AI problems

A note from the nitpicking department:
It was actually 6502 assembly code, which at the time was the CPU of several computers, most notably Apple II, Commodore 64 and my personal favorite: Atari 800XL.

You should probably watch Blade Runner (I strongly advise against the "director's cut") – where you have a whole police force for dealing with (read: terminating) rogue wetware androids (the in-movie term is "replicants").

I had a bad feeling about this; CPUs never have cute names like "Atari", there's always a numeric tail in that name, and in some notable cases, like the case in point, there's actually no alphabetic head... Sloppy. Well, I'll comfort myself with the thought that "Atari-based gadget" has a better sound to it than "6502-based gadget".

I should definitely watch everything combining high class drama with in-depth treatment of the psychophysical problem. That always stirs my faith in the human race. And my faith in the human race is definitely something I like stirred.

"Physicists claim that a bullet can fly out of one’s body with the wound closing and healing in the process, because observations indicate that you can get shot and wounded, and if a process is physically possible, that same process reversed in time is also physically possible."

Are you so sure about that? Wouldn't that break the 2nd Law of Thermodynamics?

Of course it would, but that law is statistical. For example, a gambler who keeps gambling and winning breaks the law of "The House Always Wins", and in fact none of us will live to see that, but it's physically possible. Particles around me can combine to form an exact copy of me without breaking physical laws, it's just extremely unlikely. So is a Turing-test-passing AI :)

so really what one can come away from this with is that we have nothing better to do than to write programs to do it for us?

I don't know how this is related to any of the above, but, um, I'd say that we have better things to do than things programs can do for us.

I have a warm spot in my heart to the "replacing ourselves with machines is insane" attitude, of which Vonnegut was one prominent proponent. However, I think this world view is fatally misguided and brain-damaging.

Perhaps you've seen a machine in the throes of trying to write copious numbers of error messages to a log because of a hardware fault. That's pain. Does the machine know that it's in pain? Perhaps, if there's some monitoring software with a rule that says "When the VMM is thrashing, the disks and CPU are pegged, and the hardware error log is increasing by more than 1000 entries per second, email/page/beep loudly for an admin to fix the problem". Human pain is information meant for the lowest parts of our brains, and they take over from the rational parts and try to get away from the carnivorous predator or the forest fire that evolution taught them to deal with.

@Ben: I disagree, twice: (1) pain is not just information – try it and you'll see that you can't brush off the psychophysical problem that easily, and (2) pain is not just there "for the lowest parts of our brain" – it's a pervasive feedback mechanism without which we're almost completely blind despite having seemingly impressive reasoning faculties.

why
pain is information
and feeling,born of pain are information
if you turn off the pain center of human,he will not feel it.
lobotomy?
learn the brain(not that i know it)

If you strangle a human, he will not feel pain after some time, either. Nor will a computer process information if you short-circuit it. The ability to break something fully or partially provides no exhaustive explanation of its nature.

And you cannot implement "pervasive feedback mechanisms" in software? Take the behaviorist approach: what do exactly feelings do which cannot be implemented in a machine? Heightened heart pulse? Sweating hands? A motivation to run? A motivation to kiss? Which?

These aren't feelings, these are behaviors – which is OK from a behaviorist perspective I guess, but not from the simplest of human perspectives which kinda focuses on feelings very much.

If you strip feelings from their causes and effects, what does remain? It all leads to the old philosophical question of whether perfect imitation is the thing or not, current frontline is the p-zombie, which IMHO is just pure nonsense.

AI is more about understanding how the mind works. Copying an entire human is mostly uninteresting, but copying some of the capabilities of the brain is not. Feelings, for example, are not too interesting, they are just our gene's regulating mechanism to control the computing we do. But the fact that we are capable of building mental models from unrelated previous experience in a short time and from a small amount of data is really fascinating and mysterious.

And AI is about power, too. A machine just somewhat 'smarter' (whatever it means) than Neumann or Einstein or Perelman can do things which cannot be done by anybody else, probably ever. And that will have an impact, just like these people had.

Feelings are perhaps not too interesting if you're programming a computer or commenting on an article, but very interesting if you drop a brick on your foot.

Not all of life is confined to logic, mathematics, philosophy, physics, technology, thinking and talking.

I agree on that, but AI is not aiming to reproduce the entire human life. It is aiming to reproduce a thing which can experience the entire human life. The two is not the same.

You are mixing explanation levels here. Denying strong AI because of feelings does not work. It is like trying to prove a mathematical theorem with a poem.

Well, you're mixing two different claims here. I didn't deny strong AI because of feelings – I agree it would make a problematic argument – but because of its implausibility. There's a bet over this one, Kapor vs Kurzweil – we'll see how that works out.

(Although arguably strong AI is as much or more about poetry than math or science – Turing used the ability to write sonnets and then explain why they're written thus and not otherwise as an example of what it means to be intelligent; but there is no need to argue over small details here – it's like arguing over the fine points of what communism means, totally unnecessary when making the claim that we won't see a society where each gets according to his needs and gives according to his abilities in the next 50 years.)

There is something between human like AI and current AI research, that we don't currently do, that would be useful.

We can't currently make a computer that looks after itself, there is always a human root or admin somewhere that is supposed to be looking after the computer. Someone updating the software, installing new stuff and removing malware. We get things like botnets because that assumption is not true.

This is very different from what people traditionally think AI is about. But it is a useful trait of evolved creatures that we are missing from computers. Animals manage to improve their skills without a super user. We haven't really explored how we might make a computer look after itself and what the limitations are. Autonomic Computing is the closest but it is generally oriented towards maintaining a known system state.

I think a key non-technical reason to keep things the way they are, with a human in the loop, is to make sure that things evolve in accordance to the changing perceived purposes of humans rather than the changing perceived purposes of machines.

I agree that we will want to keep humans in the loop, as it were, but I think there are ways of doing that while not assuming that humans will be able pick the things to change or to understand all the changes made. But a human should be able to evaluate the resulting changes.

So having humans as more CEOs of computers rather than as engineers. Or another way to think about it, if we understand how the brain works there is no reason that we have to duplicate it in its entirety. We could make the bits that don't deal with goals (such as the sensory systems) and use humans to provide the goal orientation information (probably slowly to start with due to lack of bandwidth, but with the potential for more integration as we improve our ability to integrate with brains). I suppose exo-brain add ons is what I am thinking of.

I think the key non-technical reason no one is pushing for this is because it is hard, and everyone in computing can make sufficient money with the status quo.

Economically, the fact that the status quo is profitable to existing vendors does nothing to stop a new vendor from pushing this and then eating into their market share.

As to CEOs of computers – let's say that it works badly enough with humans (a rogue employee can do a whole lot of damage to his employer; for that matter, a CEO is himself an employee of the shareholders and not infrequently wipes them out for his own gain).

I meant under the status quo, new vendors/researchers can find new markets for extensions of the existing computing/AI paradigm and be vastly profitable a lot more easily than they can pushing something speculative like self-upgrading computers.

Indeed CEOs of computers will go wrong, the question is would it be worse than the current botnet/phishing identity theft stuff we have by assuming that computers have engineers to look after them.

As we don't know how they will work it is hard to say.

i am study sybca

I don't get the pain argument. As far as I ever thought of pain, it really was only a signal to indicate damage, and to limit activity that worsens damage. What more are you implying pain is? Your explanations (comment 8, etc) are vague.

@JeramieH: it's not a logical or a scientific argument. I'm just saying that pain is a feeling – that there's a fundamental difference between an SMS telling you "damage detected, limit your current activity" and actual pain; that the message can be represented and studied as a mathematical object, but pain can not be; and that everybody understands this at the gut feeling level – that pain is not a number though an image file is, which is why I use pain as an example and not, say, eyesight.

Some people dismiss feelings as "illusion" on the basis of, for instance, our ability to drug ourselves into a state where pain is not felt. I think it's about the same as saying that life does not exist because you can club someone to death – and in fact that's a view some people subscribe to (that there's no fundamental difference between a living human, his dead body and a pile of dirt).

It's fine as far as logic is concerned – feelings are impossible to objectively observe and formalize and thus do not create a problem for any sort of logical argument, being outside its realm. I think one's worldview shouldn't be restricted to things accessible to logic simply because only a tiny minority of things are, but it's of course a subjective position.

However, there are exactly zero people who actually subscribe to the life-and-death-are-the-same worldview as evidenced by their actions. Even psychopaths tend to inflict pain on the living and do not derive equal pleasure from "hurting machines".

good

I think Qualia (http://en.wikipedia.org/wiki/Qualia) is a word you would like to check out.

Qualia I know about. But what is bfrs? Brominated flame retardants?

bfrs is now sort of like xkcd, meaningless. It was the first handle I could think of when Hacker News asked for one. At that time, it stood for Ben-Franklin:Richard-Saunders (Franklin's pen name). Ben Franklin (for reasons now forgotten) is the mascot of my passport photo web app (http://freepassportphoto.dyndns.org/).

Some comments on stuff I found interesting in your post:

"...I want to pay a tribute to the best movie of the previous century, which, among other notable achievements, provided the most comprehensive treatment of the psychophysical problem in the popular culture. That motion picture is of course The Terminator, part I and, to an extent, part II."

People called me crazy for expressing the same opinion. I lost count of how many times I watched Terminator – part I. Its what got me interested in AI. Couple of years back, it got knocked off from the top position by Carl Sagan's 1997 "Contact". When I was a kid, the strange code listings seemed impressive, now I know that such low level atari assembly code would never surface to the top level. If at all any code did come to the top level, it would be lisp. Unfortunately lisp, as Larry Wall put it, "has all the visual appeal of oatmeal with fingernail clippings mixed in", unlike the orderly and imposing looking assembly.
———-

"...To pass the Turing test, a machine needs to have first-hand experience of operating a human body with all of its elaborate hardware."

This is well appreciated by AI "weenies" today. For example see Philip Greenspun and Ellen Spertus' male-female Turing test (http://philip.greenspun.com/software/watson).
———-

"...Moving on to our next problem, we meet the people who think that we actually need AI. We don’t. Not if it means “a system that is supposed to scale so that it could pass the Turing test”. And this is the only thing AI means as far as I’m concerned here...So the only thing we don’t have is artificial intelligence that simulates our own. And this is as tremendously useless as it is infeasible. Natural intelligence as we know it is a property of a person. Who needs an artificial person?...If you want a kid to raise, you can make one in a fairly reliable and well-known way."

It looks like you are overlooking the possibility of an "intelligence explosion". A frequent oversight that prompted I. J. Good (one of Turing's colleagues at Bletchey Park) to write his famous essay: "Speculations Concerning the First Ultraintelligent Machine"
(http://web.archive.org/web/20010527181244/http://www.aeiveos.com/~bradbury/Authors/Computing/Good-IJ/SCtFUM.html#6_CellAssembliesAndSubassemblies)
"...highly intelligent people can overlook the intelligence-explosion. They say that there is no point in building a machine with the intelligence of a man, since it is easier to construct human brains by the usual method...Let an ultra-intelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultra-intelligent machine could design even better machines; there would then unquestionably be an "intelligence explosion," and the intelligence of man would be left far behind. Thus the first ultra-intelligent machine is the last invention that man need ever make."
———-

Your irreverance gave me the lulz. Calling the formidable T-800/Arnold "Atari-powered-gadget", and AI heavyweights, Hofstadter and Kurzweil as AI weenies, I find very funny! Thanks for the lulz.

I also like Phil Greenspun's angle on AI there; he isn't what I'd call an AI weenie – he's very sensible about the whole thing.

Regarding intelligence explosion: I can believe it in a limited context – where you have a machine (program) living in a formal universe (of theorems or 3D models or whatever) and this machine gets better and better in its domain. I don't believe this sort of thing will lead to a Turing-test-passing AI, for reasons that I mostly discussed above.

philg, ofcourse, is no AI weenie. Some of his best satire is directed towards AI (http://philip.greenspun.com/humor/ai.text). What I should have added is that, philg being popular, people are aware of the man-woman Turing test etc.

Today's AI heavyweights, Hawkins, Kurzweil, Ng, Norvig, Thrun et al., don't consider the goal of AI to be to build a machine that can pass as a human. Instead the goal is to replicate the capabilities displayed by the cortex. The core idea goes by various names, "Mountcastle's hypothesis" being a popular one.

By the way, I found your blog while trying to decide if I should write my new project (a streaming X11 screen grabber) in C++. I started with x11vnc code, which is pure C89, and kept thinking to myself, "what a pain C is, must define variables at the beginning of scope, can't write type-agnostic functions, no OOP enabled abstraction facilities, etc. etc." However, I know C++ only in a half-assed way (like most people?), and if I did choose it, there was a steep learning curve to contend with. I think your FQA has scared me enough to stay the hell away from C++. As you put it, "an itch here or a scratch there is much preferable to having your legs blown off!". My decision, hopefully for the better, is to stick with my usual Python/C combo.

Hope it works out well to you... IMO the good reason to use C++ is if there are great libraries in it doing what you want and the interfaces are too big to wrap.

I would recommend a couple pounds of Daniel Dennett, that should be able to start your healing process...

This is awesome. Really. I didn't immediately attempt the YKMEE but I did laugh out loud after parsing the word 'damage'. I tend to hastily duck and declare that pain is an event, which solves nothing– the problem simply becomes "is an event information?" My favorite answer, for the next thirty seconds, is "Snow".

Well, keep not attempting it; it's not nearly as enjoyable when attempted.

Great article. I guess I'm considered an AI researcher, but I agree that it's pointless to consider the philosophical problems associated with "strong AI." To me, AI is about getting machines to solve harder problems.

About Terminator: I've always found it weird that the Terminator has textual information overlayed on its camera feed like a HUD (HUMAN CASUALTIES: 0.0). This means that its computer takes machine readable data, renders it as images of text, integrates it with the camera data, then uses OCD to translate it back into machine readable data. Maybe this is the Terminator's version of the internal dialogues going on in our heads.

I guess the assembly listings and HUMAN CASUALTIES is how they communicate to the audience "it's a robot".

I agree with some stuff you say. But I don't agree that AI is useless. Indeed, you almost admit it would be very nice to have, when you talk about translation. Can you seriously deny that having machine translation is very useful? Sure, you may say we will never get it, but that doesn't make it useless.

Well, now people are writing AI which can write AI, True AI soon :)

Great writing on the topic. Insightful and humane. Thank you for spending the time and effort to write this piece.

Post a comment

Author:

Site (optional):

Comment (Markdown):

Preview Post