Slashdot Mirror
A.I. and the Future
The guesses about the future above are as good as yours or mine. But Spielberg's haunting and provocative movie A.I. has opened a window into human consciousness and the moral implications of artificial intelligence.
This window is unlikely to last very long. The next Monica Lewinsky scandal is always around the corner, ready to fuel the Big Media machine and distract the public. Given the short attention span of Americans in particular to scientific issues like this (genomics, copyright, intellectual property, fertility research, alleged global warming), it's worth beginning a discussion on A.I. Where is it going? Which vision of A.I. and the future do you think is closest to reality? Will machines make us uncreasingly dependent on them, as the Unabomber suggests? Will they take us over, as George Orwell believed?
Or, as M.I.T. computer scientist and artificial intelligence researcher Kurzweil suggests, will humans and machines -- especially miniaturized, increasingly powerful computing machines -- simply become an integral part of out bodies and lives? Kurzweil envisions the distinctions between these two "species" and entities (biological and digital) rapidly blurring.
It says a lot about our willingness to think seriously about technology that no national politician has ever addressed these issues in a meaningful way. But a murderous student of technology has:
Unabomber Kaczynski wrote in his infamous manifesto:
"As society and the problems that face it become more and more complex and machines become more and more intelligent, people will let machines make more of their decisions for them, simply because machine-made decisions will bring better results than man made ones. Eventually a stage may be reached at which the decisions necessary to keep the system running will be so complex that human beings will be incapable of making them intelligently. At that stage the machines will be in effective control. People won't be able to turn the machines off, because they will be so dependent on them that turning them off would amount to suicide."
Reading that excerpt, it occurred to me -- not for the first time -- "What a shame this demented creature chose to express himself through the maiming and murder of innocent people." Because he sure has a point.
I'd like to know where he got the number for the human brain's ability.
The difference is that every tool until now has only done something that a human could already do, only better. Physical labor? Lever, power tools, cranes. Traveling? Car, airplane, spaceship. Use our senses? Media and communication technologies. An AI will be the first tool that is capable of doing something that it was not explicitly told to do by a human supervisor, and that has never happened before.
As emergent systems become more complex, "programming" their thoughts will become more and more difficult. Eventually, the three laws would have to be inserted via some form of psychological conditioning, the same way humans are given equivalent unbreakable rules.
Of course there's nothing that stops the robot from learning how his own brain works and hacking the laws out of himself.
okay, so this was intended as a joke - but who GIVES A SHIT if the machines don't have feelings?
I think it's really an academic question as to whether "intelligent machines" will arise, or whether they will manifest the phenomenon we refer to as consciousness.
We already have machines that are starting to do things - that make this question moot. The "music composing" machine that fooled the music experts. Deep Blue. Etc.
It's already happening, and it's inevitable.
Are these machines conscious? No, but it doesn't matter. To us. They are tools. Nothing more. Can an army tank be accidentally set to drive forward mindlessly without a human driver, and run over buildings and cars and cows? Yes, of course, and the same thing can and will happen with "intelligent machines" as well. It's inevitable. Especially as their workings become so complex as to be unpredictable, even to their designers. (hell, I'm supporting software that does that already).
Is humanity in danger? Of course. We've been in danger of extinction from our unnatural tinkering from the first time Oog started a campfire near some dry grass.
Why all the hemming and hawing about it now? Because it sells books, and tickets to movie theaters, and gets venture capital for companies working on "AI".
I personally believe that natural (or supernatural) human consciousness will never be duplicated by a machine. Others believe that one day, we'll have the mastery over the physical universe that will permit that. Frankly, that's not an important question, because outwardly, "intelligent machines" will be indistinguishable from conscious beings long before we might actually reproduce human consciousness (if that were possible).
Outwardly indistinguishable.
Long before that, either we'll figure out adequate safeguards for such machines, or we'll be the victim of a stupid accident, and become another species on the very long list of extinct species. No biggie. It happens.
These are my friends, See how they glisten. See this one shine, how he smiles in the light.
You're missing a very important one that was created R Giskar (I'm not sure of the english name) and taken over and exercised by R. Daniel Ilivaw: (which forces a rewrite of the 3 laws you mention)
Zero: A robot must not harm, or through inaction, allow to come to harm Humanity.
One: A robot must not harm, or through inactivity allow to come to harm, a human being, as long as it does not conflict with law zero.
Two: A robot must obey all commands given to it by a human being except when these conflict with the first law.
Three: A robot must preserve itself at all times unless by doing so it contradicts the first two laws.
So, assuming this is programmed right, the law Zero could most-certainly be applied to kill as many humans as required for humanity to progress, and most-certainly allow AI to take over humans in every aspect of their lifes.
The foundation books by Asimov illustrates this quite clearly. The master puppeteer is an AI robot throughout thousands of years. For the better of humanity, of course.
Karma karma karma karma karmeleon: it comes and goes, it comes and goes.
Or a Wand of Enlightenment. That'll work too.
All this stuff about leaving behind the bits controlling bodily functions and sex drive and things is a bit naive I think. I'm guessing that all of these messy sensory inputs and systems are necessary for intelligence and consciousness. I mean, if all you think intelligence is is a set of well defined rules linking together a set of ideas, there's Cyc, which I don't think anybody thinks is intelligent.
Need a Python, C++, Unix, Linux develop
Bit of a problem here. You seem to be assuming that the machines would have to be self-willed, inner-directed entities to "take over the world". Not so. Consider a small subset: Air-traffic controller.
... and the insurance companies will make sure that they do! Again, no intention is needed on the part of the computers. This needs a lot more intelligence than the simple case of the airplane, but nothing that we would call sentience.
20 years ago this was a dominantly human directed process, though radar, radio, etc. were already totally necessary components. Incrementally over the years more has been automated. In the last couple of years steps have started towards having planes automatically dodge each other. They already pretty much navigate from known origin to known destination. Once planes keep track of each other, and landing is automated (probably technically doable today, but not yet acceptable) then we move towards the stage where the pilot is just there as an emergency backup, who probably wouldn't be able to do anything anyway (the instruments display via the computer, active flight depends on a computer to manipulate the wing controls, etc.). Maybe he'd be able to reboot it.
This process proceeds without any intention on the part of the machines, but it causes the entire flight experinece to be totally dominated by these same machines. No A.I. needed, beyond the ability to dodge, navigate, and land. (These are either already here, or just about.) And social acceptance, which is the sticky point.
Watching drivers, I have frequently speculated how much safety would improve if computers were driving. Not really practical yet, except on specially prepared test tracks, but slow steps in that direction are visible, if you look for them. Again, a bit part of the limitation is social acceptance. Without that, any automated car will be 100% responsible for any accident that occurs, regardless of the circumstances. So nobody works very hard on developing it. (And it is a tricky problem, no question!) But eventually it will be here. Then the computers will have taken over the cars
Caution: Now approaching the (technological) singularity.
I think we've pushed this "anyone can grow up to be president" thing too far.
The lack of a desire to might stop that.
So much of the argument is a projection of our own purposes onto a computer. This is a bit wierd, as I find it quite difficult to get a computer to see things the way I do on purpose. To assume that it would happen by accident strains credulity.
Caution: Now approaching the (technological) singularity.
I think we've pushed this "anyone can grow up to be president" thing too far.
I don't think that the problem is that the computers being used aren't fast enough or don't have enough RAM.
I think the fundamental problem is that we don't know what intelligence is, and so are, understandably, finding simulating it quite difficult.
People have been saying since the fifties that, in 10 or 20 years, we'll have sufficient computing power for a machine to become intelligent. Well, I have more computing power than those people could have dreamt of sitting under my desk right now. It didn't turn out to solve the problem.
What AI research needs is plenty of "RI" to crank through the conceptual problems too, not just the biggest supercomputers money can buy.
Cheers,
Tim
It's official. Most of you are morons.
As someone pointed out in the A.I. review, some of the questions of malicious machine intent can be solved by Asimov's Laws of Robotics:
1. A robot must not harm, or through inactivity allow to come to harm, a human being.
2. A robot must obey all commands given to it by a human being except when these conflict with the first law.
3. A robot must preserve itself at all times unless by doing so it contradicts the first two laws.
A.I. didn't reflect this very well - David put humans at risk at several points. This gives an inaccurate, overly frightening picture of the intelligent machines we would likely create - it somewhat serves as FUD.  "Oh no, the robots will only act in their own best interests, and we'll die."
0x0D 0x0A
But Spielberg's haunting and provocative movie A.I.
Didn't Katz say it was a POS last week? Heh, Katz reread last weeks column before writing this weeks!
And to answer your question.
Intersections used to have a policeman to direct traffic. Now we have automated lights. Ever tried to get around a big city when the lights go out? It's tough, but he policemen on the corners try to keep things going.
Oh, and what do people do after a hurricane? They cook on grills, and live in tents.
My point? People adjust to the point of least resistance. We rely on machines and automation because it is easier. If/when the machines die, we go back to doing the old/hard way. It sucks for a while until we get used to it, but such is life. I feel an undertone of "we'll all die out after the machines are gone" in your column. Let me reply simply, "No, we'll just adapt and start inventing new machines."
Aah, change is good. -- Rafiki
Yeah, but it ain't easy. -- Simba
..just what does this have to do with Enlightenment?
----------
----------
Check out my blackbox styles
If you want to truly 'protect' humanity from this technology, get a PhD in a relevant field and start doing serious research. Because that's the only place where you are going to be able to institute controls, at the development level. And by control I do not mean creating Ethics Police, I mean developing techniques that stop problems before they start, or at the very least, clean up minor messes before they become serious threats.
I'll point to computer viruses as the hackneyed example: What could have been a terrifying constant danger has become no more than a nuisance. Why? Because there are very intelligent people working in the field, and there are such people because there is a demand for protection.
Also, where does it say that humans and computers fusing is a bad thing? Explain to me what is so special about a given classification of elements. So what if my far future descendants make significant use of inorganic chemistry in their physiology? Big whoop. I'd still call them kin.
"Avast! Prepare for the rodgering!" THWACK! "Arrr.. me nards.."
David Kaczynski, Teds' brother, was profiled in the Washington Post Magazine on Sunday.
Best Slashdot Co
The first artificial intelligences would probably need a lot less computation power than that. Not being real organisms, they wouldn't have to concern themselves with the biological systems we do (digestive, cardiovascular, endocrine, nervous, reproductive, limbic, movement/navigation, etc, etc).
It's 10 PM. Do you know if you're un-American?
What AI research needs is plenty of "RI" to crank through the conceptual problems too, not just the biggest supercomputers money can buy.
I could not agree more. IMO, the most exciting research in AI right now is the work being done at MIT by Rodney Brooks and his students and colleagues. Dr. Brooks is also keeping a close eye on progress in computational neuroscience and I expect a few conceptual breakthroughs to come from that sector in the not too distant future.
The traditional AI community has conviced itself that AI will come gradually. They're in it for the long haul. I completely disagree with that assessment. I am convinced (as is Dr. Brooks) that there is something important that we are not seeing. Once we see it, AI will be upon us like a flood, almost overnight.
Could someone please explain to me what Godel, Escher, Bach is all about? I've heard about it, but I would like to know what the theme or main point of it is. Thanks.
(hehehe)
I think you may have it backward. Your reasoning assumes that the brain works like a computer, and each neuron is a process that takes up a given amount of RAM and a given number of CPU cycles. I think you could equally claim that each neuron is itself a processor, so you have 100 billion processors running in parallel, rather than 100 billion processes tended to by one CPU. And even that is obviously wrong, since it doesn't take into account how adjacent firing neurons affect each other, how a memory seems to be stored holographically throughout the brain rather than sequentially, etc. If true A.I. (whatever that is) is possible at all, it isn't going to run on a Pentium or the like.
And yet, humans are still on top. We have not yet reached the point where the creation is greater than the creator, and it's unlikely that we ever will -- As machines take over certain mundane aspects of human life, we move on and put our time into other things.
Got Rhinos?
"One World, one Web, one Program" - Microsoft promotional ad
The Anti-Blog
Agreed.
We're on the road to Tycho.
Speaking to the former point, however, you remind me a bit of David Pearce's thinking (which he calls the Hedonistic Imperative); if we can recreate consciousness, surely we can leave out the ability to suffer. I said to him, as I say to you, it may not be possible. And I mean, fundamentally, impossible. You are treading over interesting ground with respect to fundamental aspects of consciousness and subjective experience that we do not understand yet. If it were possible to systematically prevent suffering, however, I would tend to agree with him that by allowing suffering when we could choose not to, we would be cruel. Regardless, I am certain that we, as a people, would do it anyway.
I define the notion of "soul" as the idea that there is some agency beyond the brain which is responsible for our consciousness, our decisions, or our identity. I would hold that this has nothing to do with "good" and "evil," a dichotomy which is arbitrary and based, as much as we have a species-wide consensus on the subject, on our instincts, our genetic heritage.
We're on the road to Tycho.
I will go farther to say that I believe our machine consciousnesses will do what we make them to do, just as we do what evolution requires of a successful species.
We're on the road to Tycho.
I believe that the soul is sentimental superstition, and that the notion of human consciousness as somehow fundamentally "unique," "indomitable," or "unassailable" is insecure and adolescent. I have no doubt in my mind that we can and will make machines "in the likeness of a man's mind," and that these systems will, whether we grant it or not, be every bit as "human" in their thoughts as I am - they have my sympathy in advance.
We will, of course, learn a great deal of very important and revolutionary things about ourselves along the way. I believe human consciousness, not genetics or space, is our next great frontier, and we may see revolutionary developments there in our lifetimes. Cognitive science is a remarkably well-funded academic discipline, and has been the subject of massive and relatively quiet investment for several decades.
However, right now it's mired in very un-sexy pursuits, needling sea slugs and flies and mice, and we're still hammering away at nerve cell biology, chemistry, and physics. Pure theory of consciousness is pretty much at a standstill, after the great claims and great failures of the computer science-based AI folks, who showed pretty uniformly that, while they could do a lot of neat tricks, they had little fundamentally in common with the operation of human or animal intelligence, thereby at least giving us a slightly better definition of it.
And, in the meantime, we have "luminaries" who love to sit around in masturbatory celebration of what the future will be like, although this has the feeling to me a of a popular science magazine speculating about how we'll all travel around in air cars and eat food pills and vacation in space. It has nothing to do with the real implications of AI, and after the 100th or so run through the science media grinder, these tired old speculations are poor company whether they turn out to be true or not.
We're on the road to Tycho.
Yeah, the one good thing about those Packard Bells is that you could intimidate them into working..
But try that with an IBM and you'll get nowhere... IBMs need to be cajoled or bribed into working. Just say loudly, "Well, I *was* going to double the memory on this machine, but since it won't boot..". Works every time.
Compaqs, however, require a judicious application of precussive maintenance. They just won't listen to reason at all.
Also, NEVER NEVER NEVER screw the case cover back on before testing the card / memory you just changed. This shows the machine that you lack humility, and it will of course refuse to work. Turn it on and test it, THEN replace the cover. This shows the machine the proper respect.
--
PaxTech
All movements for social change begin as missions, evolve into businesses, and end up as rackets.
I'm a sysadmin. Duh. But let me tell you, Skippy, I bite those machines back HARD.
As machines become more intelligent, it is likely that such simple and direct laws will be difficult to program into them, if not impossible. For instance, if an intelligent system is a neural net that learns from it's environment, it will be problematic to have the Laws Module (LM) monitor the state of mind of the neural net to determine whether or not it is about to break the Laws. The LM will have to be "outside" the neural network of the AI (since neural networks are altered by the environment) but at the same time be able to interpret the network with sufficient understanding stop it from any transgression. In effect, the LM will have to be an extremely intelligent agent in its own right. It must understand the definition of "harm", which seems easy to us, but is extremely difficult to program. Then there are the ambiguities that crop up -- preventing the harm of one person causes another harm -- is some emotional harm more damaging than physical harm -- etc. It is questionable whether or not an intelligent entity that approaches human capabilities will be able to function 30 seconds with such simple rules constricting it in our extremely complex world without locking up.
Humans do not run by simple ethical rules. I suppose the most simply stated ethical rule is the Golden Rule -- "Do unto others as you would have them do to you." That won't work for AIs until they have emotions. It still doesn't work for most humans.
The Moore-Murphy Law: The number of things that will go wrong will double every 2 years.
Techno-hype.
Problem solved.
Honorary Member of Jackie Chan's Kung Fu Process Servers
I fear the day when IDA* search takes over the world! Aaaaaaaaaaaaah! It's complete! It's optimal! It's powerful!
For information on the man himsdelf, you can visit,:
http://web.mit.edu/invent/www/kurzweil_bio.html
His company's website can be found at:
http://www.kurzweilai.net
Tom
Oh arse
1: Gödel, Escher, Bach: An Eternal Golden Braid by Douglas R. Hofstadter
6 7/ singinst
Artificial intelligence. Cognitive science. Mathematics. Music. Art. Language. Computer programming. Zen. Philosophy. Self-reference. Genetics. Paradox. Logic. Everything.
http://www.amazon.com/exec/obidos/ASIN/04650265
May I suggest a few things? Read Kurzwiel. Read Hofstadter's Godel, Escher, Bach. Perhaps you'll come to understand the mindset of those who are developing this A.I. technology that every one else fears will run amok and distroy humanity. (I also thought I was supposed to be chained to a machine 24 hours a day working for the machines by now, too.)
There was a guy on IRC a few nights ago who I took for granted was a bot, and not a very good one. (Wild non-sequiturs, bursts of random abuse...) I thought people were putting me on when they insisted he was a real person, until the guy/bot made some reading-between-the-lines responses that could only have come from a human or a really superb AI.
Is there some kind of inverse Turing test to designate a human who is indistinguishable from a buggy Perl script?
Unsettling MOTD at my ISP.
Seriously, who most do you fear the most producing the AI units humanity would be dependent upon?
Microsoft
AOL/Time Warner
Disney
The Church of Scientology
Evil Mutant Communist Space Wizards©
Intel
Sun
Anything Steven Jobs is involved with
Her
Me
Cowboy Neal
-- .sig are belong to us!
All your
A feeling of having made the same mistake before: Deja Foobar
As long as the dominant operating system is a Microsoft product, I have no worries about "smart" machines.
---
I disagree. Intelligence is the ability to figure out a task. You want an intelligent robo-maid. Intelligence can exist without sentience. They are not the same thing.
As x approaches total apathy I couldn't care less.
I'll make this short and simple. If it makes money, It'll happen. If it doesn't, It won't happen *for long*
For those of you who want to understand how much CPU power and memory it would take to simulate a human brain, here are some figures.
Est. 100 billion neurons
Est. 60 to 100 trillion synapses
Est. 1 khz clock speed (times a neuron fires a second)
Assume we assign 32-bits for the given state of a neuron for 100 billion neurons.
Required memory for neurons alone: 400 gigs.
Now, synapses connect two neurons. So we need 2 pointers or index per neuron. Now 32-bits isn't enough as we can only index up to 4 billion some items.
Aftering playing with Excel, I figured we need at the minimum number of bits per address is 50. But because it's faster to work with bits divisable by 8, we'll use a 56 bit addressing system.
So, to connect a synapse to two nuerons, we need 14 (56 / 8 * 2) bytes, for atleast 60 trillion nurons.
Required Memory for synapses: 840 terabytes.
Now, you're job is to write a program that enumerates 840 terabytes of memory, one thousand times a second, performing calulations along the way.
"Communism is like having one [local] phone company " - Lenny Bruce
For those who want to see the current state of AI you might want to try Alice Bot. It's very good and I tricked one of my friends into thinking it was a chat room....
A good ChatterBot site is The Simon Laven Page. It has listings of interesting ChatterBots. My favorite is NIALL. It learns from what you tell it and comes up with some very funny responses.
--Volrath50
As for my contention about the (non-)feasibility of AI with current technology, it's impossible to prove a negative. The burden is on the other side to prove that it is possible. I have yet to see examples or evidence based on current technology of true intelligence of the sort that Katz says we should be worried about.
None of your examples are evidence enough for me:
You can certainly disagree about what constitutes "intelligence"; like I said, there is a great deal of healthy debate about this. But I have yet to see evidence of anything that looks like the kind of A.I. I would worry might take over the world as Katz describes.
Just because a person is wrong about a single thing does not invalidate all of his ideas. And anyway, I was citing Searle as an example of one person who is more skeptical in this debate, not as the utmost authority on the topic.
Clearly the trend favors the notion that one day we will have intelligent machines.
Well, we have also been able to build faster and faster vehicles. So by your reasoning, "the trend favors the notion" that one day we will have faster-than-light travel. I admit it's a bad analogy; what I'm trying to say is that just because computers are completing tasks that appear more and more like intelligence to observers, but still fall short of true intelligence, doesn't mean that one day machines will actually attain true intelligence.
However, I was not claiming that we will never, ever have intelligent machines. I said that given today's technology, intelligent machines are so far off in the future that they are not a matter for practical debate, as Katz claims they are.
If George Bush starts talking about how we need to have a worldwide dialogue on whether the machines will take over, we will really know he has gone off the deep end.
Hmm... 50/50 odds... not bad. I'll say... Truth?
I think that more or less, our society is already dependent on technology. For example, the Y2K issue which surfaced caused a certain amount of panic, as we began to realize exactly how much of our world is driven by technology.
As we move forward, technology becomes such an integrated part of our lives that we forget how to live without them. Who today can survive without the telephone, without our cars, without computers, our dishwashers, and hell, without condoms? The belt on our dryer broke this week and my parents couldn't do the laundry all of a sudden. Forget what people did for millennia -- my family needs the machine or we're helpless. The dryer has since been fixed, and it's business as usual. God forbid we ever lose electricity for a week.
As new things come out onto the market we increasingly use them for their convenience, and we forget how to do without them. As we become dependent on artificial beings to take care of the mundane daily tasks in our lives, we will soon realize that we cannot live without them.
Should the damned thing ever become smart enough to plot to overthrow me behind my back, or to steal from me, after months or years of faithful service I would have been trusting enough to let the thing do whatever it wanted. Because I never would have expected it.
Anybody else see 'The Score' last week?
You can accomplish anything you set your mind to. The impossible just takes a little longer.
I'm more worried about how people will abuse technology rather than how technology will abuse people. Computers and technology will always serve people. Which people is open to debate.
Disconnect your television. Do your own research. Draw your own conclusions. They're probably lying. Don't be a sheep.
When we try to convert our Napster .nap files to MP3's, call the new app KReader and have Adobe arrest us for breaking the DMCA after we've given our talk about our app at DefCon, I think oppression by machines would be a welcome replacement.
"Anonymous Coward" is for whistleblowers, not unpopular opinions.
Yes, these 'thinking machines' will develop quickly. Yes we will rely on them for many many things. The new IBM HelpDesk system is evidence of that. Yes, many dark futures may await us.
But in the end, it will be us who make the choice of retaining superiority or handing it over to machines. They are still our products, they will still only have the abilities and functions that we give them. Some may say that the essence of AI is learning by experience, thus enhancing the subject's own abilities - but we still, at this point, and will for some time to come, retain ultimate control over what these things are capable of. We have to make a choice. They will not exceed our own abilities unless we make the choice to either give them skills or give them skills without limits which would have prevented them from developing past a certain point. Of course, no vote of humanity could be taken and no laws could be passed that would be effective - it is a choice the engineers themselves must make every time they sit down to some code or some circuitry.
I'm not anywhere near skilled enough to do that kind of work, but I can say this with some fair amount of certainty: geeks love pushing technology, but are generally fiercely protective of their freedoms and liberties. I don't see many of them choosing to invalidate their own existence.
think for yourself, you won't like the results if others do it for you.
working Ray Kurzweil Link: http://www.kurzweilai.net/