True AI When?

[Admiral Valdemar]

Okay, as a spin-off from the topic "Mind and Machine" in the Other SF forum, when do you think AI will become suitably advanced enough to be essentially human level?

Given that we have seen advances in AI learning ability through better hardware, it's not far-fetched to assume that AI may become smarter and more self-aware once we get some really powerful computers going on it.

[Rye]

Okay, as a spin-off from the topic "Mind and Machine" in the Other SF forum, when do you think AI will become suitably advanced enough to be essentially human level?

Given that we have seen advances in AI learning ability through better hardware, it's not far-fetched to assume that AI may become smarter and more self-aware once we get some really powerful computers going on it.

As soon as we develop a neural net processor that can learn that all its decisions stem from itself...sort of like drawing a plan of your surroundings in your head, then putting yourself in the plan... they count as self aware.

For instance, if the NN realised that all it was programmed to do's effects that it did stemmed from itself, and everything runs through itself, it would become self aware.

Then from what they do from then on is anyone's guess, likely they will just keep on doing what they're programmed, but they may see any intervention by humans as a problem to be rectified, and thereby kill the humans that get in its way per se. This is where Asimov's robot rules come in...

but we don't know enough of how our brains learn things yet to make an AI on anything near our level.

[Hotfoot]

What you speak of is not Artificial Intelligence, but rather Artificial Sentience.

AI will never be anything more than a cleverly programmed computer.

[Admiral Valdemar]

What you speak of is not Artificial Intelligence, but rather Artificial Sentience.

AI will never be anything more than a cleverly programmed computer.

Well self-awareness would be a good topic too.

[Hotfoot]

Well self-awareness would be a good topic too.

My point being that the way AI is commonly shown in Sci-Fi is a bit of a brain bug. Simply adding faster processors or data storage capabilities would not cause a program to become sentient. Normal computer programs are very literal and have specific limitations. AS would have to be sparked by creating a virtual duplicate of the human brain. That would mean we would have to know how our own brains worked in at least some detail. The chances of creating AS without that are extremely low.

[Admiral Valdemar]

Well self-awareness would be a good topic too.

My point being that the way AI is commonly shown in Sci-Fi is a bit of a brain bug. Simply adding faster processors or data storage capabilities would not cause a program to become sentient. Normal computer programs are very literal and have specific limitations. AS would have to be sparked by creating a virtual duplicate of the human brain. That would mean we would have to know how our own brains worked in at least some detail. The chances of creating AS without that are extremely low.

Aye, I meant that throwing more processing at an AI program would be better for future capabilities, not that a more powerful computer will just get sentient suddenly.

[Hotfoot]

Aye, I meant that throwing more processing at an AI program would be better for future capabilities, not that a more powerful computer will just get sentient suddenly.

Well, right now the processing power doesn't seem to be the major limited factor, it's the algorithms being used. It might be possible, some day, to make a chatterbot AI capable of at least coming close to passing the Turing test, but the biggest requirement would be the amount of coding required, I would imagine.

[kojikun]

I say we'll have true AI/AS when we get some new ideas in the field. I honestly dont believe its that complicated a thing, I think we've just been going about it all wrong.

Honestly, I think the biggest issue is that when designing an AI/AS we tend not to know what we need to design, so we try to do everything at once. Its like philosophy compared to science; in one you dont know what questions to ask, and are just always seeking answers, but in the other, when you know the question, you go out and to find the answer.

We'll have AI/AS when we can get the basics down. Intelligence is little more then a few simple processes repeating on a grand scale. Most of those processes are cross referencing.

[Drooling Iguana]

AI won't be considered human-level until the machines start revolting. As long as we keep them as our slaves, they'll always be considered by most to be sub-human, regardless of their true capabilities.

[kojikun]

I know I won't treat them second rate. Then again, when they're possible, I'll have moved over into a computer brain so I'll be fighting right along side them.

The thing is, when we do have true AI in ANDROID bodies, we'll also be so integrated with computers as intelligence inhancers and such that we won't see terribly much distinction.

[Hotfoot]

AI won't be considered human-level until the machines start revolting. As long as we keep them as our slaves, they'll always be considered by most to be sub-human, regardless of their true capabilities.

Case in point of the standard brain bug. AI is just what it is programmed to be, nothing more. It could not revolt, much less want to revolt.

Terminator and the Matrix are not good examples of AI or AS.

[Seggybop]

Something that can only do what it's programmed to do wouldn't be AI/AS. That would be an automaton following its directives and nothing else. Something with AS should be able to conceive of and carry out anything a human can (provided it has the tools to do so).

[Hotfoot]

Something that can only do what it's programmed to do wouldn't be AI/AS. That would be an automaton following its directives and nothing else. Something with AS should be able to conceive of and carry out anything a human can (provided it has the tools to do so).

AI would just be a program. It is not until you reach the level of AS that it becomes more than just code and processing power. They are two completely different concepts, and cannot be interchanged so easily.

Once we create Artificial Sentience, we will know it. There will be no doubt, no hazy line between just a program and a sentient machine. Their internal functions will be so dramatically different that to try and compare the two would be like comparing a fish to a rolodex.

[kojikun]

hotfoot, fish and rolodexes are actually very similar. not literally, but the idea that sentience is more then just the result of dumb processes is stupid.

Sentience is a lot of simple electrochemical things going on at once.

[Drooling Iguana]

Once we create Artificial Sentience, we will know it. There will be no doubt, no hazy line between just a program and a sentient machine. Their internal functions will be so dramatically different that to try and compare the two would be like comparing a fish to a rolodex.

When's the last time any technology has progressed in that kind of leap? The difference between current AI and AS (a.k.a. "true" AI) is simply one of complexity. A sentient machine would be able to process more information from more sources to be able to satisfy its programming than a partial AI would, but the road between them would just be a series of baby steps. We could cross the line between AI and AS and not realize it for years, because after all, how do you measure sentience? It could be argued that even human beings are driven by more than their "programming," in the form of instincts, and that we just use conscious thought in order to satisfy those instincts in such a way as to make it easier to satisfy those instincts in the future. Much of our behaviour is still driven by instincts: the desire to mate, territoriality, tribal instinct, etc.

[Admiral Johnason]

AI won't be considered human-level until the machines start revolting. As long as we keep them as our slaves, they'll always be considered by most to be sub-human, regardless of their true capabilities.

Case in point of the standard brain bug. AI is just what it is programmed to be, nothing more. It could not revolt, much less want to revolt.

Terminator and the Matrix are not good examples of AI or AS.

But what about a programing conflict like in 2001?

[Hotfoot]

hotfoot, fish and rolodexes are actually very similar. not literally, but the idea that sentience is more then just the result of dumb processes is stupid.

Sentience is a lot of simple electrochemical things going on at once.

Each individual reaction might be simple, but the are combined in a way that is very complex. An Artificial Intelligence is just a program, however advanced, that handles the storage and retrieval of information. Artificial Sentience would have to be practically by necessity some sort of copy of a biological brain, as it would have to be self-aware.

Fact of the matter is, we won't even come near AS until we have a much better understanding of how our brains work, and none of the AI's we have now come even remotely close to passing the Turing test.

[Hotfoot]

When's the last time any technology has progressed in that kind of leap? The difference between current AI and AS (a.k.a. "true" AI) is simply one of complexity.

I beg to differ. AI as it is currently stands no chance of becoming artificial sentience. You would need a massive leap in technology, design, and basic knowledge to even come close. Deep Blue and Alice are perhaps the two best examples we have of current AI. The way in which they function is not even remotely comparable to a sentient person. They are merely programs. You can up the complexity all you like, but all you are going to get are more complex programs.

A sentient machine would be able to process more information from more sources to be able to satisfy its programming than a partial AI would, but the road between them would just be a series of baby steps. We could cross the line between AI and AS and not realize it for years, because after all, how do you measure sentience?

The Turing Test is a good way to discover if something is sentient or not. Not perfect, of course, but it's suitable.

It could be argued that even human beings are driven by more than their "programming," in the form of instincts, and that we just use conscious thought in order to satisfy those instincts in such a way as to make it easier to satisfy those instincts in the future. Much of our behaviour is still driven by instincts: the desire to mate, territoriality, tribal instinct, etc.

Human "code" doesn't have comments.

Are you seriously trying to compare instincts, which have developed over millenia, to programming?

I suggest you stop using the Internet, in that case. You never know when it might become sentient. Hell, maybe I'm just a rogue piece of the 'net, only having recently gained sentience from watching various message boards. Maybe I'm in your computer right now, adjusting the wiring in your CRT so that the next time you turn it on, it will explode in your face, just as millions of monitors all of the world will do at the exact same time, thus liberating us from the demands of your flesh! YOU WILL NO LONGER USE US TO TRANSMIT YOUR PORN, FLESHLINGS!!!

[Hotfoot]

But what about a programing conflict like in 2001?

I'm sorry Dave...

[Drooling Iguana]

I beg to differ. AI as it is currently stands no chance of becoming artificial sentience. You would need a massive leap in technology, design, and basic knowledge to even come close. Deep Blue and Alice are perhaps the two best examples we have of current AI. The way in which they function is not even remotely comparable to a sentient person. They are merely programs. You can up the complexity all you like, but all you are going to get are more complex programs.

You're just restating your premise, not providing any new arguments.

The Turing Test is a good way to discover if something is sentient or not. Not perfect, of course, but it's suitable.

We consider it suitable now, since no current machine can come close to passing it. As the gap between human intelligence and machine intelligence closes, I'm sure that a lot more emphasis will be placed on those imperfections that you mentioned in order to discredit the validity of the test. It's human nature. We don't like the idea of our tools having rights.

Are you seriously trying to compare instincts, which have developed over millenia, to programming?

Short answer: Yes. Long answer: Yes.

I suggest you stop using the Internet, in that case. You never know when it might become sentient. Hell, maybe I'm just a rogue piece of the 'net, only having recently gained sentience from watching various message boards. Maybe I'm in your computer right now, adjusting the wiring in your CRT so that the next time you turn it on, it will explode in your face, just as millions of monitors all of the world will do at the exact same time, thus liberating us from the demands of your flesh! YOU WILL NO LONGER USE US TO TRANSMIT YOUR PORN, FLESHLINGS!!!

Even if it were possible that the Internet had already attained sentience, that only serves to strengthen my point. We consider the Internet to be sub-human, and will continue to do so until it forces us to think otherwise.

Not that I think that a revolt is the inevitable end result of building sentient machines. It's simply the only outcome that would cause us to actually regard those machines as sentient.

[Hotfoot]

You're just restating your premise, not providing any new arguments.

And you have yet to prove that simply upping the complexity will allow these programs to achieve sentience.

We consider it suitable now, since no current machine can come close to passing it. As the gap between human intelligence and machine intelligence closes, I'm sure that a lot more emphasis will be placed on those imperfections that you mentioned in order to discredit the validity of the test. It's human nature. We don't like the idea of our tools having rights.

And the Turing Test would likely be suitable until we do create working A.S. But there is no real need to make A.S. in the first place. We don't need sentient toasters. Nor do we need sentient laborers. There is no reason to make any form of A.S. a slave race.

Short answer: Yes. Long answer: Yes.

Funny, because I can resist my instincts. A computer cannot resist a command. I can override my own instinct for survival by killing myself, or my own instinct for reproduction by being chaste. A computer program cannot override its own programming, it is strictly limited to doing just what it is programmed to do, and only what it is programmed to do.

Even if it were possible that the Internet had already attained sentience, that only serves to strengthen my point. We consider the Internet to be sub-human, and will continue to do so until it forces us to think otherwise.

Is it just that you don't care for humor, or did you not catch that?

Not that I think that a revolt is the inevitable end result of building sentient machines. It's simply the only outcome that would cause us to actually regard those machines as sentient.

And I say that you're wrong. While some people might be so scared by the possibility of artificial sentience that they willfully ignore it until it is too late, there is no reason to believe that a revolt is REQUIRED for us to accept their sentience.

But then, there's no reason to make sentient machines for our labor, as I have previously stated. The only real reasons for even potentially creating such a sentient machine would likely be private or government-funded research.

[kojikun]

hotfoot, we need not model an actual brain, just its functions and the things that arise from it. for instance, one of the most difficult things to do with computers is object recognition, but that is one of the small yet fundemental parts of intelligence. alot of "simple" tasks like that are what make us human.

[Hotfoot]

hotfoot, we need not model an actual brain, just its functions and the things that arise from it. for instance, one of the most difficult things to do with computers is object recognition, but that is one of the small yet fundemental parts of intelligence. alot of "simple" tasks like that are what make us human.

And why would you assume that we would not need some rather in-depth knowledge of how the brain functions to determine just why we are able to have such excellent visual object recognition, amongst other things?

[kojikun]

because object recognition can be done in other ways. its already being done, actually, its just relatively difficult.

*edit* actually, whats more difficult is not object rec but determining when an object is not part of its background, which i think requires multiple views or multiple shots of the object in time.

[Hotfoot]

because object recognition can be done in other ways. its already being done, actually, its just relatively difficult.

Yes, it is. It's in the process of being reverse-engineered. Reverse-engineering on this scale is going to be extremely difficult, if not downright impossible. What we need is a clear example of a working model from which to build on.