AI and computers

[NotTheCHEAT]

In the Creation vs. Evolution thread we started talking about the possibility of the Universe being simulated by a computer.

This got me to thinking about how a computer might simulate a single human.

Lately we have been hearing a lot about Artificial Intelligence in the news. It's the next big thing. It's hip, it's cool, it's amazing- but is it possible? The current trend is toward using computers to act as humans, or artificially intelligent things. While I admit that there have been some pretty amazing things, such as Deep Blue, which beat a human player at chess, I would not call this artificial intelligence. It's still pure logic.

Computers operate on pure logic. Everything is either true or false. You do not have maybe, or in-between, only Yes or No. Numbers can be represented with Yes-or-No, but it's still only logic. Humans do not operate on pure logic. We have creativity, imagination, conscience, morals, values, ideas, etc.
True, computers can be programmed to simulate these things, but I have a hard time believing that a computer can simulate a human. Remember, computers must be programmed. And the programmer, who operates not on pure logic but also on fuzzy logic, will put in there whatever they want. Maybe the programmer likes spicy foods. So he will program the computer to "like" spicy foods to (assuming that the computer has an eating system of some sort, whether physical or virual). So, an AI program usually is based on the human who programmed it, even if imperceptibly. To write a true fuzzy logic program which does not carry the traits of the human programmer, it would have to be written by a computer, and that computer would have to be programmed by a fuzzy-logic human... and so on. So it's impossible to write a "perfect" program which has no recognizable traits of the human who wrote it. Two humans will write a complex program which performs the same function, differently.

Then too is the fact that there is no clear correspondence between certain parts of the brain and computer parts. Computers don't have millions of neurons, they have one CPU. A neuron could be thought of as a simple CPU, so maybe if we want our computers to be similar to humans they should have millions of simple CPU's instead of a single complex CPU. Multi-threading, multi-programming, and all of these things will not do. Neurons do not follow programs either, they communicate with each other. Memory and Disk roughly correspond to short-term and long-term memory, but only roughly. And these should also be made of "neurons", or in other words simple CPUs.

If we want to make a human-computer, we should IMVHO be working at it from a different perspective than the perspective we are currently working at it from. I don't think that normal computers (except maybe supercomputers) that exist today are suitable for this type of application. I foresee great advances in AI, but I don't think we will ever have a human-lioke machine at this rate, or at all.

Anyone have a rebuttal?

[AGI1122]

A while back I had placed some code into my forum that implimented an A.I. that responds to people's posts as a user named "Bob".

Cool thing is it learns as you talk to it.

You can talk to it here by replying or starting a topic:
http://cbb.agigames.com/index.php?actio ... ex&board=9

I definatly think A.I. is fully possible, while this bot may not be as smart as you and me. I do think it's about as smart as some of the 12 year old AOL kids that can't talk very well.

[AGI1122]

I fixed your post NotTheCHEAT, html isn't allowed in the forum. You have to use BBC instead. This is the code to make a url:

Code: Select all

[url=http://www.site.com]blah[/url]

Oh and I noticed guests couldn't post in the test board for Bob there, I changed it so guests can reply and start topics in the test board now.

[Crazed123]

Theoretically, a genetic algorithm for fuzzy-logic AI (possibly neural network based) could be written and evolved to try and approximate animal intelligence.

[Ushma]

I would not call this artificial intelligence. It's still pure logic.

So are you, you're just very very complex.

Humans do not operate on pure logic. We have creativity, imagination, conscience, morals, values, ideas, etc.

Yes they do. It's just very very complex logic with a lot of inputs.

True, computers can be programmed to simulate these things, but I have a hard time believing that a computer can simulate a human.

Humans being lots of molecules, atoms, etc. following physical laws (or on a higher level, neurons following rules), I don't have a hard time believing that. I have a hard time understanding though why anybody'd want to. What's far more useful is to make a program that can handle requests in a somewhat intelligent way to make it more useful to humans. We do a fine job of simulating humans, having a general-purpose computer do it is a waste of cycles/time/etc.

Then too is the fact that there is no clear correspondence between certain parts of the brain and computer parts. Computers don't have millions of neurons, they have one CPU. A neuron could be thought of as a simple CPU, so maybe if we want our computers to be similar to humans they should have millions of simple CPU's instead of a single complex CPU. Multi-threading, multi-programming, and all of these things will not do. Neurons do not follow programs either, they communicate with each other. Memory and Disk roughly correspond to short-term and long-term memory, but only roughly. And these should also be made of "neurons", or in other words simple CPUs.

This is true, our brain is a very specializing processing system; the computers we have built are general-purpose machines.

There's no need to build neurons; you would be simulating them on top of a general purpose machine. The simulation would cause them to behave as they should.

If we want to make a human-computer, we should IMVHO be working at it from a different perspective than the perspective we are currently working at it from. I don't think that normal computers (except maybe supercomputers) that exist today are suitable for this type of application. I foresee great advances in AI, but I don't think we will ever have a human-lioke machine at this rate, or at all.

I don't see why we'd *want* a human-computer. The way we are working on it now is appropriate to what we want: programs that can respond in somewhat intelligent ways to requests to make themselves more useful to human users.

[AR]

The yes/no thing is not really correct, take audio for example, audio is analog (in waves), how could a computer possibly store it? Simple - sampling. Sampling is where you can't store the exact pattern so you approximate it by just recording the frequency every X milliseconds, it would be plausible to do the same thing with intelligence/conciousness.

It is important to distinguish between intelligence and conciousness/sentience. A computer program which can adapt by rewriting itself could be considered sentient (adaption has a prerequsite of self-awareness, don't ask me how to program that, and computers have always been intelligent which easily fulfills that criteria).

So really all it comes down to by current criteria is understanding what constitutes self-awareness and writing a program to emulate it, give that program the ability to understand basic information (ie. read English) and rewrite itself and then it's "sentient". (It could then trawl the internet rewriting itself to include new information, find a book on learning other languages and rewrite itself to understand those and so on). Alot of people seem to what to simulate a human mind, I don't see why you'd want to being as inefficient as it is.

[Freanan]

I actually agree with all in ushma's post.
"Consciousness" and so on are all somehow "simulated" by different states of neurons, maybe there is also some hormon chemistry in the mix, but all that should be simulatable.

I would like to add that there are two different kinds of AI:
The kind that emulates intelligent behaviour (logical programming, chess-computers, computer game ai, chat-bots...)
and the kind that simulates the actual physical, chemical and biological processes that drive intelligent behaviour (neural nets, genetical programming...).
Of course they are both just logic (as we are - logic driven by the laws of physics), but i think it is not hard to imagine that kind 2 can simulate intelligent beings like humans, given the needed computing power (Or maybe implementing it directly in special-purpose neural net hardware and the like).

I think really implementing the second kind of ai is not very interresting (except for research) because it is ineffieciently designed (as our brain is).
I mean... the human brain might use it's low level circuits to perform arithmetics on fast lowlevel, just like a computer does.
Instead the human brain "imagines" numbers like "two apples and five apples... =seven apples", needing thousands of lowlevel math operations to perform the imagination of the numbers, while actually one of them would suffice.

The first kind is "not real ai" but it may be much more usefull (imagine a bot like "the librarian" from neal stephenson's snowcrash - that would be a really great research tool - and it is not THAT utopian).

[distantvoices]

Hm ... I daresay our brain isn't inefficient. It's the users who use the brainpower inefficiently.

Imagine a computer handling/interpreting 3d vision (with triangulation), 3d audio parsing, matching shapes of objects to stored samples/classes of objects ...

Sorry, no cope. Human understanding of the world mostly bases on abstraction. We don't need the exact description of a willow just to know that it's a tree, once we 've saved the correct abstraction.

Teach this a computer. Then we talk about number crunching.

[AGI1122]

Darn looks like Bob broke... now all he does is post smilies. I will have to fix him later.

[NotTheCHEAT]

Yes, the human mind is based on pure logic at the low level, but it's masked out heavily at the upper level. In essence, you would have to dig through a ton of creative and random things before reaching the underlying "pure logic" down there.

Also, there are random elements based on what we see, hear, past experiences, etc. Computers have to simulate randomness based on the timer, CPU speed, and other less-than-random phenomena.

Yes, it might be possible to simulate a human brain. But it wouldn't be very practical, and I don't think that the way they are trying to do it will work the way they want it to.

Someone once said it would take a building the size of the empire state building full of wires and circuits to simulate the human brain. With nanotechnology and such, this can be reduced to a relatively small chip. But we still have not created the human-like brain.

[AGI1122]

Working again. It was a problem with the upgrade to his brain.

[JoeKayzA]

Also, there are random elements based on what we see, hear, past experiences, etc. Computers have to simulate randomness based on the timer, CPU speed, and other less-than-random phenomena.

From inside the computer, you probably won't get random events, since everything (timer, cpu speed...) is deterministic. The only random events are actually coming from the user , which is a human (or a second AI? ).

I believe that without our senses, our brain would probably also work in a deterministic way, no magical randomizer in there (IMHO).

Well, and then there is the question whether our environment has random events at all, or whether everything that happens is a reaction to some actions. But that probably belongs to the 'creation vs. evolution' thread

cheers Joe

[Ushma]

Yes, the human mind is based on pure logic at the low level, but it's masked out heavily at the upper level. In essence, you would have to dig through a ton of creative and random things before reaching the underlying "pure logic" down there.

Also, there are random elements based on what we see, hear, past experiences, etc. Computers have to simulate randomness based on the timer, CPU speed, and other less-than-random phenomena.

The human mind is a machine. A very complex machine, but a machine nonetheless. You put in certain inputs; you get certain outputs. There are however, *very* many inputs, to the point at which it becomes *very* difficult, if not impossible, for humans to accurately predict the next output. To do so, we would need perfect understanding the logic of the brain (which is, as said, very complex, and we do not fully understand it) along with a complete knowledge of all the inputs. This doesn't happen; you would probably need to be omniscient. Indeed, you yourself do not really have a complete knowledge of your own inputs.

A computer simulating the human mind could apply inputs to the simulation just as they are applied to your mind from your senses. This would allow for the same amount of unpredicatability as a human, if there were enough inputs/sensors.

Also, these inputs are not necessarily random. The universe could quite possibly be deterministic. With neither the inputs nor the machine being undeterministic, you will receive a deterministic result.

Yes, it might be possible to simulate a human brain. But it wouldn't be very practical, and I don't think that the way they are trying to do it will work the way they want it to.

I agree that simulating a human brain would be impratical, not to mention pointless. The human mind is a very specialized piece of equipment for operating the human body. I don't think AI researchers are really trying to simulate the human mind. If they were trying to create sentient intelligence, they would do it in a way more appropriate to implementation on a computer.

[Crazed123]

IMHO, you people are wrong. The human mind is not essentially mechanical in nature because the human conscious and subconscious minds are not merely mental abstractions of the products of neurochemistry. I'll try to explain. The brain and its hardware acts like an operating system, interpreting sensory input and mental output to mediate between body and mind, right? However, as it does so it CREATES the user programs that are the conscious and sub-conscious mind. These in turn drive the interpretation of sensory data and direct the body to act. Furthermore, as the body ages, so does the brain, and either the user programs or the hardware itself can be influenced differently by an event depending on when it happened.

Therefore, because the user program can change itself OR the hardware can change it based on outside input (thus changing how it is directed), the mind is no machine. That is, outside events don't have a deterministic effect on the mind, they BLEND with what it already knows. This is how free will works. Things affect us to the extent that they match some pattern up in our heads, or affect our brain's neurochemistry in a way that has a broad, nearly random effect on the mind.

Kind of hard to explain beyond that, but suffice to say it's a lot more complicated than writing a few deterministic rules.

[Ushma]

Furthermore, as the body ages, so does the brain, and either the user programs or the hardware itself can be influenced differently by an event depending on when it happened.

Therefore, because the user program can change itself OR the hardware can change it based on outside input (thus changing how it is directed), the mind is no machine.

That is, outside events don't have a deterministic effect on the mind, they BLEND with what it already knows.

I suppose I wasn't clear. I never denied that the brain has a very large and variable state. The inputs it receives alter the state and affect future outputs. The decisions you make are based on the sum of your all experiences as well as your current inputs. Your experiences very much make you who you are, determining what your reponses to situations will be.

That is what allows the brain to seem random. The sheer number of possibilities for the state is so enormous to be incomprehensible to humans.