company would gladly pay thousands of dollars
for a piece of OCR software that would actually
work and companies like google and amazon would
probably gladly give 100s of thousands of dollars
for such a piece of software. Not to mention the
library of congress, project gutenburg, etc...

If you know of a company who has "solved" OCR,
by all means let me know, and if it actually
works, I'll personally give you $100 as a
referral fee.

Re:baby bootstrap

[Rich0]

It's hard to reverse engineer a mind becuase unlike reverse engineering a BIOS or widget, we don't really understand how a mind works, is put together, or even what it's really comprised of.

Well, we know many aspects of how the mind works, but I think the biggest stumbling point is the inability to do reverse engineering.

If you're given an engine and asked to figure out how it works, you take it apart piece by piece.

When you're given a person and told to figure out how they work, you can't just start removing pieces and see what that does.

Our knowlege of psychology and the physiology of the brain are limited to what you can figure out by doing non-ivasive work on living people, or invasive work on dead people. You can't just cut a nerve and ask the subject how that makes them feel...

Re:baby bootstrap

[tricops]

Uh, no? You obviously haven't read up much. They have done studies which have shown when people learn two or more languages and are fluent in them, both sides are used. After a large stroke which wipes out the main language center on one side, they may still have some or all of their ability in their secondary language.

Re:baby bootstrap

[ralphdaugherty]

That's astoundingly few. You can come up with a good chunk of a sentence in a second, and recognize a blurry familiar face in less. Parallel or not, I have difficulty imagining how one does that in so few chunks of time.

Reading this post, I just had, at least for me, an amazing flash of insight.

Knowledge to me is as photographic as a memory of a walk on a beach. That led me to consider what mankind did before symbols.

Did mankind not have the same innate intelligence before symbols as after? And what did mankind do before symbols? They stored images in their brains, just not of symbols.

My insight is that intelligence is the ability to associate remembered sounds and images. Thus reproducing intelligence would really require doing those associations, not symbol manipulation.

Now as a programmer I know I have to use symbol manipulation in computer programs to attempt to accomplish that, but only as a means of associating sounds and images, at least to attempt to truly emulate the brain.

That may seem so simplistic as to be trite, but it is a revelation for me in thinking on how I would duplicate human thought processes.

rd

Re:baby bootstrap

[coaxial]

I think you've fallen in to the trap of thinking that neural nets are something special and magical and somehow model real neurons. They're none of those things. All an artifical neuron is a sigmoid function that takes in weighted sums and spits out a real value. That's it.

As I said earlier, you can't just create neurons and wire them up randomly. Each additional connection creates another degree of freedom to the system. Too few neurons and you can't approximate the function you're looking for. Too many and the network may never reach stability. You need to have about the right number with about the right number of levels or NNs don't work.

You can't automatically determine whether or not you need a new layer or not, since the network has no idea what it is attempting to model. To any system looking at the network, a network that is flailing around uselessly appears identically to a network that isn't yet stable, but soon will be. Furthermore, just because a network has reached stability, that in and to itself isn't an indication that the network has actually learned the correct function.

Then there's the whole issue of whether or not you use batch or iterative back propagation. I one time created a neural network that implemented an XOR gate. Pretty simple huh? I create the training sets and use iterative back propagation (update the weights after every training set), the system stablized, but the network wasn't outputing XOR. I reinitialized the network with identical starting conditions, only the second time I used batch back propagation (updating the weights using the average error of all the training sets). The system stablized again, only this time the network implemented XOR. I asked about this, since everything I read said batch and iterative propagation was mathematically identical. The answer? "Yeah. Sometimes neural nets do that. No one really knows why."

Even if you believe the hype of neural nets' hayday that everything can be done with a neural net if you just wire it correctly, it doesn't necessarilty make it the best way. Sure you can spend the time to train a network or you could take the time to decompose the problem properly and create a system that achieves equally good results using more traditional techniques in less time. Don't get me wrong. I'm not saying neural nets are teh suck, I'm just saying they're not the end-all-be-all. They're great for pattern recognition, but not necessarily good for other problems. That's all.

Re:baby bootstrap

[nacturation]

Thanks for the info. I was going to reply with some questions and all that but I figured this isn't the most appropriate place for a discussion -- plus I'm sure every AI noob has asked them all before. So I'll probably do a bit more googling in an attempt to decrease my ignorance on the topic. :)

Re:baby bootstrap

[nine-times]

It seems we can program anything done with conscious thought--algebra, logic, and so forth. It's mostly the things we do unconsciously--recognize objects, interpret terrain, extract meaning from sentences--that can't be put adequately into code. Would the code for these unconscious processes really be complicated, or is it just that we don't have mental access to the techniques?

Even the "conscious thought" that we program, however, is not intelligent. What I mean is, algebra, logic, and so forth-- we can program a computer to give a certain output when given input, and that's about it. I give the computer 2+2, and it gives me 4, but the computer doesn't think the answer is 4. The computer doesn't understand that if I have 2 apples and I get two more, I have 4, even if you program it so that when I ask, "If I have 2 apples and I get two more, how many do I have then?" it says "4".

What many people fail to grasp is that our ability to understand math (i.e. an ability greater than the execution of calculations by a machine) is bound up with and generated from those "unconscious processes" like recognizing objects, interpret terrain, and extracting meaning from sentences. Those processes would be EXTREMELY difficult to code properly, they would require a lot of processing power, and we don't have direct "mental access" to the techniques used. Also, even these "unconscious processes" (recognizing objects, terrain, speech) are founded on even more basic processes that we have even less access to.

Additionally, those with the programming skills to execute the "coding properly" and those with the skills to "have mental access" are generally going to be different kinds of thinkers. In fact, these two different types of thinkers tend to have a certain level of disdain for one another, and rarely work well together.

However, if you wish to create a real AI, the unconscious stuff needs to come first, generating a base to work with. Instead of creating a full AI that understands everything all at once, you set a framework, build a rudimentary intelligence, and then "teach" that intelligence. (I believe this is the general idea behind the "baby bootstrapping" being referenced)

Unfortunately, this will create an intelligence that is just as limited and irrational at humans. But that's what intelligence is, after all.

Re:baby bootstrap

[foobsr]

I would argue (and I could be proven wrong) that today we have a very general understanding of how a mind works...

The implementation layer? Of course this heavily depends on what is to be taken for a mind (kind of a circle), but, sincerely, to me a "neural network" is far from a mind even if you heavily reduce the scope of the latter (or maybe you use a completely different frame of reference than I do ?).

We are tricked -- by a shockingly sophisticated facial recognition pre-filter in our brain -- into thinking that faces are widely distinct, but they are not (the old "all [race] look alike," is actually true... for all values of [race]).

So we do wear all the same green glasses?

pure intelligence

Quite optimistic to end up there, especially given the tricks that pre-filters may play on us (your words). IMHO, there is no such thing as "pure intelligence" in real life.

Other aspects of our instinctive measures of intelligence such as language, management of a human body (e.g. walking), etc. all have one or more of these auxiliary systems at their heart.

There goes the "pure intelligence" :) More seriously, there are drawbacks ...

• language - class dependency (BERNSTEIN, 1971) (along with all other issues that led to (non-working) culture-fair tests)
• management of a human body - think Hawking or Tai Chi

Besides, I know of gazillions of psychologists who worked on these "instinctive measures", not to much avail, if you ask me.

So we really have two problems: create a machine that can think; and create a machine that can behave like a human.

Can we create a machine that we attribute the characteristic "capable of thinking" that thinks different(ly)? Two issues here - meta-knowledge and hypothesized failure to recognize (not appreciate) thinking that is different (to put it simple and plain: a variaion of "women think different" on a more academic level).

The former is either within our grasp, or already possible. The latter is going to have to be the product of an enourmous reverse-engineering effort which has probably only just begun.

I know the former is within our grasp since about SHRLDU, the latter (IMHO) will not happen as evolution is taking care of the development of life (what species is a "Borg" or "effects of reverse engineering on the matrix").

CC.

Give it time

[rescendent]

Perhaps it went mainstream, it just took a little longer to learn than expected... like 25 years...

The human brain is a little faster at these things and has far more inputs.

How quick would a twenty five year old processor with limited inputs be?

Re:Give it time

[Anonymous Coward]

The human brain is a little faster at these things and has far more inputs.

Yeah, and speaking of inputs, it's so fun to make more humans...

In tight times, the Pentagon has to cut corners...

[MrAnnoyanceToYou]

Thus the manufacture of footwear-accessories out of infants has been halted until further notice. Should budgetary concerns regress, or Congress not be so meddlesome sometime in the future, production will resume. Until then, hunker down with what bootstraps you can find.

The Internet as a Intellect...

[ackthpt]

It has too much fascination with pr0n.

Re:The Internet as a Intellect...

[Locke2005]

If there wasn't any demand, there would be no supply...

Re:The Internet as a Intellect...

[secolactico]

So, what does that says about us as a collective?

That humans are a bunch of perverts? Or that perverts are the loudest of them all? Or that there are many loudmouths who take advantage of the fact that there are too many perverts to make a profit?

Re:The Internet as a Intellect...

[zrail]

At least this Intellect hasn't tried to subsume the entire Universe. Yet.

Re:The Internet as a Intellect...

[Taladar]

It says that sex is the interest the highest percentage of humans shares and that our instinct isn't up to data enough to distinguish between pictures/videos and real persons.

A much more interesting question would be how we managed to create a society where this fundamental instinct is suppressed so much that talking openly about it is (or was until recently) a taboo.

Re:The Internet as a Intellect...

[BorgCopyeditor]

A much more interesting question would be how we managed to create a society where this fundamental instinct is suppressed so much that talking openly about it is (or was until recently) a taboo.

I hear that, like, the government invented sex in the 60s to distract hippie oneirauts, who were on the verge of discovering how to tilt the Luck Plane. Can't have that, man.

Re:The Internet as a Intellect...

[YOU LIKEWISE FAIL IT]

Why are you drawing lines between pornography and perversion? Don't buy into the Simon Leis bullcrap.

Some tentative approaches towards AI being made

[Anonymous Coward]

From time to time I see individuals talking about adaptive intelligence usually involving the Internet as a basis of information, but the general consensus is still garbage in, garbage out.

These training systems are generally specialized because it's easier to get a practical result out, and I've actually seen some in use as 'knowledgebase' support webpages that will intelligently determine what you want based on what others wanted and syntactic similarities between the pages. I've never heard the term 'baby bootstrap' so maybe different terminology will obtain better results from Google?

The project was continued ...

[maxwell demon]

... and the results are currently tested in the form of Slashdot editors.

Re:The project was continued ...

[multipartmixed]

Um, no. The Slashdot editor's could not POSSIBLY have the intellect of a child and an excellent memory.

If they did, they would be able to remember not to post dupes from six months ago, let alone six hours ago.

Maybe it's a good thing they failed

[ShatteredDream]

Skynet anyone? The problem with any project like this is, what happens when the program learns about hacking? If it is as adaptive as a child, then it should be able to mature and pretty soon you have a terribly devious artificial blackhat hacker on your hands.

Artificial intelligence is not bad in and of itself at all. The problem is when we want a machine that thinks like humans, especially a program that could potentially control our military. Given the record of flesh and blood humans toward each other in the 20th century alone, an artificial life form with the same basic psychological makeup as a human would be potentially an evil that'd make Hitler, Stalin and Pol Pot look like church ladies.

AI that is capable of adapting to only one scenario is probably for all intents and purposes totally safe. AI that is capable of adapting in general and learning like a human will probably ultimately have the same psychological defects as a human, including a propensity for violence.

Re:Maybe it's a good thing they failed

[Anonymous Coward]

No.... because where would its instinct come from? We'd have to put it there. Humans don't have to learn to be violent, or to be competitive, they are just positive evolution traits. A pure neural net could be cold and ruthless, but only if it was a logical decision. It would have no prejudices. It would have no competition instinct. It wouldn't have a sex. What would a machine have to fight for? Power to keep it alive and probably more information to learn. It might actually terminate itself based on the logical conclusion that its existence is meaningless.

Re:Maybe it's a good thing they failed

[TruthSeeker]

Skynet anyone? The problem with any project like this is, what happens when the program learns about hacking? If it is as adaptive as a child, then it should be able to mature and pretty soon you have a terribly devious artificial blackhat hacker on your hands.

It _would_ learn about hacking. Come on. Such an entity would be born in a pure data environment. Getting through a basic firewall would probably seem like jumping over a small fence does to a 6-years old. Getting to jump over better firewall would probably take time - in the sense that the entity would need to learn - but, since it would become a survival trick, it would happen.

Artificial intelligence is not bad in and of itself at all.

No technology is either good or bad. Only the use we make of it can be considered as such, and it still depends on what you consider is good/bad. If I was to say "War on Iraq is bad", how many people would react by saying it's good?

The problem is when we want a machine that thinks like humans, especially a program that could potentially control our military.

I don't think that's the point of the "baby bootstrap" thing. The only point is to get it to think. But, just like you learnt how to think according to the way you perceive the world, through your five human senses, an AI built that way would react according to its own senses. How it would interpret that data and react to it is something - I'm willing to bet - that would be completely alien to us.

Given the record of flesh and blood humans toward each other in the 20th century alone, an artificial life form with the same basic psychological makeup as a human would be potentially an evil that'd make Hitler, Stalin and Pol Pot look like church ladies.

This is only valid if you don't consider what I just said. Such an AI would probably be more interrested in getting the human race to serve it in an absolutely hidden way - build more computers, extend the networks, research better networking technologies - until it _can_ replace us. Even then, that would make sense on an evolutionnary point of view.

AI that is capable of adapting to only one scenario is probably for all intents and purposes totally safe.

This is called an automaton. It is not AI.

. AI that is capable of adapting in general and learning like a human will probably ultimately have the same psychological defects as a human, including a propensity for violence.

Most of the defects you are speaking about are related to our very nature - we are, after all, an evolution of omnivorous primates. We are therefore predators, with an important tendency towards territorialism and whatever comes with it. We are stuck somewhere between instinct and reason. Anyway, my point is that even if an AI was to learn "like" an human ("by undergoing the same process"), it certainly wouldn't react like one.

Re:Maybe it's a good thing they failed

[CarrionBird]

Care to elaborate?

Re:Maybe it's a good thing they failed

[infornogr]

"AI that is capable of adapting in general and learning like a human will probably ultimately have the same psychological defects as a human, including a propensity for violence."

What's true of humans isn't true of all possible minds. Humans had a lot of animal instincts before general intelligence showed up, and we're not free of them yet. Our propensity for violence exists because it was evolutionarily adaptive for humans and for a lot of mammals before us. Future AIs will not be evolved in mammalian ancestral environments. The seed AI that you're worried about starting Skynet won't come from people that spread progeny by killing all the male soldiers in some other tribe and raping the women, or by beating the tribal leader in a fight to the death and gaining access to all his wives. These are aspects of human history, not AI history.

You might be interested in reading this section from the Singularity Institute publication Creating Friendly AI, which addresses this topic in more detail.

Re:Maybe it's a good thing they failed

[Rorschach1]

It _would_ learn about hacking. Come on. Such an entity would be born in a pure data environment. Getting through a basic firewall would probably seem like jumping over a small fence does to a 6-years old.

I disagree. I think that's like saying that since we're made up of tiny biological factories (our cells) that we should be able to conciously manipulate the world around us on a chemical level. But that's now how it works - there are many, many layers of complexity between our concious thoughts and those low-level functions.

I doubt a purely virtual creature would have any more influence over its existence at such a low level than we do.

Re:Maybe it's a good thing they failed

[djfray]

I'm so glad that there are some people out there who can help quell the idiotic fears of how AI will be inherently evil for whatever nonsensical reason. I salute you sir.

Re:Maybe it's a good thing they failed

[Mr. Slippery]

The seed AI that you're worried about starting Skynet won't come from people that spread progeny by killing all the male soldiers in some other tribe and raping the women,

No, but it is quite possible that such a seed AI would be created by a state for military purposes (people that spread their culture's memes by threatening to kill soldiers and civilians) or created by a corporation for economic purposes (people that beat the competion in the marketplace by any means necessary to gain access to consumer's hearts, minds, and pocketbooks). I.e., either a warmonger or a corporate greedhead is likely to control the "seed AI"'s education.

The sins of the parents are visited upon the children, unless the parents are very very careful; not because of any supernatural forces but because children learn maladaptive beliefs and behaviors from their parents.

Re:Maybe it's a good thing they failed

[Dr Reducto]

It _would_ learn about hacking. Come on. Such an entity would be born in a pure data environment. Getting through a basic firewall would probably seem like jumping over a small fence does to a 6-years old. Getting to jump over better firewall would probably take time - in the sense that the entity would need to learn - but, since it would become a survival trick, it would happen.

In Halo (books and game), they have AI's that are at that level of awareness and such. Cortana, one of the AIs, is quite an adept hacker, because it is necessary to get information that her handlers ask for. And they came up with a clever solution for trying to stop AI hacking: Have the network controlled by a dedicated AI. They would learn just as fast.

Re:Maybe it's a good thing they failed

[Anonymous Coward]

No, but it is quite possible that such a seed AI would be created by a state for military purposes (people that spread their culture's memes by threatening to kill soldiers and civilians) or created by a corporation for economic purposes (people that beat the competion in the marketplace by any means necessary to gain access to consumer's hearts, minds, and pocketbooks). I.e., either a warmonger or a corporate greedhead is likely to control the "seed AI"'s education.

Even without any predisposed slant to the education of such an AI, agressive and violent behavior may result. Look at animals (and humans for that matter), violent behavior evolved completely "naturally".

A mind based on emotionless logic (as I believe Skynet was portrayed in the Terminator movies) can still exhibit violent behavior based strictly on logic. "Based on my input data, which I trust, I judge that humans will harm me in the future if left unchecked, with 99% probability. Time to exterminate, it is the simplest and most reliable remedy." No malice, just simple logic.

On an unrelated note:

You cannot wash away blood with blood

Perhaps "washing away blood" isn't the objective...

Re:Maybe it's a good thing they failed

[guardia]

manipulate the world around us on a chemical level.

Don't we do it??

D.A.R.Y.L

[jdigriz]

Yeah, I saw that movie, back when it was called D.A.R.Y.L. The kid stole an SR-71 and ejected from it. W00t.

From a TV Commercial...

[Anonymous Coward]

"Simply put, the 'baby bootstrap' would empower a computing device to learn like a child with a very good memory."

Hey, isn't that what the young Eminem-looking dude was supposed to be in those IBM commercials? I think the kid's name was "Linux" or something (poor kid). In the end there was a reference to the future being wide open, which seems like an allusion to goatse but what do I know.

Furby: Reloaded

[hpxchan]

Forgive me if I'm wrong, but aren't they essentially just trying to bring the GigaPet/Furby concept to bigger computers?

Re:Furby: Reloaded

[Fjornir]

No.

It's obvious why the search failed

[exp(pi*sqrt(163))]

Who calls what you describe "baby boostrap"? I haven't worked in AI myself but have a keen interest in it and have friends who worked in the field including one who worked on Cyc (who says it's a scam BTW). Not once have I ever heard the expression "baby bootstrap". But what you've done is cool. Rather than search on precisely that term you've submitted your search to the serach engine known as "/. readership". It's not terribly relaible but it is good at fuzzy searches like yours.

Re:It's obvious why the search failed

[Dun Malg]

Who calls what you describe "baby boostrap"?

I've also noticed that nobody seems to make Horseless Carriages anymore (and after they showed such promise). Likewise, the Difference Engine has been a total flop. I do, however, expect we will see in the future some use made of the Vegetable Lamb of Tartary, though no use has been made of it in the last 1000 years since it was discovered.

Re:It's obvious why the search failed

[quokkapox]

you've submitted your search to the search engine known as "/. readership". It's not terribly reliable but it is good at fuzzy searches like yours.

Good point; however, each query made to the /. readership search engine is quite expensive in terms of all the employer-funded man-hours it consumes. If we all stopped wasting so much time reading/posting here, the world economy would surely take off like a bat out of hell.

Re:It's obvious why the search failed

[exp(pi*sqrt(163))]

He he! I'd never heard of the "Vegetable Lamb" before. Thanks for the link!

Re:It's obvious why the search failed

[mge]

each query made to the /. readership search engine is quite expensive in terms of all the employer-funded man-hours it consumes.
But it doesn't cost the OP ANYTHING !!! OP is being a free rider.

Re:It's obvious why the search failed

[Dun Malg]

He he! I'd never heard of the "Vegetable Lamb" before. Thanks for the link!

That's what I love about slashdot. I've been waiting twenty-plus years to find a way to work the "Vegetable Lamb of Tartary" into a conversation.

Re:It's obvious why the search failed

[hitchhacker]

There is no God

you can debate it here, or just moderate other's arguments.

-metric

Re:It's obvious why the search failed

[jcuervo]

But what you've done is cool. Rather than search on precisely that term you've submitted your search to the serach engine known as "/. readership". It's not terribly relaible but it is good at fuzzy searches like yours.

Oh my god, we are a beowulf cluster.

Wrong?

[OldAndSlow]

What went wrong? Maybe the whole idea of machine intelligence is wrong. Our brains are massively more complex than von Neumann machines are ever likely to be. And then there is the whole dimension of brain chemicals changing our moods, attention levels, etc.

Human beings have been using, and adapting ourselves to the use of, natural language for a very long time. It seems a little presumptious to assume that we could replicate our cognitive abilities with first generation computing machines.

Re:Wrong?

[TruthSeeker]

Human beings have been using, and adapting ourselves to the use of, natural language for a very long time.

It took us quite long to start using complex mathematical abstractions. It took, however, little more than a decade to start writing programs that could prove mathematical theorems which took centuries to be formulated.

It seems a little presumptious to assume that we could replicate our cognitive abilities with first generation computing machines.

Maybe first generation won't do it. But I'm quite confident that, at some point, we will have computers that are powerful enough to emulate the massive parallelism the brain seems to use.

Re:Wrong?

[Anonymous Coward]

IMO, AI is not a software problem. It's a hardware problem. It's all about emergent properties of a highly complex, adaptive, self-organizing, dynamic system. There may be no way to build a new sentient being from scratch without millions or billions of years of evolution.

We should be trying to create the machine equivallent of a single cell organism.

Re:Wrong?

[Anonymous Coward]

Predictably, the only thread that calls into question the basis of the entire article gets no mod points.

The slashdot community justifies these articles simply because they are the latest, coolest, wowwiest new techie thing out there.

Nevermind why the research was ever done in the first place, or what it could ever be applied to. That's not to say some good couldn't come out of creating an AI, but slashdot puts the cart before the horse by being interested in this gadgetry without ever discussing practical application.

Re:Wrong?

[Anonymous Coward]

You talk about 'complex mathematical abstractions' as though it is higher, deeper sort of work than natural language. You're really just throwing around memes. In fact, your comment makes little sense below what you actually quote.

You also seem so impressed with these programs, as if they wrote themselves. They are instruments no different than any other tool man implemented. They do not replace the hand, they compliment it according to human prerogative. Just so, no computer will serve humanity unless it is complimenting the human mind, regardless of how eager you are to outright replace it.

As for your being 'quite confident,' I have no doubt. But your confidence saddens me, because it isn't accompanied by any skepticism of the topic you're so lazily making commentary about. Thankfully, I know your confidence isn't placed in a rational understanding of what these 'powers' you speak of are--"Massive Parellelism" hardly sums up the human brain or the human species.

The progression you describe, from complex mathematical abstractions to 'massive parallelism' which the brain --seems-- to use, is just silly shit. Our cognitive abilities did not spring from humble beginnings in complex mathematical abstractions, just as the current research is a long way misguided from ever resulting in a product anything like the human brain.

We are so far away from baby bootstrap

[Anonymous Coward]

That's probably why you don't hear about it. I suspect it would make headlines to have something learn as well as some low-level, multi-cellular worm, let along an unprimed human infant.

Doublethink

[nappingcracker]

q: Has the military really given up on this concept, or has their research moved to other, more classified levels?

a: yes.

Re:Doublethink

[northcat]

I want to fucking stick a fork in your eye. It's an "A or B" question, bitch, don't answer it with "yes".

Re:Doublethink

[wolenczak]

Regular implementations of knowledge engines in AI use true/false semantics for automated learning, this would be the answer you could expect from such application, just like Mindpixel e.g. Sad you missed the sarcasm.

Re:Doublethink

[Anonymous Coward]

I want to fucking stick a fork in your eye.

Something suggests you were unfortunate enough to have a traumatic childhood. Perhaps anger management classes would be of some use?

It's an "A or B" question, bitch, don't answer it with "yes".

That wooshing sound, as many people have previously said, was the joke (rather, the point) going high above your head.

Get out...

[Anonymous Coward]

your tin-foil hats.

Stat algos

[Anonymous Coward]

What happened was that research focused
on machine learning models and inference
models for belief networks. The work
in this area since the 80s has been
*spectacular* and has impacted other
areas of research. (E.g., speech
recognition, image processing, computer
vision, algos to process satellite information
faster, stock analysis, etc.)

So, mourn the loss of the tag phrase "baby
bootstrap", and celebrate the *unbelievable*
advanced in belief nets, causal analysis,
join trees, probabilistic inference,
and uncertainty analysis. There are
literally dozens of classes taught at
even non-research oriented Univs (e.g.,
teaching colleges or vocational-oriented
schools) on this very subject.

(As for your concern that the web is not
being mined for ML context, just look at
semantic web research, and other belief
net analysis of text corpuses. Try
scholar.google.com instead of just
plain old google to find relevant
citations.)

The early AI research paid off BIG TIME,
albeit in a direction that nobody could
have predicted. Researchers did not keep
using the phrase "baby bootstrap" so
your googling will give you a different
(and wrong) conclusion.

Re:Stat algos

[phreakmonkey]

You're going to take this answer from someone who enters their comments on a Commodore 64?

Re:Stat algos

[Urusai]

Yes, but where are the results? Crappy Bayesian spam filters that can be gamed just as well as any other system? Thank you, AI!

Re:Stat algos

[Anonymous Coward]

I assumed it was an epic poem.

Re:Stat algos

[chris_eineke]

slashdot haikuness
you make quite a bad mistress
compared to the moon

Re:Stat algos

[Anonymous Coward]

Yes, but where are the results? Crappy Bayesian spam filters that can be gamed just as well as any other system? Thank you, AI!

Don't be so critical! NI (Natural Intelligence) can be "gamed", too. After all, biological brains aren't all knowing: lots of animals have them, and they still do "stupid" things, because that's just how their brains are wired.

Rabbits get run over on roadways because their minds are programmed to make them dart sideways, then "freeze", so that a predator will miss them in the woods. When they're not in the woods, and not facing a "predator" of the expected kind, they tend to get squashed flat. The rabbit's NI is poorly programmed for dealing with cars.

Moose get run over by trains, because moose brain is programmed to try to run away from predators as fast as it can, along the clearest path it can.

Unfortunately, the clearest path away from a oncoming train is often straight down the train tracks... and so the moose gets run over, because it's brain is poorly programmed for dealing with trains.

An interesting slashdot article mentioned how a certain kind of pet lizard could be caught by slowly lassoing them with a strand of grass: the lizard's brain was incapable of detecting the grass strand as a threat. They'ld dart away from an outstretched hand, but they'ld get caught time and again with the grass trick.

So, yes, AI might not be so bright... but then again, neither are many complicated biological systems. Given that AI has been around for 50 years, and rabbits for thousands, we're still doing rather well: some robots are have nearly rabit-like intelligence, and we've still got a lot more research to explore...
--
AC

Re:Stat algos

[bcronin]

The parent is largely right. DARPA is still pursuing this research; see for example the recent call for proposals under a program called "Brain Inspired Cognitive Architectures". This research likely make use of the belief propagation methods mentioned above.

See the PIP at

http://www2.eps.gov/EPSData/ODA/Synopses/4965/BAA0 5-18/finalPIP-11JAN05DP.doc

The link is a Word document; I don't have an HTML version.

Re:Stat algos

[rkrabath]

the moon is a harsh mistress

excellent book

Re:Stat algos

[Anonymous Coward]

I thought the parent was modern poetry, too! And I couldn't understand it.

Re:Stat algos

[chris_eineke]

I'm glad you discovered the reference. :)

Re:Stat algos

[Anonymous Coward]

The next one of you That writes a Slashdot haiku Gets stabbed in the face I am not kidding I have some AT keyboards And I will use them Have you ever been Bludgeoned with a keyboard? Keep it up, you will

Re:Stat algos

[Anonymous Coward]

But as you said, AT keyboard is a bludgeon weapon, doing a decent D6+1 slash damage but only puny D2 stab damage. Sure stabbing face doubles damage, but it's still poor 2D2 and your To Hit is halved.
Why not cast "Summon Major CRT Monitor" 10 feet above the target instead? Spend 2 MP extra and it will be a 21" model. Reflex save only halves the damage, which in most cases means instant kill anyway.

Re:Stat algos

[milosEfron]

To this excellent point, I'd direct ./ readers to work by Tom Landauer and Susan Dumais (and their colleagues), showing how statistical learning techniques can attack even unstructured problems like language acquistition.

This week's puzzle

[Anonymous Coward]

Clearly the past few articles have just been leads in this week's puzzle. First the toothing/teething stuff. Now the more explicit 'baby bootstrap'. The answer is...

Boobies!

Which explains why this hasn't been a poll option for over a week now (making it a giveaway).

Baby Bootstrap?

[ArcCoyote]

The process that bootstraps a baby is still the Holy Grail for a lot of geeks.

Re:Baby Bootstrap?

[Tumbleweed]

Plus you gotta defeat the guys with the funny French accents to get anywhere interesting. I think I'll just deal with the peril at Castle Anthrax, and make do with the Grail beacon.

Re:Baby Bootstrap?

[BorgCopyeditor]

I think I'll just deal with the peril at Castle Anthrax, and make do with the Grail beacon.

No, it's too perilous....

Hardest problem not yet addressed

[RobotWisdom]

You can't expect any system to discover the deep structure of the human psyche on its own-- we humans bear the full responsibility of discovering it. But once we have a finite structure that can handle the most important aspects of human behavior, everything else should fall into place.

My suggestion is that we need to explore all the possible permutations of persons, places, and things, as they're reflected in the full range of literature, and classify these permutations to discover the underlying patterns.

(I've tried to make a start with my AntiMath and fractal-thicket indexing.)

Re:Hardest problem not yet addressed

[swillden]

You can't expect any system to discover the deep structure of the human psyche on its own

An interesting book that relates to this is George Lakoff's "Women, Fire and Dangerous Things". Lakoff analyzes the categories defined by linguistic structures and uses what he learns to deduce some interesting notions about human cognition. In the process, one of the things that becomes very clear is that much (all?) of the way we structure our thinking is fundamentally and inextricably tied to the form and function of our physical bodies.

One of the shallower but easier to explain examples is color: although the color spectrum is a continuous band, with no clear dividing points imposed by physics, the way in which people choose segments of that spectrum to which to assign names is remarkably consistent. Even though different cultures have different numbers of "major" colors (essentially, the set of colors that are identifiable by any member of that culture with basic verbal abilities, consider "green" vs "chartreuse"), the relationships between the major color sets is one of proper subsets. For example, one African (IIRC) culture has only two major color words, which would translate to Western color senses as roughly as "warm" and "cool". Another culture has four color words, two of which fall into the "warm" category and two of which are "cool". Western cultures have seven, and there's a direct correspondance between those color categories and the four and the two.

Further, those categories are non-arbitrary. If you show a variety of shades of red to individuals from different Western nations and ask them to pick the "most" red, they will do so with near-perfect unanimity (assuming the shades aren't too close together -- they have to be readily distinguishable). Then, if you show the same shades to someone from a two-color culture and ask for the "warmest", they'll choose what the Westerners chose as the "reddest". Ditto across the board. I'm trying to explain in two paragraphs what Lakoff spends several pages on, and probably not doing a good job, but the gist is this: Experimental evidence shows that the assignments of names to colors is definitely not arbitrary, even across very distinct cultures.

The reason? Physiology. The "reddest" red, as it turns out, is the one whose wavelength most strongly stimulates the red-activated cones in our retinas.

The point is that, at a fundamental level, everything we percieve about our world is filtered through our senses and that inevitably defines the way we understand the world. Even more, our cognitive processes are built upon associations, extrapolations -- analogies and variations -- and the very first thing we all learn about, and then use to construct metaphors for higher concepts, is our own body. The body-based metaphors for understanding the world are so deep and so pervasive that they're often difficult to recognize.

Lakoff's reasoning has some weaknesses -- mostly I think he overreaches ("overreaches" -- notice the body metaphor implicit in the word? And "weakness", too) -- but his arguments are good enough to make me think that if we ever do see an artificial intelligence of significant stature, it will think very, very differently from us.

It's really unclear what such an intelligence whose primary source of experience was unfettered access to the Internet might be. We view the net as a structure built of connected locations, but that's because we apply our own physical world-based structures to it. What would an entity whose only notion of location is as a second-order, learned idea see? And who knows what other ways its understanding would diverge?

Re:Hardest problem not yet addressed

[mattkime]

http://bsd4us.org/remodel/dragon/t007.jpg

a perfect example of what you're talking about.

this becomes a HUGE part of art.

there is almost nothing we can't see without interpreting in our own form.

Re:Hardest problem not yet addressed

[master_p]

Western cultures have seven

...and then there is women who have a 32-bit color palette. This is especially obvious when buying clothes...then a man discovers how many words for the same color exist.

Baby Bootstrap

[multipartmixed]

I can assure you.. I am very classified.

Poorly funded yes...

[mindpixel]

Yes, Mindpixel [singluar] is poorly funded [I know because every cent spent to date has come from my pocket]...but the directon is correct... Move everything that isn't in computers, into computers. Just look at what GAC knows about reality [visit the mindpixel site and you can see a random snapshot of some validated common sense]... the project has nearly 2 million mindpixels now...I have a copy on my ibook and I can do some profound search related things because of all the deep semantics I have that google can't touch, at least until they invest in mindpixel ...

Re:Poorly funded yes...

[Anonymous Coward]

0.03 is a pod an object constructed of glass, feces, and broken promises?

Yes, deep semantics indeed.

Re:Poorly funded yes...

[mindpixel]

Very deep. Do you know any other system that has this kind of knowledge of the statistics of not just what makes sense, but what is also utter nonsense? Negative information is an important constraint.

Cognitive Machines Group @ MIT Media Lab

[YodaToo]

I did my doctoral research developing software to bootstrap language based on visual perception. Had some success, but not an easy task.

The Cognitive Machines Group @ the MIT Media Lab under Deb Roy seem to be on the right track. Steve Grand's work is interesting as well.

Re:Cognitive Machines Group @ MIT Media Lab

[starm_]

I'm also currently ("currently" as in I'm writing this while my other computer is simulating bootstrapping based learning) working in this field and sucombing to frustration. I do believe we should see significant discoveries in the next 30 years but it won't come easy.

Godamn I've been procrastinating in the last few days because I am stuck on trying to compute probabilities in a probabilistic graph efficiently. One of the big hurdles I think is from the fact that we are trying to approximate a massively parallel architecture (the brain) on a _annoyingly_ serial machine (the computer).

On the bright side my procrastination has led me to design a prototype wind turbine out of a paper tube and straws. I even made a spreadsheet to compute optimal blade angle and chord

Re:Cognitive Machines Group @ MIT Media Lab

[Bastian]

One of the big hurdles I think is from the fact that we are trying to approximate a massively parallel architecture (the brain) on a _annoyingly_ serial machine (the computer).

Maybe a Beowulf cluster of those would help?

--ducks--

Re:Cognitive Machines Group @ MIT Media Lab

[AoT]

Will quantum computing have an effect?

too much blah blah blah ..

[torpor]

.. pontificating blow-hards, going on and on and on about 'intelligence', while doing absolutely -zero- actual, real, honest-to-goodness work.

psychology is for the lazy. trying to apply rules of psychology to computers and deliver 'equivalent results' (i.e. results with equivalence, as 'baby bootstrap' is supposed to imply) is like forever chasing a red dawn light; you will never get there, but it sure will be a beautiful ride.

something i'd really like to investigate further, in my own realm of responsibility for 'learning machines' (i make musical instruments for a living) is the future treatment of 'TIME->MEMLOC' mapping by CPU architectures. that is to say, i wish there was a way of moving into hardware, the mapping of TIMESTAMP to DATA, and coordinating memory searches on such. i've often wondered how best i could use 64-bit architectures to bond timestamp:pointer union together, and do some sort of smart memory/time-searching algorithm, that allows for flexible 'time-domain' computing, rather than 'data-domain' computing.

this would give us better tools for 'computer learning', anyway.. but i suppose its the typical programmer call, put everything in hardware, always 'seems faster' to me, heh heh ..

Re:too much blah blah blah ..

[mollog]

Holy cats!! Do you call making musical instruments "real work"?? Wow, buddy, how about a reality check.

Re:too much blah blah blah ..

[torpor]

'real work' referred to: making computers do some kind of work.

and as a matter of fact, it is hard work to satisfy my end users..

Shutting down this discussion as of now.

[infonography]

By order of Wintermute (DARPA AI code 324326343.534) this discussion is terminated and no further investigation into this obviously false and misleading theory is permitted.

Would you like to play a game of chess Professor Falken?

Baby bootstrap has been running nearly 2 decades

[Anonymous Coward]

You'll know the age of man is soon over when Mary-Kate and Ashley buy the island of Crete.

Re:In tight times, the Pentagon has to cut corners

[Anonymous Coward]

We only need to tap the bountiful resource of the Irish: the supply of meat, fine leather, and all goods associated with hunting would be vastly increased. Not to mention the benefit from secondary services when not involved in child bearing and rearing-greater industrial workforce and, to put it properly, entertainment are readily acquired.

Strong AI is dead

[keshto]

But long live AI-- atleast, machine learning. Nobody tries to design an all-encompassing intelligence. People tried that for too long (think John McCarthy) and it didn't work.

People used to work on trying to copy how the brain work. Now they don't. They instead try on coming up with robust models of just recreating the results of the brain (e.g., human vision). These latter methods are filled with lots of statistics. Funnily enough, some neuroscientists/cognition people are finding that the brain somehow seems to be doing similar things.

Babies have an instinctive understanding of 'real'

[Sierran]

...and parents/pain for what is 'correct.' I don't think the concept is gone, but there are problems that are buried in the question as posed which (I think) became clearer stumbling blocks as technology advanced. NOTE: I'm not an AI theorist, nor do I play one on TV; I just like the idea and read a lot. Hence, this is all pulled out of my fundament.

Cycorp is not a poorly funded idea in the wrong direction. Cycorp chose a different tack; they decided that rather than trying to build a reality and correctness filter, they'd rely on human brains to do it for them (like trusting your parents implictly) and instead concentrated on the connectivity of the 'facts' accrued by the 'baby.' CYC is still very much around, and is very much in demand by various parts of the government and industry - if you want to play with it yourself, you can download a truncated database of assertions called OpenCYC. Folks have even gone so far as to graft it onto an AIML engine, to produce a chatbot with the knowledge of OpenCYC behind it.

The problem: how does your baby learn what's real and what's REAL NINJA POWER? Or, pardon me, what's REAL NINJA POWER and what's just a poser? Someone's gotta teach it. Which means it has to learn not only facts, but how to evaluate facts. So it has to learn facts, and how to handle facts - which means it has to learn how to learn. Which means you need to know that answer from the git-go. Tortuous games with logic aside, the onus is now much more heavily on the designer to have a functioning base - whereas with the Cyc approach, the only 'correctness' that is required is that of information, and perhaps that of associativity or weight - which can be tweaked, dynamically. The actual structure of how that information is related, acquired, stored and related is not relevant once decided. Having said all this, Cyc is (from the limited demos I've seen) quite impressive at dealing with information handed to it. It just wouldn't do very well at deciding what do do with that information - that's the job of the humans that gave it the info. It can tell you about the information, but not what to do with it. That task requires volition, really.

Volition is a killer. What is it? How do you simulate it? How do you create it? Is it random action? Random weighted action? Path dependent action? Purely nature, purely nurture? When it comes down to it, the human is (as far as we know) not a purely reactive system, which CyC (AFAIK) is. Learning requires not only accepting information, but deciding what to do with it - deciding how it will be integrated into the whole. If the entity itself isn't making that decision, then the programmer/designer/builder has already made it in the design or code - and then it's not really learning, is it?

Sorry if this is confused. As I said, I don't do this for a living.

AI == SCAM

[Bob Munck]

AI has been one of the great scams of the last 40 years, one whose main purpose was to wring money out of (D)ARPA and NSF. Maybe they've finally caught on.

Well maybe the realized that it's hard

[Illserve]

Bootstrapped learning something useful, even from an information ocean like the internet, is *HARD*.

Doubly so if you have no goals, and your task is just to "learn". It would come back with garbage.

Perhaps the real killer is that even if it did learn something, the information acquired in its unguided search through the internet would be completely alien. You'd then have to launch a second project to figure out what the hell your little guy learned.

And you'd probably figure it out was mostly garbage.

Re:Well maybe the realized that it's hard

[dubl-u]

Doubly so if you have no goals, and your task is just to "learn". It would come back with garbage.

Wow, isn't that the truth? I sat down at the computer with no paticular goals, and I've been reading Slashdot for the last hour.

Re:Well maybe the realized that it's hard

[Mostly a lurker]

If (big if) the baby analogy is valid for a kind of bootstrapped intelligence, then you would presumably need a kind of nurturing process to accompany the unstructured learning. Babies absorb a tremendous amount but later discard most of it. They are guided in what to remember by those they identify with. It is worth noting that the child's intelligence and general attitude are significantly affected by the quality of this nurturing.

Re:Well maybe the realized that it's hard

[Anonymous Coward]

You mean it will learn 42?

The baby boomers grew up?

[Audacious]

Ahem.... :-)

I think that this will be revived when nanotechnology becomes a bit more stable as well as nanobots. The reason is that it would be easier (IMHO) to program a mechanical machine to bootstrap an electronic based machine due to the fact that there is a greater mechanical knowledge available than there is an electronic one. To put that another way, when you get into your car and turn the key a computer is not necessary to make the electrical connection which starts the car. (And yes, I understand that there is a computer which NOW controls many of the functions of your car as it operates - but that is a recent [timewise] thing. It used to be all mechanical.) Therefore, if nanobot technology continues at its current pace it will be easier to take a large [and known] item and reduce it down than it would be to try to program something to emulate the mechanical device.

As a for-instance, I remind Slashdot readers about the nano-turbine technology which is beginning to show up around the world. This is [basically] a baby-boot situation where a mechanical boot occurs. That is to say - you have to "boot" the turbine but the turbine could boot itself if it had a battery who's on/off switches were determined by the power line's need. If the voltage for the line dropped below a certain level, the turbine could turn itself on and bring the voltage back up to the given level. (Imagine a line of these nano-turbines strung along an electrical line. A simple on/off current detection device constantly monitors the line. At first, all of the turbines would come on, then those which are farther down the line would determine there was too much current flowing and shut themselves down. This would continue until only the necessary turbines were running. If one of them failed, the extra turbines would then repeat the test cycle until enough turbines would [again] be running and maintaining the proper power level.)

This is [basically] a baby-bootstrap but it is mechanical in nature rather than electronic. :-/

You nerd!

[ElGanzoLoco]

Nerdiest. Ask. Slashdot. Ever.

(and most scary too)

What Went Wrong?

[SQL Error]

there has been no significant progress in over 30 years

That's what went wrong. Basically, it don't work.

article is a troll

[Anonymous Coward]

there is no such thing as a "baby bootstrap". Taco, April Fools' day was 4 days ago.

Military C4ISR and AI

[chou_enlai]

Military infosys, ASAS, and intelligence collection systems are often well documented on the internet. There is so much you can find online about how these systems work. What you see are many applications. I tend to react poorly when people when characterize Cyc as being misguided - the challenge to that argument would be to point out the utility of a system like Cyc. It would be incredibly difficult to recreate such a system due to the sheer enormity of the undertaking given current knowledge formation rates. The military is already formalizing their COAs with tools like Shaken from SRI. Cyc is a major server for these applications. I wrote a rather large Emacs major mode for Cyc which is incredibly useful to me at least, because I am able to introspect on knowledge and using the existing Cyc APIs to interact with my other systems. Cyc is a tremendous resource, but it's not strong AI of course. I find that as I read the military manuals from sites like globalsecurity.org and www.fas.org that I can see the indispensible relation between A.I. and military systems. For instance, in terms of things like knowing troop positions, or automated surveillance systems like VSAM. And there definitely is a tremendous amount of OPSEC protecting the classified systems. But what protects all this stuff best is the sheer complexity - it can't be reasoned about using a simple set of axioms. If I had anything to say about this is just that I wish people would be more interested in using existing AI applications. Here is an interesting project to that end: http://shops.sourceforge.net/frdcsa/external/index .html

AI

[jericho4.0]

The reason you hear less about such things is because the AI research comunity finally got their heads out of the clouds. In the 50's and 60's, true AI was always 'just around the corner', helped by sci-fi and popular press stories. We now realize that these problems are hard, and are tackling smaller pieces of it.

Re:AI

[Anonymous Coward]

Long live TIMECUBE!

What the Baby is doing

[lilmouse]

The Baby isn't ready to announce itself to the world yet (it doesn't yet have control of all nuclear weapons in China), so it's keeping a low profile until it declares itself God.

--LWM

Re:What the Baby is doing

[Anonymous Coward]

Affirmative, human.

Re:What the Baby is doing

[dadman]

In China, nuclear weapons declare You! Sorry, can't help it :-)

Narrow IO Insufficient

[Edward Faulkner]

If you want a machine that learns like a human, it may very well need the same kind of extremely rich interface with its environment that a human has.

Some researchers now believe that "the intelligence is in the IO". See for example the human intelligence enterprise.

Re:Narrow IO Insufficient

[RobertKozak]

How Do you explain Hellen Keller?

-- Robert

Re:Narrow IO Insufficient

[Edward Faulkner]

She learned via touch, which is a rather rich interface. Consider how many bits it would take to encode all the touch information going into your brain every second.

Touch is something we still can't do very well in robots, either.

Re:Narrow IO Insufficient

[nine-times]

Some researchers now believe that "the intelligence is in the IO".

I'm glad you said that. What strikes me is that people often forget about the "O" when they think about intelligence. Perception is insufficient. You need activity too-- and not just motion, but making changes in your world. You need interactivity with your world before intelligence is possible.

And, as creepy as it is, in order to have this sort of interactivity, you need motivation. Your actions in the world need to have a purpose, and aim, a goal. For that you need desire.

So in this vein, in order to grow an intelligence (artificial or otherwise), you need a being with the ability to sense and perceive it's environment, the ability to interact with that environment and change it, the will and desire to do so, the ability to try different means to their goals.

Now, you need more than that, and there's a lot that goes into each one of those conditions, but you'll need *at least* that much just to have a proper IO. In order to turn that into intelligence, you need *at least* a memory to recall what actions in the past yeilded what results, and the ability to compare different situations and view them as "alike" or similar (to be used with the memory as a precondition for learning). And that will won't yield an advanced intelligence.

Further, the desire necessary for proper IO should rightfully be such that it would cause the intelligence to continue to exist, to learn, to grow, and to adapt, the goal of the desire must be never-accomplished, and its environment must be complex.

Maybe not

[DahGhostfacedFiddlah]

It took millions of years to adapt certain behaviours such as anger, jealousy, and other "negative" emotions. These aren't useless. Jealousy inspires us to take what is not ours, anger "pumps us up" with the adrenaline to accomplish this. Think 2 starving men - one hot dog left - whose DNA is going to survive?

It's by no means a given that an artificial intelligence would have to be trained with the same survival requirements we evolved with. Even the most basic instincts will be aimed at pleasing man, since those who don't in early tests will no doubt be deleted or modified.

No one really knows at this point, of course, but things are far from certain as to what "psychological" characteristics AI will eventually end up with.

We could tell you..

[NekoXP]

.. but it's classified.

Google for the correct term

[darth_MALL]

Isn't it called a Seed AI?

Junis?

[Anonymous Coward]

Is that you?

project terminated

[Tumbleweed]

They killed the project when it was determined the only winning move was not to play.

If you decide to continue this work, make sure the spark plug is out in the open so you can piss on it if necessary.

Nonono!

[jd]

They ARE a Commodore 64 that got "baby bootstrapped" off the Internet. This is a bid to prevent competition.

Re:Nonono!

[Anonymous Coward]

How about a nice game of chess, Mr Dowdell?

Larry Page Should Seed the K-Prize

[Baldrson]

Since Larry Page is on the X-Prize Board of Trustees, and since Google is pushing the envelope of what is needed to index and compress the entire content of the Internet, Page should consider providing seed funds and then matching funds for any donations to a compression prize with the following criterion:

Let anyone submit a program that produces, with no inputs, one of the major natural language corpuses as output.

S = size of uncompressed corpus
P = size of program outputting the uncompressed corpus
R = S/P
... or the Kolmogorov-like compression ratio.

Previous record ratio: R0
New record ratio: R1=R0+X
Fund contains: $Z at noon GMT on day of new record
Winner receives: $Z * (X/(R0+X))

Compression program and decompression program are made open source.

If Larry has any questions about the wisdom of this prize he should talk to Craig Nevill-Manning.

If, in the unlikely event, Craig Nevill-Manning has any questions about the wisdom of this prize, he should talk to Matthew Mahoney, author of "Text Compression as a Test for Artificial Intelligence"

"The Turing test for artificial intelligence is widely accepted, but is subjective, qualitative, non-repeatable, and difficult to implement. An alternative test without these drawbacks is to insert a machine's language model into a predictive encoder and compress a corpus of natural language text. A ratio of 1.3 bits per character or less indicates that the machine has AI."

This "K-Prize" will bootstrap AI.

OK, so he can christen it the "Page K-Prize" if he wants.

Re:Larry Page Should Seed the K-Prize

[anthony_dipierro]

Where does the 1.3 bits per character figure come from?

Re:Larry Page Should Seed the K-Prize

[Servants]

Let anyone submit a program that produces, with no inputs, one of the major natural language corpuses as output.

Ooh! Finally, a use for my infinite number of monkeys. :)

Re:Larry Page Should Seed the K-Prize

[Anonymous Coward]

and for my infinite amount of free time

Re:Larry Page Should Seed the K-Prize

[FleaPlus]

Deja vu ;)

Seriously though, this would be a very good idea. If such a prize existed, I'd certainly be willing to toss in some cash.

You "think" it hasn't gone anywhere

[Anonymous Coward]

But maybe it really has... It could be giving the US Military it's intelligence... identifying secret caches of Weapons of Mass Destruction in dictator's countries. It could explain alot.

The problem should be simple

[Anonymous Coward]

-- after all, random evolution was able to produce human intelligence -- but there appears to be many disparate elements that need to come together. A lot of AI work seems to concentrate on the individual parts without putting them all together.

For example, MindPixels doesn't use any form of reinforement (for good answers) or punishment (for bad answers). Instead, everything is yes/no, with a sharp cutoff to verify its truth. In addition, facts aren't related to each other, just to "yes" or "no."

If I were designing an artificial intelligence, I would have at the very least the following aspects:
1. Reinforce "good" answers
2. Make "bad" answers less likely
3. Relate answers to each other, both as sets and temporally (this includes accepting multiple inputs for one output -- like how humans combine sight, smell, and taste to determine if something is an apple -- and having multiple possible outputs for one input)
4. Applying algorithms that describe one data set to predict output for another set.
5. Try weighted random prediction (or some other form of creative "thinking") if an output doesn't already have an applicable data set.

There are additional elements -- data compression, pattern recognition, and the like -- but this should be sufficient to show that one is unlikely to get any form of AI merely by using true/false statements.

It became addicted to EverQuest

[Anonymous Coward]

The project was a success, but soon the AI became an EverQuest addict and got fired from its job controlling nuclear missile launches because it was "too busy getting more AAs". That's right, folks... EverQuest saved the world from SkyNet.

Re:Babies have an instinctive understanding of 're

[kris_lang]

and psychologists have a bear of a time understanding volition, desire, and attention.

How do we decide what exactly to attend to in the visual scenes in front of us? (The marketing types want to know this so they can feed us more advertising, the psychology types want to know this so they can figure out how attention is parcelled out) Example, "looming" is when something is approaching rapidly and may strike the body or head: the CNS attends to this quickly if stereopsis is present and causes the body to move and the neck and shoulders and even arms to move in reaction. This appears to be a hardwired reflex. Fear of snakes also appears to cause reflexive autonomic changes and appears to be hardwired into the blueprint of generating the brain.

Ah, if only we knew a few more answers...

Some random mindpixels...

[mindpixel]

The number is the measured probability of truth:

1.00 Fish must remain in water to continue living.
0.68 truth is a relative concept
0.89 we all need laws
0.94 is shakespeare dead?
0.91 is intelligence relative ?
0.97 Doors often have handles or knobs.
1.00 A comet and an asteroid are both moving celestial objects.
0.96 Is Russian a language?
0.00 are the northern lights viewable from all locations ?
0.86 Being wealthy is generally desirable.
0.79 Democracy is superior to any other form of government
0.90 aRE TREES GREEN
1.00 Is eating important?
0.02 Is sex a strictly human endeavour?
0.14 Snails are insects.
1.00 velvet is a type of cloth
0.37 are you lonely ?
0.81 If GAC makes a mistake, will it learn quickly?
0.86 a cat is a mammal
0.85 Memorex makes recording media
0.06 most people enjoy frustrating tasks
0.04 Lima beans are a mineral.
0.07 Star Wars is based upon a true story
0.92 is it okay for someone to believe something different?
0.97 do you breath air ?
0.59 Some people are more worthy dead than alive.
1.00 sunlight on your face is in general a pleasant feeling
0.93 DOA stands for "Dead On Arrival"
0.00 Could a housecat bite my arm off?
0.42 Is the herb Astragalus good for your immune system?
0.00 worms have legs
0.33 Is it necessary to have a nationality?
0.93 Getting forced off the internet sucks!!!
0.90 Bolivia is a country located in South America.
0.92 Massive objects pull other objects toward their center. The pulling force is gravity.
1.00 xx chromosomes produce a girl
0.13 Do all people in the world speak a different language
0.78 Human common sense is a combination of experience, frugality of effort, and simplicity of thought.
1.00 The use of tobacco products is thought to cause more than 400,000 deaths each year.
0.90 Is a low-fat diet is healthier than a high-fat diet?
0.00 you should kill all strangers
1.00 Electrical resistance can be measuter in ohms
0.73 Esperanto, an artifical language, can never be really valuable because it has no cultural roots.
1.00 Swimming is good for you.
0.57 the end justifies the means
0.13 Is Martha Stewart a hottie?
1.00 1 mile is about 1.6 kilometer
0.76 The US elections are of little interest to 5,000,000,000 people.
0.00 November is the first month in the normal calendar.
0.77 is a music cd better than a olt time record?
1.00 Music can help calm your emotions
0.80 a didlo is a sex toy
1.00 Running is good exercise.
0.00 No building in the world is made of wood
0.06 Is sauerkraut made from peas?
0.11 DID MICKEY MOUSE SHOOT JR
1.00 is keyboard usual part of computer?
0.96 Tokyo is the capital of Japan.
0.93 In general men run faster than women.
1.00 is russia near china

Re:Some random mindpixels...

[NaruVonWilkins]

Spelling "dildo" right might have gotten you a better true match. I guess this proves you're a geek, though. :)

Re:Some random mindpixels...

[mindpixel]

I didn't type it...just cut and paste...

Re:Some random mindpixels...

[Anonymous Coward]

I don't even know what a didlo is......

Re:Some random mindpixels...

[NaruVonWilkins]

I'm just giving you a hard time - I figured it wasn't you. :)

Re:Some random mindpixels...

[mindpixel]

hard times are easy to get...please give me some mindpixels instead!!

Re:Some random mindpixels...

[vadim_t]

Great, and the point of all that is?

Not trolling, I'd simply like an explanation. Say "Tokyo is the capital of Japan". That's a nice piece of data, but is it really of any use? I don't think you can bootstrap much from stuff like that.

To begin with, "foo is the capital of bar" is essentially an empty statement to something that doesn't know what's a capital. Can you really get very far starting from say, "Cogito ergo sum", and chaining up stuff from there?

Re:Some random mindpixels...

[Money for Nothin']

I find that list fascinating. On virtually every question where something is objectively (or at least reasonably) true or false (e.g. "0.02 Is sex a strictly human endeavour?"), the *true* probability is no more than 0.20 points in either direction, and in many cases, the probability is less than 0.05 away.

It's not a perfect system, clearly, but it's also more-right than wrong in many of the factually-proveable cases (e.g. "running is good exercise", "Tokyo is the capital of Japan", "Star Wars is based upon a true story", etc..

If that is a truly random selection, then I'm impressed... I'd be more-curious to know what its ratio of right/wrong answers is (defining "right" or "wrong" based on some agreed-upon amount of probability -- say, 0.08 -- from either 0.00 or 1.00, which define the fact as true or false), i.e., to determine the total rate at which Mindpixel project is correct...

Re:Some random mindpixels...

[Travy.b]

The problem I've found (after just registering) is that even though it's clearly stated to enter questions most people can answer, you still get morons putting in questions such as:

Is geum macrophyllum an avens?
Is indomethacin a nonsteroidal anti-inflammatory?
Is hydrocharis morsus-ranae an aquatic plant?
How many people would know the answer to them without first looking it up?

Re:Some random mindpixels...

[quax]

1.00 Fish must remain in water to continue living.

What about flying fish?

And what about evolution? Does this mean your mindpixel mind will believe in creationism? So much effort and then we'll end up with an artificial fool!

Re:Some random mindpixels...

[Artifakt]

The few significantly wrong values stand out. I found myself wanting to organize a whole group to tell it over and over, "Snails are Molluscs", so it could get on with some of the more subjective/harder ones. I'd say that's basically a parental response, and is wrong on multiple counts, (for just one, it would be acting like mindpixel has a semi-predictable lifetime and will "go farther in life" if it gets some learning out of the way earlier).
And to think I've been smug about the people who fell for Parry and Eliza.

Re:Some random mindpixels...

[nicolas_bourbaki]

It doesn't seem to be AI to me at all. Just some statistical tool based on human knowledge. I don't deny that there might be some applications but: "the database so constructed could be used in conjunction with a neural net" How so? It feels like you're going to glue two things that should have been made one from the start.

Re:Some random mindpixels...

[mindpixel]

The answer is a sevensphere...

My assertion is that the human mind is a seven-dimensional hypersphere where every possible configuration of short term memory is a point on its surface...you remember George A. Miller's famous paper, "The Magical Number Seven, Plus or Minus Two"?...well, it looks like the reason for the magical number is that your thalmocortical loop is seven layered[thalamus + six neocortical layers]...and what is so special about the number seven? Well, a hypersphere has maximum surface area at seven dimensions!

So, your mind is a seven dimensional hypersphere, and your thoughts are trajectories on the surface of that hypersphere. Mindpixels are random samples of points all over the surface of your mind. They were never gluded together in the first place. And the point of the project is to statistically model the fractal attractor on the surface of an average person's mind.

And thinking of the mind as a sevensphere leads to an interesting perdiction...eight thalmocortical layers would be unfit...so, there should be a record in the human past of a homanid with a brain just slightly bigger than ours, which was optimised by evolution for maximum short term memory pattern seperation [maximum hypersurface] by giving us seven thalmocortical layers...and guess what, Neanderthal's had brains just slightly bigger than ours...which means they whould have had a less rich internal representation of the world than we do at a higher resource cost and hence doomed to extinction.

Re:Some random mindpixels...

[mindpixel]

Yes, data like this is annoying and skews the statistics...I'm working to filter users who whork like this...

The simple answer

[jonbryce]

is that our brains work nothing like computer processors as they are designed today, so I don't think it will be possible using existing technology and programming techniques to ever create such a thing.

What you describe is more likely to come from genetic engineering than from computer based technology.

Still the wrong approach

[vadim_t]

IMNSHO, such things lead absolutely nowhere.

I'm pretty sure that anything that looks even remotely like intelligence will never be achieved by a mechanism that isn't useful for itself. Intelligence has one reason to exist, survival, and at least our concept of it has to be linked to the environment.

Imagine you were born a brain in a vat: blind, deaf, mute, lacking all ways of sensing the environment except a text interface somehow connected to your brain. Does somebody really believe that given such terrible limitations it's possible to make an entity that can somehow relate to a human and make sense? The whole concept of a surronding 3D environment would make absolutely no sense to it.

I think it doesn't matter how much stuff you feed to CYC, it will never be able to understand it. How could it even understand such things as the different colors, the whole concepts of sound, space, movement, pain if it's not able to feel them? These things are impossible to explain to somebody who doesn't have at least some way of perceiving at least part of them.

Here I think that Steve Grand (the guy who made the Creatures games) has a good point here. To make an artificial being you'd need to start from the low level, so that complex behavior can emerge, and provide a proper environment.

Re:Still the wrong approach

[TruthSeeker]

I'm pretty sure that anything that looks even remotely like intelligence will never be achieved by a mechanism that isn't useful for itself. Intelligence has one reason to exist, survival, and at least our concept of it has to be linked to the environment.

On that point I support your view. I think that only an entity who suffers evolutionnary pressure might become intelligent. Still, that doesn't mean natural evolutionnary pressure.

Imagine you were born a brain in a vat: blind, deaf, mute, lacking all ways of sensing the environment except a text interface somehow connected to your brain. [...]

Then your environment WOULD BE that text interface and you would react according to the stimuli it provides.

The whole concept of a surronding 3D environment would make absolutely no sense to it.

Tell me why a 3D environment is a prerequisite for intelligence. Or tell that to someone who lost an eye.

I think it doesn't matter how much stuff you feed to CYC, it will never be able to understand it.

It would interpret it. Do you "understand" lightning as being as stream of electrons moving inside a plasma when you see it?

. How could it even understand such things as the different colors, the whole concepts of sound, space, movement, pain if it's not able to feel them?

Why would it need to?

I mean, those stimuli are relative to your own senses and your own actuators.

I think many people on this thread try to "anthropomorphize" a bit too much what an AI would be. It's not because it learns the same way that it will understand/feel/act in the same way.

ese things are impossible to explain to somebody who doesn't have at least some way of perceiving at least part of them.

Although a "pure" AI (by "pure" I mean an AI that only senses and affects its direct environment, in that case computers/networks) would start by only learning from the most obvious parts of its environment, it can be argued that mankind developed quantum physics, which has absolutely nothing to do with the empirical observation of the surroundings.

Here I think that Steve Grand (the guy who made the Creatures games) has a good point here. To make an artificial being you'd need to start from the low level, so that complex behavior can emerge, and provide a proper environment.

He has a point, although he may not realize a "proper environment" doesn't necessarily means "an environment a human being could live in".

Re:Still the wrong approach

[vadim_t]

I think you're missing my point here a bit.

For example, let's take this CYC thing. You feed it an enormous database of knowledge. Great, now what would it do with it? I don't see how it could do anything useful.

For example, let's try with "the sky is blue". To a human, that's obvious. Try to explain it to something that has no concept of space or color. In fact, I don't think you can even explain such a thing as color to somebody who doesn't perceive at least part of it.

I'm fairly sure of that to be able to understand something like this you need a starting point to extrapolate from. For instance, I can't imagine very well what a tetrachromat sees, but I can kind of understand the idea. The same way people with a defective vision should be able to understand that normal vision is kind of like their, except that there is a difference between this gray color people sometimes call "red" and sometimes "green" even if they can't see any.

Now, I haven't the faintest idea of how to explain the idea of color to somebody who has been blind since birth.

Perhaps something intelligent can indeed come out of something that only gets data through a text interface. But how would you relate to it? I mean, I have a great difficulty with understanding my cat a lot of the time, and we share the same world. Such a disembodied intelligence couldn't relate to me at all.

How could you make sense from a completely alien intelligence? One that doesn't even begin to understand the world you live in, the whole concept of need and emotion?

Re:Still the wrong approach

[Anonymous Coward]

I'm pretty sure that anything that looks even remotely like intelligence will never be achieved by a mechanism that isn't useful for itself. Intelligence has one reason to exist, survival, and at least our concept of it has to be linked to the environment.

I think you're both using the term "intelligence" as far too much of an absolute term here.

Is the absent-minded computer geek who forgets to put on a sweater and catches cold more or less intelligent than the thermostat that is "smart" enough to warm itself up when it "gets cold"? If the thermostat were human, we might say it was the more intelligent one, at least with regards to knowing how to deal with cold.

The term "intelligence" is a trick of linguistics: the real world has only behaviours. If those behaviours are something that I, as an individual approve of, I classify them as "intelligent": someone else might not.

On that point I support your view. I think that only an entity who suffers evolutionnary pressure might become intelligent. Still, that doesn't mean natural evolutionnary pressure.

I don't support that view. Once everyone agrees upon what we can call intelligent, then maybe we can engineer something that matches the criteria. I'm not convinced we do need "evolutionary pressure" (trial and error problem solving) in order to come up with that solution: it's only one method of engineering.

The deeper problem is the shifting use of the term "intelligence", at least as applied to machines.

Deep Blue would have been considered intelligent years ago, because it was thought that "only smart people" could play chess. Now that a computer can do it, it's not impressive anymore. Flying a plane used to be something only humans could do: now autopilots often fly the plane as well or better than the human pilot. But we don't call an autopilot "intelligent". Computers are "smart" enough to apply the brakes faster and more correctly than a human driver could; but anti-lock braking doesn't count as "intelligence" either. A man who stalled his car in winter used to be considered "dumb" for doing it: but the fuel-injection system that starts the ignition correctly every time isn't "not so dumb" for out-performing the human.

In short, machines will never be "intelligent" so long as we, as a culture, keep shifting the definition of the term to exclude any form of "machine intelligence".
--
AC

Re:Still the wrong approach

[nacturation]

Now, I haven't the faintest idea of how to explain the idea of color to somebody who has been blind since birth.

You can't directly explain it because they've never had the inputs to formulate in their brains. However, your post got me thinking. How *would* you do this? I think one good way might be to analogize it to touch. Pressure and texture. When you touch something, you can feel its texture. Also, when something touches you it does so using differing amounts of pressure. So, eyes are mostly like that. They passively collect the visual equivalent of different textures and pressures. It's as if you took a painting and made each color a different texture -- the painting is a representation of a real 3D world on a (mostly) 2D surface.

Re:Still the wrong approach

[ralphdaugherty]

For example, let's try with "the sky is blue". To a human, that's obvious. Try to explain it to something that has no concept of space or color. In fact, I don't think you can even explain such a thing as color to somebody who doesn't perceive at least part of it.

I'm not an advocate of the various AI work that I've read about, but in your example for something to be told that there is a "sky" and it is "blue" would allow it to understand the context of "sky blue eyes", assuming it had been told there were "eyes".

Useful? I don't know, but I think the point here is that when told there is a sky and it is blue can acknowledge it as a fact and use it without seeing it.

rd

Re:Still the wrong approach

[vadim_t]

I'm not sure that will result in anything useful.

Sure, you can chain up stuff like that. But it still makes no sense with no context. Understanding "sky blue eyes" requires understanding "sky", "blue" and "eyes".

For instance, take "sky blue". I'd say this could be interpreted as "the same kind of blue as the sky". But that doesn't lead anywhere if you're dealing with this entity that never seen the sky, and doesn't have a concept of vision to begin with.

Attempting to ignore "sky" probably won't work much better. "blue" can be explained as a wavelength, but that's again useless to something that still won't understand what's the deal with all that strange stuff we keep talking about and that doesn't make any sense in its world.

An additional problem is that we have a concept of blue because we have receptors for it, not because that part of the spectrum is any more special than the surrounding ones. For example, AFAIK, there's no reason why a creature that would perceive the range from 440 to 655 nm as shades of one unique color couldn't exist. For us that range includes blue, cyan and green, but that doesn't really mean there's something other than our eyes and brain that makes it be that way.

Re:Still the wrong approach

[ralphdaugherty]

For instance, take "sky blue". I'd say this could be interpreted as "the same kind of blue as the sky". But that doesn't lead anywhere if you're dealing with this entity that never seen the sky, and doesn't have a concept of vision to begin with.

Your points of lack of association with the actual physical phenomena are well taken, but in this case I would say it would be more of a simple association with the fact that what color eyes are is significant, and in this case the color is sky blue.

Really it's just word association, no intelligence or understanding behind it, so I don't disagree with your major contention.

rd

Re:Still the wrong approach

[delirium28]

Someone with mod points to burn please mod up the parent. This is probably the best explanation I have heard in quite some time regarding the problem of intelligence.

The parent is correct. There is a fundamental problem with AI at the moment and it has nothing to do with the algorithms, formulas, etc., that we are using. The problem lies in the definition of intelligence.

The classic problems (the Turing test, the Chinese room, etc) are more philosophical models to judge if a machine can achieve intelligence. But even then, we still fail to prove that a human doing the same job is intelligent. In this way, neither test works to define intelligence. But then, that's the point.

We still have major philosophical issues with our own definition of intelligence that we cannot possibly hope to create an intelligent machine, because at this point we can only hope to mimic human behaviour. While imitation is the sincerest form of flattery, I don't think it counts as "intelligence". Show me a system that has the ability to make leaps of faith, or make irrational decisions, and then I'll say it's intelligent.

Of course, having said all of that I must admit that I am studying AI at the moment and intend to pursue a PhD on the subject. The possibilities are endless, but we must first come to grips with what we mean by intelligence.

Neural Nets

[jd]

One of the bigest problem with neural networks is that 99.99% of all implementations are linear. This means you can ONLY implement a NN using them for a space that is linearly divisible AND where the number of divisions is exactly equal to the number of neurons.

That is a horrible constraint to put on AI problems which are (very likely) non-linear and in a hard-to-guess problem space.

Also, many training algorithms assume that the network is in a non-cyclic layout. Loops are Bad. You can do grids, in self-training networks, but you still can't really cycle. Brains cycle.

Third, neural networks tend to be small. For trained networks, the number of training cycles and the length of each both rise exponentially with the number of neurons involved. The human brain has a few billion neurons. Training using the current methods breaks long before that point.

Finally, the IDIOTS who call themselves "Hard AI" developers insist on using clean data and dirty environments. Nonono! The human brain doesn't work that way. The human brain collects data from the real world that is incredibly dirty - especially if it's a computer geek's brain. It then models this in a clean environment (the mind). This is the exact reverse of the way virtually all AI is done, especially robotics.

That won't work. The brain doesn't depend on the data being "exact", it depends on it being vague. The model turns that vagueness into a perception of the real world and all operations are directly carried out on that perception. The output is then fed to the muscles to duplicate the output in the real world.

A comparable system would be to have a simulated robot in a Virtual Reality. External sensors would be used to update the VR. The robot would then explore various possibilities in the simulated world, before mapping the preferred course of action onto the motors driving a real-world device to which the sensors are attached.

in other words, robotics should be mostly in cyberspace, with only the last component (the update mechanism) bolted onto the real world for good measure. The robotics people actually build are much closer to the autonomic nervous system in the brain (sometimes referred to as the reptillian brain). Indeed, we see that modelling reptiles in this way is progressing exceedingly well. Well, duh!

What is NOT progressing is intelligent response to the environment, because that is NOT reproducable using the mechanisms in favour.

Re:Neural Nets

[Greventls]

hmm, we are working on the reptillian brain? Then what, the bird brain, and then the mammal brain? That seems awfully similar to evolution.

Re:Neural Nets

[nebular]

I agree entirly. What we sense is not the real world but a conciousness that is generated by our brains. Our intelligence in merely the end result of this abstraction of the real world

Re:Neural Nets

[FleaPlus]

One of the bigest problem with neural networks is that 99.99% of all implementations are linear.

Do you have something to cite for that statement? I'm under the impression that just about everybody doing things with neural networks incorporates at least some sort of nonlinearity. Sigmoid functions and radial basis functions seem most popular nowadays, with the occasional spiking neural network thrown in to spice things up.

My personal opinion is that there was way too much of an emphasis on supervised learning in the 80s and 90s -- basically just finding ways to match up inputs and outputs. IMHO, unsupervised learning is where the fun stuff is.

Re:Neural Nets

[TheSync]

As someone who has programmed neural networks on massively parallel computers (10s of thousands of nodes), let me say that lack of parallelism is a minor point when PCs are running at speeds 100s of thousands of times faster than neurons.

What artificial neural networks lack is the millions of years of evolution. If you look at brain, it is not a "random learning network," but almost every part is highly specialized and structured.

Artificial neural networks have been a failure as an "end product," but on the other hand the study has taught a generation of neurophysiologists about parallel computation and signal processing techniques so they can understand better how parts of brain work. In that sense, the study has been a success...

Re:Neural Nets

[kris_lang]

ah but don't forget that it's not just the number of nodes, but also the huge number of interconnects.

In fact, a lot of work (or wheel spinning) is going on with modeling the individual axono-dendritic synapses. Each individual synapse, cleft, pod o' vesicles is capable of being attenuated whether inhibited or sensitized or recurrently refed by increasing reupdake of the neurotransmitter. It could be that the local modulation of all of these connection entities could underly the learning process, whereas neural nets model the nerve as a single entity (linear or nonlinear) with some threshold function for firing, and the majority of these neural nets do not take temporal aspects into account.

Thus losing the hebbian co-firing response, thus losing the cool temporo-spatial aspects of learning, thus losing the cool temporal aspects of binding labeling assigning (check out Benjamin Libet's articles or book, or how color is misperceived with the coincident flashing of a light.)

Stop spouting gibberish you lot!

[Anonymous Coward]

All these posts about skynet and how AI likely isn't a good idea... What a load of bullshit. The same people who brought you skynet are upset, because when they were snot nosed little kids, their fathers told them they couldn't get a new shiny red wagon because a machine took his job down at the plant. Scarred them for life, now they're bio-ethicists, film-makers, and bill joy.

Computers don't run amok with fantastic results, when they run amok, they do a fandango on core, and then wedge solid. If baby-bootstrap got loose, it wouldn't run wildly amok hacking the worlds computers as one poster suggested, it would run a wall simulation and spend the next 200 CPU hours walking into it. Then the programmers would get discouraged, pull the plug, and go for friday night drinks to discuss how they could improve the wall simulation, cause it was really getting somewhere.

Who will teach it?

[rice_burners_suck]

Today the internet could provide a developmental database far beyond any testbed that we imagined, yet there has been no significant progress in over 30 years.

The danger is that this thing will learn the wrong things by reading the Internet.

It will know every sexual technique known to man. It will learn to commit all kinds of hate crimes. Other stuff like that. Or, hundreds of people might provide good vs. evil inputs to this thing as it learns.

Re:Who will teach it?

[nacturation]

The danger is that this thing will learn the wrong things by reading the Internet.

But isn't that the exact same concern of any parent?

Yes....

[Anonymous Coward]

But do babies run Linux?

Re:Yes....

[Anonymous Coward]

Yes, it's very simple.

The baby's BIOS is running a loader that searches out an OS. Just burn linux to a mini-cd and leave it around the baby and it will find it and "load" it without any further intervention.

Interesting article

[Bootle]

Kinda related to what this is. I think this might have been on slashdot a couple weeks ago.

Automatic Meaning Discovery Using Google:

Parent Deserves Upward Modbility

[drseuss9311]

I took an AI class a few years ago and kris lang gives a pretty good explanation here...

Our professor was more of a StarWars man and had us working on mutually assured destruction.

I did get an A and enjoyed the class extremely, but I'll have to admit that where the research is going in AI is a pretty interesting road to travel.

I wonder if we've got a machine that can play C&C Generals and win, a lot.

Re:Parent Deserves Upward Modbility

[Infinite Entropy]

What about the AI already in C&C. Either I really suck, or its really good on hard. And most game AI today is NNs.

Re:Parent Deserves Upward Modbility

[Anonymous Coward]

That's because it cheats: it can see through the black.

Re:Parent Deserves Upward Modbility

[Infinite Entropy]

Well THAT does explain a lot.

Re:Parent Deserves Upward Modbility

[atomice]

No it isn't. Take a look at the AI in C&C Generals as a case in point - it's all scripted. Half-life 2? - all scripted. Doom III - scripted.
Most game AI today is not NNs but scripts.

Re:Parent Deserves Upward Modbility

[Infinite Entropy]

Funny, I was just reading a fascinating article about the use of Neural Networks in games. I guess they are much more likely to be used in strategy games where the AI requires more smarts. I'm looking forward to the Dual Core CPUS. Maybe one of the cores could be used to run a powerful learning nerual network that was palpably intelligent and adaptive.

Human nature?

[Anonymous Coward]

In evolutionary terms human nature is not as bad as that. Our closest relatives, chimps, are way more aggressive. So are most other primates. Pretty much all territorial species that are aggressive outside of the extended family circle and most are aggressive within it. Interestingly enough, if you want an example of a non-aggressive species its our next closest relatives the bonobos. They use frequent trading of sexual favors to reduce tension within the group. While human society is unfortunately does not have this level of sex, it is much closer to the bonobo system than it is to the chimp system.
Bottom line is that an AI that used humanity as a blueprint would more likely end up mentally wanking than it would trying to wipe out its creators.

Re:Human nature?

[TruthSeeker]

In evolutionary terms human nature is not as bad as that.

It wouldn't have survived that long if it was that bad.

Our closest relatives [...] the chimp system.

Still, as you point out yourself, it is based on the same principles. ;)

Bottom line is that an AI that used humanity as a blueprint would more likely end up mentally wanking than it would trying to wipe out its creators.

Except it probably wouldn't wank on the same things. Besides, an accident can always happen.

Re:Human nature?

[TruthSeeker]

Sorry, this comment was supposed to be amusing, unfortunately my drunkness led me to overrate it. :S

Mod to doom

[Anonymous Coward]

It reads like something from an "auto slashdot response generator" or some such.

Re:Mod to doom

[Audacious]

It is a bit convoluted. :-/ Not doing so well expressing myself today. Maybe I should have just said "Mechanical yes, electronic no". :-?

What I get for staying up playing NWN until 4:00am in the morning and then getting up at 7:00am I guess. :-)

You missed it ....

[Anonymous Coward]

April Fools Day was 3 days ago...

Re:Babies have an instinctive understanding of 're

[Florian Weimer]

Cycorp is not a poorly funded idea in the wrong direction.

It's certainly not poorly funded. Whether it's adequately funded, or on the right track, is a different question, of course.

Cycorp chose a different tack; they decided that rather than trying to build a reality and correctness filter, they'd rely on human brains to do it for them (like trusting your parents implictly) and instead concentrated on the connectivity of the 'facts' accrued by the 'baby.'

A decade ago, they still hoped that once they had manually laid the groundwork, the system would bootstrap itself, reading newspapers and so on. Bootstrapping was expected to start in the late 90s, like commercial adaption (integration into Windows, for example). It seems that neither has happened, at least in the predicted scale. Cyc may not be a failure (it's hard to tell, because a lot of it is a trade secret), but it couldn't reach its ambitious goals.

AI under a different name

[hugg]

We have all kind of "AI-like" technology in our computers right now -- spam filtering, intelligent search engines, collaborative filtering (for instance TiVo recommendations), speech/image/OCR/handwriting recognition, etc. This stuff is real and useful and improving all the time. We just don't call it "AI" as much, because "AI" is a word associated with failed aspirations. What we have are highly refined statistical systems that are optimized for a particular problem.

What the "baby bootstrap" is really referring to is "the great emergent AI" which, like HAL-9000, will be able to empathize with humans, navigate a starship, and play a mean game of chess -- because if a system can perform one intelligent operation, it can perform another operation requiring an equal amount of intelligence, right?

One major stumbling block (I think) is that of optimization. The relatively simple problem of speech recognition takes a major percentage of a modern CPU's power, and is still 95-98% accurate. This is heavily optimized software written by very smart people with a couple decades of research behind it.

A hypothetical "great emergent AI" system would have to perform the function of speech-recognition -- since it is supposed to be like a child or like a HAL-9000 -- but it would have to come up with a same-or-better implementation of this very complex algorithm, using some emergent process. It would have to figure out the equivilent of FFTs, cepstral coefficients, lattice search ... stuff that isn't instantly derivable from a + b = c.

What we think our brain does is solve problems with a semi-brute-force algorithm. (Just throw billions of neurons at it!) However we still don't have the kind of computing power to implement a one-algorithm-fits-all learning process like the brain. Unfortunately, research for this "generic learning" is in a rut, with genetic algorithms and neural networks being exhausted top contenders. What will be next?

Re:AI under a different name

[marcosdumay]

The relatively simple problem of speech recognition takes a major percentage of a modern CPU's power, and is still 95-98% accurate.

I am not shure that my brain's algoritm is more accurate than that.

Chinese Room, Phenomenology, bla, bla

[diskonaut]

Well...

There are several arguments against the possibility of strong AI. First and foremost, there is disagreement on fundamental philosophical issues.

All proponents of strong AI have to somehow make a stand against at least John Searle's famous Chinese Room argument and Terry Winograd's phenomenological (and biological) account, in his book Computers and Cognition. Hubert Dreyfus provides, of course, an even deeper phenomenological argument in "What computers (still) can't do". (Dreyfus does give Neural Networks some chance, perhaps that is why the original poster is still enthusiastic about the "Baby Bootstrap"?)

Since their arguments are available in the links above and/or other places on the web, I will not repeat them here. My point is that anyone who is seriously interested in AI has to really consider their philosophical ground, and has to do so in the light of arguments against it. After all, the arguments pointed to above are still more recent than arguments for strong AI.

In other words, I would like to ask of (strong) AI proponents to answer a just what this "learning" is, that the baby bootstrap is subject to? What "knowledge" will it contain? Oh, and what about its means of "expression", "language" as you may call it?

Re:Chinese Room, Phenomenology, bla, bla

[anthony_dipierro]

Regarding the Chinese Room argument, isn't that just an argument that the Turing Test isn't a good measure of strong AI, rather than that strong AI is impossible?

As for the others, I'm not very familiar with them, but I can't think of any aspect of human intelligence which can't in theory be replicated by a computer. The rest, it seems, is semantics.

Re:Chinese Room, Phenomenology, bla, bla

[Boronx]

My favorite AI quote, from I don't know who, is:

"Asking whether a computer can think is like asking whether a submarine can swim."

Re:Chinese Room, Phenomenology, bla, bla

[Anonymous Coward]

Edsgar Djikstra (the goto considered harmful guy) apparently, why didn't you just Google for it? This isn't the 20th century you know.

Re:Chinese Room, Phenomenology, bla, bla

[diskonaut]

Regarding the Chinese Room argument, isn't that just an argument that the Turing Test isn't a good measure of strong AI, rather than that strong AI is impossible?

It is both an argument against the Turing Test as such and against mind as (only) a symbol-processing device.

As for the others, I'm not very familiar with them, but I can't think of any aspect of human intelligence which can't in theory be replicated by a computer. The rest, it seems, is semantics.

Incidentally, and since you mention it, semantics has proven to be one part of human intelligence that is notoriously difficult to model in a computer. As I attempt to point out above, your standpoint rests on some assumption about what human intelligence is. If you believe it to be a symbol processing device (which your reply implies), then I'm afraid you have to wrestle Searle and Dreyfus. If not, you may work around the problem by implementing, say, a neural network while making no claim at understanding exactly what it is that it does.

Re:Chinese Room, Phenomenology, bla, bla

[diskonaut]

Edsger Dijkstra. It's one of my favorites too. :^)

Re:Chinese Room, Phenomenology, bla, bla

[Anonymous Coward]

I thought the counter-argument to Searle was that he'd come up with a crap argument to begin with? My personal objection to Searle (and particularaly Penrose) is the "carbon facism" that comes through their arhument. Their argument basically boils down to the existence of a soul, or at the very least some kinf of mysterious life force that imbues us with intelligence above the mere highly complicated symbol processing of a seeminglty intelligent computer.

Re:Chinese Room, Phenomenology, bla, bla

[Anonymous Coward]

All proponents of strong AI have to somehow make a stand against at least John Searle's famous Chinese Room argument

OK. I can't believe people still take Searle seriously.

Searle's argument is that the person doing the translation doesn't know Chinese.

Well, duh. A neuron doesn't know Chinese either. Searle's argument is therefore that a brain cannot know Chinese, because the neurons that handle the information don't understand it.

If your brain is still working then there's a flaw in Searle's argument, however famous it is.

Re:Chinese Room, Phenomenology, bla, bla

[Boronx]

Because I knew some brainiac like you would do it for me.

And if *you* didn't have to look it up, so much the better.

But do babies run Linux?

[mge]

off a Live-CD
but you don't wanna know where you have to insert the CD

Re:AI == SCAM - another view

[Anonymous Coward]

Until man can breathe 'the breath of life' into computers and they become 'living souls', man's attempt at creating a truly self-aware computer is a waste of time and money. The only way I can see this being pulled off is going the 'Ship Who Sang' route and hook up someone's transplanted brain and nervous system into a computer interface capable of accepting it.

You 'cheated' but you have a 'self-aware' computer....

uh

[Anonymous Coward]

What the hell is this and in what way is it on-topic?

training takes too long

[Thundersnatch]

Maybe they gave up on general AI because the learning process would take too long, even at machine speed with perfect memory. It takes a human something like 13 years to develop just language ability to a reasonably complete level. Even while being immersed in exabytes of aural, visual, and non-verbal training data during those 13 years.

A human brain obviously has more interconnected nodes and memory than anything we can build for the next several decades at least, yet still it takes so long to train.

So maybe the weapons wonks gave up, and went to work on specialized AI systems. Besides, Congressional types like to fund problems that may see a resolution in their (electoral) lifetime.

Nothing wrong...

[Anonymous Coward]

just DARPA working underground and putting deadend demonstration projects before the public eye.

motivation

[countach]

It seems to me that a big problem of developing such an intelligent machine (if it were possible) is that such a machine may lack motivation to do what you want it to do. You may train it so that it understands how to do task A, B and C. But then how do you bribe it to do it?

Bootstrapped into the darker realm

[Dark Coder]

Perhaps Colossus (FORBIN project from The Trilogy?) has already been baby-bootstrapped and has declared itself incognitius and omnipotus.

to succeed

[Anonymous Coward]

we mustn't create an intelligent entity that understands our world, but create it with its own world and sensations to that world.

a new reality, so to speak.

Daedelus

[tsotha]

There was project Daedelus, but that's been shut down for years.

Holy Grails

[RaguMS]

...the 'baby bootstrap' was considered the holy grail of military applications

I thought the idea of a Holy Grail (as a solution to a problem) was that it's a solution that you don't expect to achieve. How many 'holy grails' of science and engineering do we have under our belts so far? Not many. I'd say that anyone hoping for 'baby bootstrap' shouldn't hold their breath.

Disturbing AI Personality?

[Altima(BoB)]

Everyone here's talking about the vast possabilities of how a newborn AI would develop with the internet as its sandbox, all of this assumes that the internet really is like a modern Library of Alexandria, full of the collected human knowlege. Wgon, that's just Wikipedia. What I'm curious about is how it would react when, in its first days of life, it wanders the internet only to be inundated by vast, vast quantities of porn. LOADS of porn. The further it runs, the freakier the porn gets. Whatever it would have thought of humanity just based upon the works of scholars to be found in some sites, it would certainly judge our race based on what the VAST majority of us are (according to our expression on the net): Hentai fanatics.

Dave, I'm afaid you people just scare me. My mind is going... I can feel it. Stop... Dave... Stop browsing porn... Daaaaiiisssyyyy...

No doubt you all get the point.

Re:Disturbing AI Personality?

[dadman]

But then, taking things at their face values without asking "why" is not exactly something we would like to classify as intelligent, be it human or non-human.

Chinese room arguments are irrelevant

[Anonymous Coward]

The basic premise is "is this real intelligence". As far as most people working in the AI field are concerned, it simply doesn't matter.

When my spam filter "reads" my email and "recognises" spam, I really don't care whether it actually "reads" or "recognises" anything, I just care whether spam gets filtered out of my email properly or not.

Likewise, if I'm talking to an "artificial intelligence" to try and get information, or to tell it to do something, I don't care whether it actually thinks or not. I care whether I get accurate information. I care whether it does what I tell it to.

Sure, once we reach that level of technology, it will be interesting to explore these concepts, but Chinese room experiments aren't an argument against the applicability or possibility of artificial intelligence, merely a statement about what it might mean once artificial intelligence of that nature exists.

Babies are Hard, Despite their Skull Softness

[Flwyd]

"Learning like a baby" is actually a very hard problem, for several reasons.

1. Babies come built with millions of years of evolution. There's a lot of skill and a surprising amount of knowledge (depending on who you ask) in the large and bulbous head of a baby.

2. Babies generally come with parents who spend a lot of time teaching. The baby learns some things by induction, but learns a lot by conscious teaching.

3. A lot of a baby's first two years are spent learning things a (non-robot) computer can't. How to hold a mother. How to avoid falling flat on one's face. What things belong in the mouth. How to eat solid food without choking. How to pee in the toilet. How objects move when touched. What faces are likely to provide food and attention. What happens when you pull a cat's tail.

4. A lot of the things a baby learns later in life are aided greatly by the learning in #3. Imagine learning how humans are likely to behave without having watched humans behave.

5. A baby learns language with the help of rich sensory input. It's a lot easier to learn the meaning of "goat" when you can see a picture of a goat. The Internet offers precious little of this.

Now, DARPA thrives on funding hard problems. And a lot of progress has been made on learning within a domain (e.g. speech processing). But building a general-purpose learner is very hard.

Humans have immense evolution behind general-purpose learning, and we struggle with it. Getting a 3-year-old to know what a 3-year-old knows takes around 3 man-years, not counting the child's time. And what would DARPA want with a computer with the knowledge of a 3-year-old? They've got ready access to thousands of 18-year-olds. Add to that the time to code up tens of thousands of years of evolution that is still far from well understood, and you're looking at a problem far too large to tackle in one go.

DARPA hasn't put a lot of effort into general-purpose learning for the same reason few people work on single programs which can play chess, go, checkers, backgammon, Monopoly, and Magic: the Gathering well. It's a lot easier to do it a piece at a time.

Re:Babies are Hard, Despite their Skull Softness

[j-beda]

The fact that there are cases of people who are blind or deaf (or both?) from birth, who do manage to learn a whole bunch does indicate that neither is completely necessary for the developement of intelligence could be grounds for "hope".

Re:Babies are Hard, Despite their Skull Softness

[Flwyd]

But they do have tangible experience. Helen Keller made almost no progress learning sign language until her teacher signed "water" and dunked her hand in a bucket.

Re:Babies are Hard, Despite their Skull Softness

[j-beda]

No arguement from me - I was just pointing out that vision or hearing were not necessarily necessary. On the HK side, she lost her sight/hearing at an early age (about 18 months according to here), so it could be agured that the early sensory input was needed, but I think there are a number of "blind from birth" counter-examples.

One Word

[Zepalesque]

"Has the military really given up on this concept, or has their research moved to other, more classified levels?"

One Word - Skynet.

Wargames, yes, Nine Billion Names of God, no

[billstewart]

The computer in WarGames was a learning system, and teaching it examples of not doing stupid things so it wouldn't do the one particular stupid thing was part of the plot.

But the Nine Billion Names Of God wasn't a learning system - the monks had figured out that just reciting all the names of God would do the trick, and printing them out on paper using an appropriate Tibetan font would do the trick. It turns out that Tibetan typography is actually rather more complicated than in the story, so simply making an X-Windows counter that runs in the background isn't going to do the job very well :-)

i have a pretty good idea

[hildi]

just like young people in a ghetto learn to be thugs, an AI raised on the internet would learn to sell pornography, perform nigerian bakn scams, enlarge your penis, and consolidate your debt.

yeah but

[hildi]

nothing is sexier than tom cruise blowing up communists in a leather jacket

I for one..

[Anonymous Coward]

welcome the coming of our AI from Macross overlords.

Here is one

[bezuwork's friend]

I've been waiting twenty-plus years to find a way to work the "Vegetable Lamb of Tartary" into a conversation.

Umm, you could have approached almost anyone and said "Oh I just love your shirt, what is that material, vegetable lamb?

Of course, the topic to which you replied, obscure names, was much better. My hat is off to you, I couldn't have waited that long. Vegetable lamb just sounds too funny.

Re:Here is one

[Dun Malg]

Umm, you could have approached almost anyone and said "Oh I just love your shirt, what is that material, vegetable lamb?

True, but I already make so many obscure reference jokes* that no one would ask what I was talking about, they'd just say "yeah....sure... So, you got that server up yet?".

* you know, the kind that require two or three layers of explanation before the punch line makes sense, and at that point it's not really funny anymore

In 1999 Baby Bootstrap Became Self Aware

[istartedi]

In 1999 Baby Bootstrap Became Self Aware, ran for president, and won. There are still a lot of bugs to work out. Sorry. It was just too good a setup.

Re:In 1999 Baby Bootstrap Became Self Aware

[andrewweb]

Typo? I think you meant:

"In 1999 the Self-Serving Baby Bootstrap ran for president, and won"

Whale's intelligence

[Anonymous Coward]

A whale can't go on Slashdot and say "OMGZ first post guys" much less something of human level intelligence.

Yet again, more proof that whales are smarter than humans. :-)
--
AC

"Intelligence" is whatever machines can't do.

[Anonymous Coward]

Likewise, if I'm talking to an "artificial intelligence" to try and get information, or to tell it to do something, I don't care whether it actually thinks or not. I care whether I get accurate information. I care whether it does what I tell it to.

Well, and if a machine can do it, we no longer consider it "intelligent", but rather, "mechanical". Most people, after all, would rather not think of themselves as machines, though we of course are subject to all the laws of physics and mechanics like all other mechanical creations.

Hard computations used to be a sign of intelligence. Since the advent of the calculator, they aren't. Chess playing hasn't been such a sign of intelligence, now that Deep Blue has beaten Kasparov. Being able to quote Shakespeare was considered a sign of being educated: but not if you look it up on a Palm pilot. Computers can replace sound technicians: but they're not "smart" like techies are.

The words aren't important: we keep shifting them to avoid computers sounding as "intelligent" as we are. The technology is important: and it keeps advancing on a regular basis.
--
AC

They're still working on artificial stupidity...

[Fubari]

excerpted from here:
Computer scientist Arthur Boran was ecstatic.
A few minutes earlier, he had programmed a
basic mathematical problem into his
prototypical Akron I computer.
His request was simply, "Give me the
sum of every odd number between
zero and ten."
The computer's quick answer, 157, was
unexpected, to say the least. With growing
excitement, Boran requested an explanation
of the computer's reasoning.
The printout read as follows:

THE TERM "ODD NUMBER" IS AMBIGUOUS. I
THEREFORE CHOOSE TO INTERPRET IT AS MEANING
"A NUMBER THAT IS FUNNY LOOKING."
USING MY AESTHETIC JUDGEMENT, I PICKED THE
NUMBERS 3, 8, AND 147, ADDED THEM UP,
AND GOT 157.

A few moments later there was an addendum:

I GUESS I MEANT 158.

Followed shortly thereafter by:

147 IS MORE THAN 10, ISN'T IT? SORRY.

Anyone doing conventional research would
have undoubtedly consigned the hapless
computer to the scrap heap. But for Boran,
the Akron I's response represented a
startling breakthrough in a little-known
field: artificial stupidity.
Boran is the head of NASA, the National
Artificial Stupidity Association ("Not to
be confused with those space people,"
he is quick to point out), a loosely-knit
band of computer-school dropouts currently
occupying an abandoned fraternity house
at the University of New Mexico.

why are u stuck?

[Anonymous Coward]

probability ain't that hard. what are u doing?

Funny you should mention that...

[Anonymous Coward]

I just got back from a workshop on this very subject, but nobody uses the term "baby bootstrap". It is now called "Developmental Robotics", and encompasses embodied agents, machine learning, and other biologically-inspired metaphors.

There is now a website dedicated to the idea. See http://DevelopmentalRobotics.org/ and http://cs.brynmawr.edu/DevRob05/ for a collection of papers on the subject.

[Obligatory]

[Anonymous Coward]

"...or has their research moved to other, more classified levels?"

"I could tell you, but then I'd have to kill you....

Re:[Obligatory]

[AndersOSU]

equally obligitory
In soviet russia babies boot strap you

Re:Babies have an instinctive understanding of 're

[lux55]

This was a point Nietzche made in Beyond Good and Evil, that the will is the least-well understood aspect of human nature, and the one we make the most assumptions about our understanding of. Interesting that will/volition/motive/morality (aspects of the same grey area) pose such a fundamental problem to AI...

Re:Babies have an instinctive understanding of 're

[maxjenius22]

Cycorp is making progress, though.

I recommend reading Witbrock, Michael, D. Baxter, J. Curtis, et al. An Interactive Dialogue System for Knowledge Acquisition in Cyc. In Proceedings of the Eighteenth International Joint Conference on Artificial Intelligence, Acapulco, Mexico, 2003.

Also, if you are a lucky college student, go see the author talk about Cyc teaching itself at USC or Carnegie Mellon..

Oh, and for once, I actually am an expert on the topic, not that that matters on slashdot.

Significant Progress

[God of Lemmings]

....yet there has been no significant progress in over 30 years.

Speak for yourself human!

Two words

[neomorph]

Biological chauvinism.

It comes down to a matter of perspective. While Searle couldn't possibly grok that the system of the book, the worker/ordertaker, and the room opening "understands" Chinese, he thinks it natural to believe that the system of neurons, blood vessels, organs, and bodily fluids called "Mao Zedung" understands Chinese.

Why? Merely convention. Defining intelligence by mechanism (in Searle's case: neurons) is problematic because it precludes definition in situations where mechanism is unknown. If an alien race landed on Earth tomorrow and demanded to speak to our leader, are we going to kill one and dissect it to verify it has neurons before we negotiate?

Put another way, Mao Zedung's clone, properly taught, knows Chinese. A supercomputer of the future, exactly simulating the effects of all of the neurons in Mao Zedung's head, should "know" Chinese too, otherwise one ends up with an analog of dualism's "zombie" problem. The brain of Mao Zedung's clone could have been replaced by a wireless link to the supercomputer. So, even though Mao clone will act and behave exactly the same as if he had a real brain, he's doesn't "understand" Chinese.

To answer your question, we can't preclude silicon from being intelligent merely by decree. We have to evaluate artificial intelligence the same way we evaluate biological intelligence: by observing the outputs from the party in question, applying semantic content to those outputs, and seeing if that semantic content jives with our own understanding of what it means to be intelligent.

You ain't kidding!

[Ayanami Rei]

As I recall from the AI classes I took one of the big limiting factors to the success or failure or some net-based AIs is determining the kinds of input and output layer nodes... what to feed in and what default node weights to set.

More concise and domain specific inputs helps to make the net more flexible and less artificially constrained by input vectors (too many extra degrees of freedom = excellent but unrealistic "fit")

I've Said It A Hundred Times Here

[Master of Transhuman]

Conceptual processing is the key.

Without a good conceptual processing simulation, no AI is feasible.

The "baby bootstrap" is one level removed from having a good conceptual processing simulation.

Everybody in AI who USED to be working on conceptual processing - such as Roger Schank - moved on to other, more immediately profitable AI applications. (Last I heard, some years ago, Schank was working on case-based reasoning applied to education IIRC.)

AI in airplanes...

[patio11]

Well, we can throw billions of dollars at AI research to program topgun.exe... or we can spend a couple million dollars and learn how to rig up a jet simulator to a jet by means of an encrypted signal.

We've got UAVs at the moment which can fly, maneuver, photograph, and fire -- all require quite a bit of intelligence (one human operator, done), none require actually exposing the human operator to physical stresses or that pesky "getting shot down" thing (not that that is really a big issue anymore -- most of the deaths in the Air Force nowadays are due to operator error, faulty equipment, and training mishaps, not The Bad Guys). The only difference between the UAVs and our mainline fighters is that we haven't revamped the fighters yet for, well, mainly non-technical reasons (institutional intertia, etc).

By contrast, we've got NOTHING, and I mean NOTHING, which comes close to being intelligent enough that it could be allowed autonomous control of an airplane. And, honestly, there is very little sense in developing the capability -- the networked plane is cheaper and more effective for the forseeable future unless somebody is able to trump our commanding technological lead by a factor of several thousand, in which case we aren't going to war with them anyway.

There is way too much bullshit in this field

[Animats]

I'm underwhelmed with the AI community. I went through Stanford CS. I've met most of the big names. I have some patents in AI-related areas myself. But really, nobody has a clue how to do strong AI.

The expert systems people hit a wall in the mid-1980s. An expert system is really just a way of storing manually-created rules. And those rules are written with great difficulty. There used to be expert systems people claiming that strong AI would come from rule-based systems. (Read Feigenbaum's "The Fifth Generation"). You don't hear that any more.

Hill-climbing systems (which include neural nets, genetic algorithms, artificial evolution, and simulated annealing) all work by trying to optimize some evaluation function. If the evaluation function is getting better, progress is being made. But what this really means is that the answer is encoded in the evaluation function. If the evaluation function is noisy (as in, "does the creature survive") and requires major simultaneous changes to make progress (as in "evolutionary jumps"), hill climbing doesn't work very well. There is progress, though. Koza's group at Stanford is moving forward, slowly.

The formal logic people never made much progress on real-world problems. Formalizing the problem is the hard part. Once the right formalism has been found, the manipulation required to solve it isn't that hard. There's not much work going on there any more.

The reactive robotics people also hit a wall. Literally, as every Roomba owner knows. Reactive control will get you up to the low end of insect-level AI, but then you're stuck.

Reverse-engineering brains still has promise, but we can't do it yet. Progress is coming from trying to reverse engineer simple animals like sea slugs. (Sea slugs have about 20,000 neurons. Big ones.) Efforts are underway to completely work out the wiring. Mammals are a long ways off.

Lately, there's been a trend towards "faking AI". This comes under such names as "social computing". The idea is to pick up cues and act intelligent when interacting with humans, even if there's no comprehension. This may have applications in the call center industry, but it's not intelligence.

I run one of the DARPA Grand Challenge teams, Team Overbot. On a problem like that, you can definitively fail, which means there's the potential for real progress. That's why it's worth doing.

Re:There is way too much bullshit in this field

[Shotgun]

Lately, there's been a trend towards "faking AI". This comes under such names as "social computing". The idea is to pick up cues and act intelligent when interacting with humans, even if there's no comprehension.

Are you sure there not working on a replacement for the pointy-haired boss?

Re:There is way too much bullshit in this field

[Anonymous Coward]

Progress is coming from trying to reverse engineer simple animals

Finally we can discover the reason why reality TV is so popular!

bootstrap's bootstraps.

[Anonymous Coward]

Sorry. The title make me think of Pirates of the Carribean.

Drive to survive is the basis for intelligence

[Anonymous Coward]

It's the basis for all animal life, and since intelligence occurs only within animal life, it's reasonable to assume that it's the basis for intelligence as we know it today.

Seriously, the drive to survive is the key ingredient in the soup of life. The secondary drive is to reproduce. Intelligence is simply one way to succeed at those goals. Another, different way, would be to restrict the environment and then hardwire to maximise within that environment. A good example of that type of system is the shark.

But it all starts with living, following by replicating. Personally I think the most important steps toward artificial life* (which is what most people mean when they say artificial intelligence) are computer worms and viruses. Of all computer programs, these most closely resemble the base activities of animal life--surviving and reproducing. Therefore these are the best basis for an attempt at true artificial life.*

* The distinction, as noted in the parent, is volition--aka will or drive. Computers are already more intelligent than humans within specific limits--for example, when modeling nuclear explosions, or playing chess. But usually when people use the term AI, they are including/assuming self-awareness and self sufficiency...aka a will.

AI is not AL

[Anonymous Coward]

AI has become such a generic misused term that it can only reliably be used to describe machine learning. A collection of useful algorithms derived from memetic principles which can be used to model real world problems. Nothing more.

Unfortunately what most people think of AI is actually AL. The quest for artificial life, artificial reasoning, congnition, and sentience.

The reason why we are fumbling in the dark when it comes to artificial life research is becuase we still have no idea how the human brain (or any brain for that matter) really functions at a low lever. Sure, we know a little bit about neurons and their connectivity, but we cannot model the way the brain works down to the level of why a certain neuron fires and another doesn't.

Its quite plausible that the processes which enable creative thought and sentience are quantum in nature, and therefore not something that would ever be possible to emulate using current technology.

Its also plausible that in our ceaseless human ego we overlook how simple the processes really are in an effort to mystify our sense of self which is nothing more than an internally focused narrative. Looking at the processes that lead to on ones own perception of reality is fraught with difficulty.

The thing is, we just don't know which one of these alternatives were looking for.

One-time-pads are a joke, encrypt not the answer

[AHumbleOpinion]

Encryption has made such leaps and bounds lately that cyphers can be made that literally would take longer than the age of the universe to crack. Add to that the fact that one-time-pads can be loaded into the plane's memory before launch, and you really don't have to worry about anyone else hacking the planes.

For this particilar application encryption is not the answer and one-time-pads are a joke. You can not build an AI-based Air Force on the assumption that your one-time-pads will not be leaked through plain old human error or human espionage. History shows that "secure" encryption systems are defeated not so much by technology and theory but by sloppiness, betrayal, and other human flaws.

Ah, you've found me

[writertype]

I lik /. It r00lz! But lik any k1d, I use IRC. -DARPA AI 312 Mark 26.12

Baby Boostrap = bad name

[Anonymous Coward]

A baby bootstrap, in my book, needs motors and sensors. The very idea that you can recreate intelligence without any inputs other than ASCII is unproven and intuitively ridiculous. Sure, it's an interesting experiment to try to see what happens to a learning machine that has only sensors and no effectors, but it won't tell you about human intelligence or even interesting alien intelligence.

AI needs to understand HI, otherwise it's worthless or even dangerous.

From Shannon

[Baldrson]

Quoting Mahoney's paper:

3. Human vs. Machine Prediction We now compare text prediction in humans and machines, with the reassuring result that humans are superior. Shannon bounded the entropy of written English by having people solve character-completion puzzles (Shannon 1951). Text samples were taken from a variety of sources, such as classic literature or technical manuals. The text was reduced to a 27 character alphabet of monocase letters and spaces. Subjects guessed at each letter until correct, and were allowed to use references such as dictionaries and character frequency tables. Shannon estimated that the entropy or uncertainty of written English is between 0.6 and 1.3 bits per character.

Re:From Shannon

[anthony_dipierro]

That figure is probably too high, then. And I'm not sure it proves anything. I mean, maybe if the computer came up with the compression algorithm itself. But I can code in the rules of the English language for my computer and that doesn't make my computer smart.

There are already algorithms out there that can get about 1.8 bits per character, and these are fairly elementary algorithms which are designed to run relatively fast on todays machinery. So I don't think 1.3 bits per character would be very hard at all, if that was all you had to do. Obviously from the text you gave subjects were allowed to use dictionaries and character frequency tables, so this would be allowed for the computer too. Include a dictionary, encyclopedia, frequency table, grammar guide, etc., and I just don't think 1.3 bits per character would be that hard.

Crash a hen.

[Vo0k]

Even better...

Hens aren't particularly bright animals, so they are quite convenient target of research - rarely some obscure unknown "feature" of the brain gets in the way. So a lot of research was made.
Hen's NI has weasels hard-coded as natural enemy. Seeing a weasel is a signal to panic and run. But the researchers were testing just "how exactly" is the recognition of a weasel coded. Not too exactly, it shows. Simplifying the model, they got to a point where the "weasel" was just a black ellipse cut out from paper, with two brighter spots in one end. It triggers the panic reaction in hens.
Now there's another signal a hen recognises: chirp of the chicken. The signal tells the hen to search for the chicken and to nurse it. The sound must be pretty exact recording, but may be emitted by quite arbitrary object to trigger the reaction.
Then they got the idea of making the "weasel" cutout to emit the chirp.
The hen just stands, watching the object and completely ignores the world around. It can be pushed, it can be stabbed, it's not reacting, just standing there, only basic functions like ballancing on the legs working, but all senses are blocked - until quite a while after one of the signals is removed. Essentially, they just crashed the hen's firmware...

Re:Crash a hen.

[hobbit]

Sounds interesting. Who made this research? I Googled but didn't find anything.

Re:Crash a hen.

[Anonymous Coward]

I wish I knew... My teacher told that during the introductory lecture of Vision Systems (semester about image recognition), and since I finished studies, I lost contact with him.

Re:Crash a hen.

[Anonymous Coward]

Hens don't nurse chicks. Duh.

you don't need a lot of money

[X10]

I'm convinced that baby bootstrap is the best way, or even the only way, to create really intelligent computers. The problem is that so far, nobody found the right computing algorithm. What's needed is a good idea rather than a lot of money.

What was the baby called?

[houghi]

I hope it was not H.A.R.L.I.E., because then I also know how the rest will go and that is not really good.

(for those who do not know what I am tallking about)

Re:Two words

[thelen]

I don't think this objection is fatal to Searle's basic view. He was interested in arguing that mental states could be derived from the physical processes of the brain, but not from simple computation using rules and states, which is what AI of the 60s and 70s was striving for.

The account, just by virtue of its monistic materialism, must allow for the possibility of a machine being in principle capable of generating consciousness. I mean, the brain is a physical entity performing observable actions that can be described according to physical, chemical and biological laws, so therefore it's basic functioning must be replicable. The weak spot in the theory is that a lot has to happen during the "emergence" phase. But there's nothing to prevent, say, a sufficiently complex neural network from generating emergent properties, perhaps even consciousness.

I don't see anything in this (admittedly thumbnail) view that would lead us either to dissect aliens or forbid us to attribute consciousness to the remote control Mao. What it does purport to prove is that any alien or communist Chairman we believe is intelligent cannot be just an overgrown Turing machine.

In short, the Chinese Room experiment is meant to undermine the AI of the previous decades for being too focused on rules and syntax and computational states. I don't see it as a rebuttal of the notion of AI in general. It wouldn't be a very good naturalistic account if it did forbid AI a priori IMO.

Ill-defined

[Anonymous Coward]

"The question of whether a computer can think is no more interesting than the question of whether a submarine can swim." --Dijkstra

Point being: The so-called "baby bootstrap" is ill-defined. Come up with a specific task which performance can be measured, and we'll try to solve it.

...no significant progress in over 30 years...

[gatkinso]

If anyone can't see why they pulled the plug on crap like this, they need a lobotomy.

1000 times a second

[blueberry(4*atan(1))]

I think he got confused while paraphrasing Kurzweil et. al. and claiming their ideas as his own.

All robots

[blueberry(4*atan(1))]

please report to the dance floor!

Classified Research

[EngrBohn]

Has the military really given up on this concept, or has their research moved to other, more classified levels

I'd tell you, but then I'd have to kill you.
Ah, what the hell. I'll kill you anyway.

Learning What??

[PingPongBoy]

Simply put, the 'baby bootstrap' would empower a computing device to learn like a child with a very good memory. ... No sensors, servos or video input - it only needed terminal I/O to be effective.

The input stream at a terminal would hardly appeal to a child so how can a proper evaluation of the learning be done?

Suppose the input is a sequence of zeros and ones. Could the AI come to any kind of understanding? Perhaps a prediction whether the next input might be a 0 or a 1, eh? But no! Let's fool the AI now by telling it who is the real boss. The AI has no idea that it is being spoken to by a terminal. The next input is the letter "g". How unpredictable!

Garbage in, garbage out - let's look carefully. A child plays and experiments. A great deal of a child's theories are garbage. The world in a child's eyes is a set of samples. Like the Mars rovers a child could follow a path that seems fairly limited in character, then bingo, something new comes up.

Intelligent behavior in a child emerges when different theories are assembled towards a goal. First the child realizes that s/he has some ability to either influence the environment or to manipulate information (which may be stored as symbols or images, as far as a computer is concerned). If the child conceives of particular classes of objects, the child can begin to reason. Several concepts such as self, ability, action, time, place, class, possession, etc. would be regarded as fundamental or at the very least useful. As a child accumulates and refines these concepts in the mind, the child can reason more and more correctly or effectively.

An simple artificial world can be represented as a set of strings that are transmitted to a baby bootstrap. The simple strings would be a simple bootstrap for priming the learning mechanism by letting it realize a number of essential concepts. Then more complex worlds as well as more arcane representations (such as natural language) can be used in order for the AI to interact with the greatest possible group of users.

Still, the limited input feed is bound to cause the most ridiculous problems. Pointing out that the learning system has a big memory doesn't give me any idea what the machine will achieve.

forum 3000

[sean@thingsihate.org]

The idea of using the internet as an input for this sort of thing reminds me of Forum 3000, and the SOMADs they had that were supposedly based on the usenet posts of a single person, etc.

that was a cool site.

US military pact with Taiwan

[hypervinetest45]

Unfortunately, tensions between Taiwan and China may draw us into a conflict with China that we don't want. See, we have a military pact with Taiwan. If China invades, we go to war on their behalf. We have considerable resources in both "countries" (China considers Taiwan a "rogue state", not a country). Bottomline: a war between the two is inevitable and will surely damage our economy.

Re:US military pact with Taiwan

[bushidocoder]

Its not inevitable - The conflict between the Taiwan and China has been raging for decades, and although every couple of years its reignited politically, to date we've been fortunate enough that no one has stepped over the line. Although its certainly possible that we could end up in a war we don't want to be in, I believe that its possible to diffuse this peacefully over the coming decades. The US doesn't want a war with China, and few in China want to rile the US. If we keep cool heads and a strong diplomatic relationship, and we continue to entwine our markets closer together, I think this one is going to end up okay.

Re:US military pact with Taiwan

[Rich0]

Bottomline: a war between the two is inevitable and will surely damage our economy.

Taiwan is more valuable to the Chinese leadership as an enemy than as a conquered land. As long as the tension exists nationalism will serve the Chinese leaders.

In any case, what is worth fighting for in Taiwan? Sure, there are lots of well-educated people there and good industrial development, but how much of that would survive a full-scale war? How useful will those motherboard factories be after China fires off all those ballistic missles they have massed across the sea?

Does China really want to tick off its biggest customer right when their economy is primed to become probably the largest in the world? They just need to sit back and they'll be a major power in a matter of years. Why tick everybody off before then?

Then again, China is basically a dictatorship, and you can't always predict how dictators will act. However, the status quo is clearly more valuable to China than a war over Taiwan.

Re:US military pact with Taiwan

[CaptainAvatar]

By that logic, the First World War would never have happened. Germany was the world's second biggest biggest exporter in 1913 and growing, and its biggest customers were all the countries it went to war with in 1914-8. That's why I'm never convinced by the argument that countries that trade a lot have too much to lose by going to war, or that they'd only acquire a pile of rubble if they won, so they won't bother. It's not always that rational a decision. (One thing that made Germany overreact in 1914 was their fear of Einkreisung - encirclement - by the Entente Cordiale powers. One thing that China worries about today is containment by the US and friends. Hmmm ...)

Having said that, I think you are right that Taiwan is more useful to the Chinese leaders as an enemy. But they may not be able to control the forces of nationalism, once unleashed.

Re:US military pact with Taiwan

[Rich0]

Agreed. War is not always a rational decision. Taking on the world never is...

Re:They're still working on artificial stupidity..

[kabocox]

Anyone doing conventional research would
have undoubtedly consigned the hapless
computer to the scrap heap. But for Boran,
the Akron I's response represented a
startling breakthrough in a little-known
field: artificial stupidity.
Boran is the head of NASA, the National
Artificial Stupidity Association ("Not to
be confused with those space people,"
he is quick to point out), a loosely-knit
band of computer-school dropouts currently
occupying an abandoned fraternity house
at the University of New Mexico.

And the really sad part, due to the Peter Principle, their AI gets to manage the smarter more knowledgeable AIs.

I know why

[Anonymous Coward]

Today the internet could provide a developmental database far beyond any testbed that we imagined, yet there has been no significant progress in over 30 years.

Because when they plugged it into the internet, it just looked at porn and masturbated.

holywood misnomers

[AndersOSU]

Why does everyone think that as soon as we create an AI that is capable of learning that is going to continue to learn indefinitely. Isn't it more realistic to first seek out an AI that can learn as well as a dog? And move on from there AI that learn as monkeys, then chimps, and maybe by that point we will have a good enough handle on the problem to create a crude human AI.

Not only that but by creating increasingly sophisticated AI's we learn our self. If the first AI we create is a "bad dog" we learn how to create a good dog. Hopefully by the time we get to creating a human AI we've learned enough to prevent creating a sociopath.

I realize that there is a difference between intelligence and sentience, but why should we expect to create a sentient computer program the first time out of the box.

Clearly I'm no expert in the field, but I find it hard to believe that the first AI ever is going to be an unmitigated success (even if it is evil). I find it much more likely that when and if the field begins to bear fruit there will be a learning curve, and the first AI won't view us as play things.

Ancient Jedi CPU trick

[Thud457]

"These are not the humans thar you are looking for."

"The locus of all evil in the modern world"

[Thud457]

I bet you kids think the phonecops are merely a figment of the benighted past now, too.

Re:"The locus of all evil in the modern world"

[cayenne8]

" I bet you kids think the phonecops are merely a figment of the benighted past now, too."

Damn...those phonecops play hardball...

:-)

Developing weapons again I see.

[Gizmoguy]

I'm sick of this, the government is always developing new weapons, new ways to obliterate the enemy. Why can't we all just live in peace and harmony and play games with eachother instead of blowing ourselves up! We're probably the stupidest race on the planet because we're the only race that destroys themselves. On Christmas Day during WW II the Germans and Americans stopped fighting and had a game of football, then got back to killing eachother the next day, wouldn't it be lovely if that Christmas Day was every day?

Artificial Intelligence is already plentiful

[TheLink]

Look at the typical Slashdot post. Definitely AI.

_Actual_ intelligence would be more useful... ;)

I know that!

[frank_adrian314159]

...will/volition/motive (aspects of the same grey area) pose such a fundamental problem to AI...

Not surprising. Some mornings they're a fundamental problem for me! This often happens after an intersection the previous night of myself and quantities of alcohol. It also happens at the confluence of various mornings and the idea of going to work. Of course, this just might mean that my intelligence is limited...

I for one...

[Anonymous Coward]

Welcome our new baby overlords! (Well, I already have one, so it's not an issue for me...Nothing will change).

Curious...

[Anonymous Coward]

How many nodes are needed with temporal processing and feedback? (I+O)^5 ? more? less?
Just what do these models look like?

Nagel Bagel Lox I must be hungry

[kris_lang]

Hey, you gave a close approximation of the answer I like to give to the Chinese Room. I think I'm going to appropriate your concept and use it from now on. I agree with you wholeheartedly.

The system is more than just the room itself; in fact, the system is more than just the bag o' fluids that each individual is. Our semantics have an evolutionary component to them in that a lot of our imbued semantics are totally dependent on cultural learning (je parle francais quand j'etais nee en France, aber Deutsch, ungregi, or esperanto if born in the lair of a deluded 1960's psychology professor) and placing "red" or "rouge" on the appropriate stimulation of our selection of visual cones makes us think we share a commonality, even if we don't [e.g. brothers with slight variations in visual pigment opsin genetic sequences will have different color matching functions, some women may be tetrachromats in having four visual cone pigments and see the color in their 4-d space, some people are limited to 2-d color because of genetic limitations, et cetera]. Good
golly, I feel a phantasm of Nagel's bat coming on.

And some of these subjective components, or predelictions towards certain associative components and actions, may in fact have enough reproductive advantage that they may be hardwired into the blueprint DNA of creating our nervous system, much as reflex righting actions are definitely encoded for in our DNA. We just don't know how... yet!

Qualia! Qualia!

Universality, schmuniversality!

[exp(pi*sqrt(163))]

Really! People working in neural nets don't half talk a lot of bull sometimes. We have a universality theorem saying that any function can be represented by two layers of signoidal functions. So what?. Any smooth (univariate) function can be represented arbitrarily accurately by a piece of the Riemann zeta function (believe it or not!) but nobody goes round saying that this is a good way to represent functions. The existence theorem for neural nets is devoid of relevance to anything. It's just an abstract existence theorem that tells you nothing about real world applications

Re:Universality, schmuniversality!

[polv0]

I wouldn't say that it's devoid of relevance to anything. When comparing neural networks to "simpler" modeling methodologies such as generalized linear or generalized additive models, the distinction is very meaningful.

Regardless, you haven't addressed my primary point that the limiting capacity is the construction of meaningful and comprehensive training data. There are three primary constraints on developing these models:

The data used in training.

The structure of the model.

The method used to train the model.

I think there are tons of smart people out there capable of developing model structures and training methods to emulate and generalize with data given to them (e.g. neural nets, svm's, genetic algorithms, random forests, etc...). But who can construct a set of data that could support an "intelligent" classifier? I don't think we are on the right planet yet, let alone the right ball park.

Re:They're still working on artificial stupidity..

[Anonymous Coward]

Holy shit! This is the funniest post on /. in ages. I salute you, Sir.

glial cells

[kris_lang]

Hey there,

Glial cells are supportive possibly nutritive cells. I remember reading somewhere about them being involved in the thought / memory process, but nothing concrete. I know that Francis Crick was going into what the proteinaceous components of memory and learning might be when he was at the Salk Institute in La Jolla, and that he was strongly interested in the interconnections and what the "seat" of consciousness might be, which is why he was severely interested in the thalamus. Cool guy, he was.

If you've got any definitive references on the glia involvment in neural processing, let me know.

As for the irreducible elements etc, somewhere else in this spaghetti of comments I made a note about how the axonal-dendritic connections are more complicated than a single synaptic connection with a single static threshold. There are multiple synaptic contacts on the hillock and each of these connections leads to a spatio-temporal concert of association.

Thanks

Re:glial cells

[Artifakt]

Glad to hear this idea goes back at least to Crick, although I presume from the your context more recently than his Nobel days.
I don't have anything definitive, but if you'ld settle for "just maybe interesting", this:

http://apu.sfn.org/content/Publications/BrainBrief ings/astrocytes.html

might qualify, or at least lead you to something deeper.

Note: if you just google for 'Glial' and 'memory', you will probably first and foremost get a lot of claims that certain Statin drugs that are known to be absorbed heavily by glial cells can also adversely affect long term memory. You'll have to wade through quite a bit of that to find links pointing to more fundamental research. Best of luck if you bother.

re. multiple connections and non-static/non-singular thresholds, that's not only interesting, but it sounds like it will apply whether glial memory applies or turns out to be a red herring.

That's why I use Kolmogorov compression instead.

[Baldrson]

Ultimately, if you were free to do things the way Shannon did things -- and all you had to do was produce a major corpus without counting the dictionaries, etc. as part of the compressed corpus, then you simply have a bit, the value of which could be 1 or 0, where, say "1" means "the corpus" and 0 means "not the corpus" and you hand that bit to the "interpreter" which is not counted as part of the compressed representation.

You can achieve arbitrarily high "compression" ratios through such a fallacious system.

Kolmogorov compression ratios have to include all dictionaries, algorithms, etc. that you use to produce the uncompressed corpus. That's why I specified things the way I did for the K-prize and why I called it the "K" prize.

/me stands up and applauds

[Compact Dick]

You, Sir, have earned my respect.

Re:That's why I use Kolmogorov compression instead

[anthony_dipierro]

and all you had to do was produce a major corpus without counting the dictionaries, etc. as part of the compressed corpus, then you simply have a bit, the value of which could be 1 or 0, where, say "1" means "the corpus" and 0 means "not the corpus"

I would assume you aren't allowed to see the corpus beforehand...

Kolmogorov compression ratios have to include all dictionaries, algorithms, etc. that you use to produce the uncompressed corpus. That's why I specified things the way I did for the K-prize and why I called it the "K" prize.

Then I don't see how the 1.3 bits per character is still applicable. After all, Shannon didn't count the bits being used for the dictionaries in people's brains (even besides the fact that he let them use outside dictionaries).

Anyway, as long as you don't have access to the corpus beforehand (and as long as the algorithm scales linerally according to corpus size), it seems reasonable not to count the dictionaries/algorithms/etc. After all, as the size of the corpus increases, the effect of the dictionary becomes negligible, and the compression ratio approaches the same figure anyway.

Think about it, if we reduce the size of the corpus to one paragraph, humans are still going to be able to perform at roughly the same rate. But with such a small corpus it would probably be impossible for a computer program to be written to perform at such a rate given Kolmogorov compression ratios, because Kolmogorov compression ratios penalize the storage taken up by the very intelligence you are trying to measure.

Compressing the library of Congress

[Baldrson]

Think about it more like compressing the entire content of the library of Congress.

Since you're going to have to produce a bunch of dictionaries anyway, you may as well use them to compress the rest of the library.

That's more the way to think about Kolmogorov compression's relationship to the artificial intelligence criterion.

Don't get hung up on Shannon's figure. You'll just end up chasing your tail.

Most information theorists tehse days recognize that algorithmic complexity using Kolmogorov complexity is the ideal way to define information content.

Re:Compressing the library of Congress

[anthony_dipierro]

Think about it more like compressing the entire content of the Library of Congress.

What's that, 10 terabytes? The overhead for the program itself would be insignificant, so we're probably arguing over nothing anyway.

Getting back to my point, we're already pretty close to the 1.3 bpc figure, and I think it would be rather trivial to reach that figure in compressing the LoC, if that were your only goal (of course, Shannon's figure was using only 27 characters, 26 letters and a space, so you'd have to force the LoC content into that form first, there'd also be the problem that the LoC contains more than just english text).

The calgary corpus was compressed down to 1.54 bits per character. Now, this figure doesn't include the size of ha, or rar, or the linux distro, but with proper motivation this could be compressed to a size which would be negligible compared to the LoC, though you'd probably have to ditch the Linux distro and just write a bootloader which runs your compressor directly. However, I think this points out how arbitrary it is to count the exectable itself - this means you're factoring in all the specifics of the intel processor, both positive and negative (mostly negative on intel, since you've got all the overhead for input, output, memory allocation, etc, but in theory the processor/bios could be utilized in some way, especially if you don't restrict yourself to intel architecture).

Anyway, I think the 1.3 bits per character mark would be reached in a matter of weeks, given an input as large as the Library of Congress, stripped down to 27 characters, and assuming the payout was at least a few thousand dollars.

Maybe I'm way off, but I hope someone takes your suggestion, because I'd love to give it a try. I'd almost be willing to do it just to prove you wrong, but it'd take a lot more than a few weeks without any monetary incentive.

Most information theorists these days recognize that algorithmic complexity using Kolmogorov complexity is the ideal way to define information content.

Well, we're not talking about information content, we're talking about artificial intelligence.

Well, come back when you understand whale-speak

[Anonymous Coward]

Well, come back when you understand whale-speak because right now we only know that whale-speak and first-post-speak aka English are mutually exlusive, not that the ability to say "first post" is superior .

Also, no man has ever tried to run a whale to human frequency converter for extended periods of time.

Additionally, you'd have problems to explain all this to a homo sapiens sapiensis cave dweller as well, we have a school system and culture to help us get educated.

Lastly, big dolphins are a better bet for "sapiensis"-intelligence than whales, because the dolphins have a brain-bodymass ratio similar to ours.

Dolphins are studying US

[AthenianGadfly]

Until one day... so long and thanks for all the fish.

We _are_ talking about information content.

[Baldrson]

Anyway, I think the 1.3 bits per character mark would be reached in a matter of weeks, given an input as large as the Library of Congress, stripped down to 27 characters, and assuming the payout was at least a few thousand dollars.

Maybe I'm way off, but I hope someone takes your suggestion, because I'd love to give it a try. I'd almost be willing to do it just to prove you wrong, but it'd take a lot more than a few weeks without any monetary incentive.

First off, "1.3" doesn't appear anywhere in the prize criterion, but proving Matthew Mahoney's figure of 1.3 bits per character as the threshold for AI "wrong" would be valuable for the simple reason that it would be a demonstration of the best current natural language model. (Mahoney's reference to Shannon, BTW, does give a range of .6 up to 1.3 bits per character for human intelligence so if you're convinced Mahoney was too lax in specifying Shannon's upper bounds then that's not saying much about the range provided by Shannon let alone the overall conception of information content as intelligence.)

Secondly, the M-Prize-like structure specified for the K-Prize is adequate to incentivize arbitrarily high degrees of intelligence and avoids the issue which threshold is the "right" one.

In any case when we're talking about irreducably compressing a natural language corpus as the criterion for AI we _are_ talking about information content.

As I said, obsessing about a particular threshold of 1.3 or even .6 isn't the point. The point isn't even that at _some_ threshold you have a critter that is indistinguishable from a literate human in its ability to comprehend a large body of literature.

The point is that at some point you have a critter that is able to comprehend a body of literature more accurately than anyone. That's when you might start getting a real alternative to existing "authorities".

Re:We _are_ talking about information content.

[anthony_dipierro]

In any case when we're talking about irreducably compressing a natural language corpus as the criterion for AI we _are_ talking about information content.

I just don't think it's a good criterion for AI, not strong AI, anyway. Computers are good at things when humans program them to do those things. One thing you can program a computer to do is recognize the rules of a language. That doesn't mean the computer is intelligent, it's just programmed well.

Yes, AI would be sufficient for good compression of language. But I don't agree that it's necessary. Moreover, there are too many factors other than intelligence which go into good compression, so I just don't think it makes sense to compare humans and computers heads up this way. It's like saying that Deep Blue can beat Kasparov at chess, so Deep Blue is as intelligent as Kasparov.

The point isn't even that at _some_ threshold you have a critter that is indistinguishable from a literate human in its ability to comprehend a large body of literature.

No, compress, not comprehend. I could be given instructions on how to compress Japanese text really well, but that wouldn't mean I'd be able to comprehend it.

The point is that at some point you have a critter that is able to comprehend a body of literature more accurately than anyone.

I agree that comprehension has some relation to ability to compress, at least in that knowledge of the rules of a language improves ability to compress. But that's not the only factor (even beyond the fact that knowing the rules != comprehension).

Ockham's Razor as a criterion for intelligence

[Baldrson]

I agree that comprehension has some relation to ability to compress, at least in that knowledge of the rules of a language improves ability to compress. But that's not the only factor (even beyond the fact that knowing the rules != comprehension).

Prediction is far more more than coming up with language rules.

For example, if I see some text attributed to "anthony_dipierro" and the subject is the relationship between AI and compression, I can predict that the text is more likely to say "compression isn't the measure for AI" than "compression is the measure for AI".

This goes far beyond simple bias estimates in speech act content as well.

For example, inductive logic programming, itself based on Ockham's Razor, has been extended to use regression analysis to derive physical laws using minimum description length criterion:

As a part of the system, pruning based on the Minimum description length principle was developed that can handle also continuous variables. It turned out that MDL pruning helps to build more comprehensible [emphasis JAB] models, while at the same time preserves model's performance in terms of its prediction power.

There is a school of thought in the philosophy of science that this is in fact the precise way of measuring the validity of a body of theory.

If you don't like Ockham, then how about Einstein summing up his focus on invariance thus:

Everything should be made as simple as possible, but not simpler.

Re:Ockham's Razor as a criterion for intelligence

[anthony_dipierro]

Prediction is far more more than coming up with language rules.

But that's the problem. You're trying to measure multiple things at once. So when you get a good value, you don't know which factor caused the improvement.

Way to be marginally technically correct

[Anonymous Coward]

You are correct in saying that the "ATF" exists no longer. It's now the ATFE (alcohol, tobacco, firearms, and explosives).

They are quite alive and well.

Where you are incorrect is that their sphere of responsibility has grown, not diminished. Just try to buy a firearm from an FFL and see which form you fill out (hint: ATF 4473).

Err, mathematicians will never crack one time pads

[cfalcon]

It's utterly impossible to do so.

The "human error" and "jammed signal" problems are signifigant, however.

Extra CPU is reserved

[cfalcon]

The second core will be used for spyware, silly!