Technology of the future

[ZiP]

There'll finally be the swishing doors like on star trek!

[AnsemK_R]

Don't they already have that?

[Doubleshadow]

It's a little late, but I also loved that pic. you found Technomancer.

[Fsiphskilm]

They've already

[ClosetOtaku]

They've already GOt Robots that can create Virtual Worlds, from looking at rooms and hallways. I've seen highschools building them, Artificial Intelligence and everything.

Since it's part of my professional field, I feel comfortable saying without hesitation that Artificial Intelligence, as it was originally conceived and usually defined, has yet to be achieved.

Of course, Artificial Intelligence as a marketing term is bandied about like it's been around for years. That's Marketing, which I've always viewed as a euphemism for "an impressive set of believable lies".

As an example: open your kitchen cabinet, the one with all the drinking vessels in them. You'll probably see: cups; mugs; glasses. Maybe you'll see tumblers and snifters and steins (oh my!). But there's probably little doubt you can tell the difference (if you know the difference) - and, more importantly, you say, "These are the things I drink liquids out of".

We've learned that for an AI routine to learn this on its own is akin to asking it to build a rocket ship to fly to the moon: it can't be done. You can't give a routine some general rules and expect it to extrapolate that it wouldn't drink out of, say, a tire hubcap. And, if you told the average computer routine not to drink out of a Cadillac tire hubcap, it would go and try it with one from a Nissan. You'd know better not to; more interestingly, a child would know better. How do we make these logical leaps? Answer: we don't know. As long as we don't know, we can't program it directly. If we program it directly, it isn't Artificial Intelligence. If we can't make AI draw the same conclusions from a limited data set like humans do all the time, it isn't AI.

So... we're still far away from AI, and many professionals have given up any immediate hope that we're going there. We've tried expert systems, complex algorithms, recursive programs in LISP, neural networks (now there's a parlor trick if there ever was one)... nothing seems to work. We may not have the fundamental knowledge we need to make that happen yet. Someday, we might.

[Technomancer]

Since it's part of my professional field, I feel comfortable saying without hesitation that Artificial Intelligence, as it was originally conceived and usually defined, has yet to be achieved.

You do robotics/AI? What sort of development do your work on?

So... we're still far away from AI, and many professionals have given up any immediate hope that we're going there. We've tried expert systems, complex algorithms, recursive programs in LISP, neural networks (now there's a parlor trick if there ever was one)... nothing seems to work. We may not have the fundamental knowledge we need to make that happen yet. Someday, we might.

No question, true AI is something that has been beyond us so far. But it's not all bad, since all of these areas do have worthwhile (if much more limited application). Neural networks have been especially useful both for industrial applications and for gaining practical neurobiological insights. Slow progress, but steady (unless Roger Penrose be proven right someday!).

It's a little late, but I also loved that pic. you found Technomancer.

What pic? I don't remember psting one in this thread.

[shooraijin]

I think they meant termyt. Unfortunately, that picture is a well-known hoax.

http://www.snopes.com/inboxer/hoaxes/computer.asp

It's still pretty amusing, though. ^^

[Doubleshadow]

Gah! I did it again! Nice pic. termyt. Sorry Technomancer.
And it's a hoax? Daggone it. 0 for 2.

[Fsiphskilm]

AI is all around u

[ClosetOtaku]

You do robotics/AI? What sort of development do your work on?

I do Informatics - especially user interface design, trying to figure out how users will tolerate data input and output modalities, both at the novice and expert level. Much of my AI work is therefore in speech recognition and natural language processing for medical facilities. I did some AI work in college (Carnegie-Mellon U, home of the Robotics Institute), but that was theoretical]No question, true AI is something that has been beyond us so far. But it's not all bad, since all of these areas do have worthwhile (if much more limited application). Neural networks have been especially useful both for industrial applications and for gaining practical neurobiological insights. Slow progress, but steady (unless Roger Penrose be proven right someday!).[/quote]

Not saying we shouldn't, and not saying that progress isn't being made, but all of this talk about AI is, in my mind, premature. Or maybe I'm being too picky over a phrase that has had its definition changed right out in under it; call me a purist.

I'm a skeptic about neural networks, especially as it pertains to the functioning of the mysterious "hidden layer"; to me, it is little more than an expert system where we didn't bother specifying the underlying program parameters and we're giving the output more credit than it is due. It's the cyber equivalent of a rat trained to do tricks: we can attribute intelligence, but any similarity between intelligence and what is coming from the rat is purely coincidental. But, I have to admit, it's pretty cool when it works right.

AI is all around us... in everything from robots to videogames.

Too bad currently it's nothing but a bunch of "if___ then___" statements.

Once upon a time, this would have qualified as AI, then it wouldn't have qualified as AI, now it does qualify as AI. The language changes, I suppose.

I'm just being harsh because deep down inside, no matter how unemotional something is... i still give it human characteristics, This one time i droped a rock on the floor and I said "i'm so sorry" to it. ... ... ... poor rock.

Anthropomorphizing. See my rat comment. I'm beginning to think it's a heuristic humans are all too vulnerable to... or are we designed that way?

[Yeshua-Knight]

well, just keep in mind that machinery is just that, machinery, when something in software goes wrong it's because at some point somewhere a human messed up, either in using the software or in the development of the software, so for a machine to attain true ai would only be a bad thing as far as the emotion factor as well as the fact that in some cases things aren't just black and white, there are mitigating circumstances and that kind of thing, however it's when a human can control a sophisticated piece of machinery that i am truly on my guard, you'd need some sort of killswitch or magnetic grenade or something to kill one of those, oh well, just the ranting of someone that reads too much popular science and popular mechanics i guess

'nuff said

[Fsiphskilm]

Yes we're all afrai

[Technomancer]

I do Informatics - especially user interface design, trying to figure out how users will tolerate data input and output modalities, both at the novice and expert level. Much of my AI work is therefore in speech recognition and natural language processing for medical facilities. I did some AI work in college (Carnegie-Mellon U, home of the Robotics Institute), but that was theoretical]

Cool, you probably already know, but my own area of research is neural networks and signal processing.

I'm a skeptic about neural networks, especially as it pertains to the functioning of the mysterious "hidden layer"; to me, it is little more than an expert system where we didn't bother specifying the underlying program parameters and we're giving the output more credit than it is due. It's the cyber equivalent of a rat trained to do tricks: we can attribute intelligence, but any similarity between intelligence and what is coming from the rat is purely coincidental. But, I have to admit, it's pretty cool when it works right.

I suppose expert systems are one way of viewing the hidden layer, but there is a more disciplined approach. It's more useful I think to picture the action of the hidden layer in terms of a form of non-linear principal components analysis. This means that the network is learning statistical associations, and very importantly higher-order associations at that. This incorporation of higher-order statsistics also ensures that the network can cope with non-gaussian data, a real bugaboo for signal processors. Such problems are often intractable by classical means. Where possible, of course a priori information should be incorporated in order to simplify the problem. However, that information may not always be available to us.

I really wouldn't call connectionist models "intelligent" in any real way. While they have proven their worth in accurately emulating some neurobiological behaviours, they have only done so on the level of individual systems and not beyond that. As for whether an integrative model could be called intelligent, I don't know; noone has the computational power to try such an experiment. In any event, there are still too many unknowns in brain function for such a model to be accurate.