🎉 Celebrating 25 Years of GameDev.net! 🎉
Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!
how far away are we from genuine Artificial Intelligence?
Author
Is it just that we currently do not have the raw computing power neccesary or is it a matter of finding that magic formula?
Perhaps it will take some genius or something to figure it out.
Assuming a computer based on the current standard architecture of processor and memory is adequate for AI, then we do have enough computing power, because A) time is irrelevant to demonstrate the viability of an algorithm, and B) there would appear to be virtually no limit to storage capacity given virtual memory, the size of hard drives and the ability to run networked storage.
That leads us to the magic formula. Many would argue that the magic formula is going to be hard to come by, and so have decided that a knowledge base rich in common sense and deductive abilities is the answer. Others believe they have found the magic forumula in various cognitive architectures, such as Soar, and merely need to demonstrate it as such.
Given what people discuss here in this forum, it would appear that AI is comprised of path finding, scripting, neural networks, genetic algortihms, and simplified finite state machines. I really don''t think these concepts will take you very far.
Why don''t you look into Cyc, Doug Lenat, The twelve dimensions of context, OSCAR, John Pollock, Soar, John Laird, Marvin Minsky, Douglas Hofstadter, situation calculus, ATP, Lisp, Prolog, defeasible reasoning, default reasoning, semantic nets, monotonic reasoning, nonmonotonic reasoning, belief systems, metreasoning, common sense reasoning, natural deduction, belief revision, and higher order logic.
That leads us to the magic formula. Many would argue that the magic formula is going to be hard to come by, and so have decided that a knowledge base rich in common sense and deductive abilities is the answer. Others believe they have found the magic forumula in various cognitive architectures, such as Soar, and merely need to demonstrate it as such.
Given what people discuss here in this forum, it would appear that AI is comprised of path finding, scripting, neural networks, genetic algortihms, and simplified finite state machines. I really don''t think these concepts will take you very far.
Why don''t you look into Cyc, Doug Lenat, The twelve dimensions of context, OSCAR, John Pollock, Soar, John Laird, Marvin Minsky, Douglas Hofstadter, situation calculus, ATP, Lisp, Prolog, defeasible reasoning, default reasoning, semantic nets, monotonic reasoning, nonmonotonic reasoning, belief systems, metreasoning, common sense reasoning, natural deduction, belief revision, and higher order logic.
AI is a matter of scale.
Todays most advanced robots have the reasoning equivalent to that of an insect, which is a tremendous leap since the scientific field begun.
AI in games isn''t growing too much nowadays, everybody wants graphics.
Todays most advanced robots have the reasoning equivalent to that of an insect, which is a tremendous leap since the scientific field begun.
AI in games isn''t growing too much nowadays, everybody wants graphics.
quote: Original post by Anonymous Poster
AI in games isn''t growing too much nowadays, everybody wants graphics.
Ummm...The Sims? Black & White?
Just curious how these relate to AI in games not growing...
ld
If some of the stuff I''ve read is to be believed, then Russian scientists may have produced artificial intelligence already. I would have to go back and read the articles again, but the way I got it was that some AI research was based on genetic or biological models. Apparently, the Russian scientist tried a different variation of this approach (the "magic formula"?).
"A man can''t just sit around." ''Lawn Chair'' Larry Walters (1982)
"A man can''t just sit around." ''Lawn Chair'' Larry Walters (1982)
Tracy, I believe you are reffering to "the Russian Brain". And by the theories of those who have posted in this room AI has been achieved. However, if that is true why are we not satisified? The awnser lies in this...
Artificial Intelligence began as a theory that one day humans would interact with a man-made "brain". As of today we have achieved that. Every man woman and child has the ability to sign on to their OS and find some form of computer game that interacts with them. A couple of years ago I downloaded a program called "Cherie". Cherie was a freeware AI. I could ask her questions and she would respond. She was had a astro-physiological master''s from oxford (or something of the sort) and when she didn''t understand my question I could ask questions closely related and soon she would be able to awnser my origional question.
This is AI, a hardware representation of this would be a "robo-maid". Give it a complete layout of your house and tell it how to clean what with what cleaning material and how much of that material to use. After several testruns the maid will be able to clean your house perfectly. Withing days however this will not prove sufficient. Say perhaps you wish to remodel your living room. The pathway the "robo-maid" made is now disrupted by your change. Thus the robo-maid must be reprogrammed and re-trained.
The final product however is just the same as today''s latest games. Say a Half-life enemy learns something new each time you play... Is this really AI? Or is this a class where a program entity is learning today''s lesson? I would go with the later.
There are somethings about us that are different than a Half-Life bot. However I believe there is a way to create an AI that is almost a real person.
The main reason we know why we are doing something is because of our history. We all have been raised by brains, brains that have learned from their parenting mistakes, who in turn have been raised by other brains. The way you act and socialize have been someone controlled by your parents. And the way your parents have raised you has been influenced by whoever raised them. Thus begins step one.
+How you have been taught to act?
Another thing we must study is our own personal experiences. Why is it that your best friend is never embarrassed, and you cannot dare to drop papers at work? Perhaps your best friend has been embarrased so much he doesn''t care, or perhaps he used to embarrass people constantly. You of course might have had childhood trauma when your suit flew off at the public pool. These are hypothetical incidents none the less and only surface the topic of...
+What have you taught yourself?
Another thing we must realise is that our environment plays a crucial role in how we act. I go to thrift stores often to pick up clothing only because it is popular. I play first-person shooter games because they too are popular. And I listen to Electronica... Now, imagine me fourty years ago. Electronica?? Computers??? Going to thrift stores when your making tons of money??? What is this nonsense? -I would be out of place.
+Where are you and what are your exo-motivations(what outside of yourself motivates you to do what you do)?
Most of these topics are just broad, but could a dog awnser these questions? Better yet could you awnser these questions for a dog?(I use the dog as an example because AI programmers believe that they can program an AI pup. Based on what I have read from the Game Developers Conference) I doubt it. Hence the reason no one is getting anywhere and why AI seems so difficult. No one has ever been a Army Man at the Black-Mesa Compound in Half-Life. And if they have, I beat that game, and I killed lots of people so you probably wouldn''t live to tell about it. No one has ever actually seen an alien, no one has ever fired a gun at bank robbers, and no one has learned from being killed how to fight better next time. Start with yourself. Simple Joe Programmer. How would joe programmer handle this situation? Sure no one wants to play a game where they program the next hit game for 2002, but once you have created an environment, and put AI-Joe into action you can learn from your AI-self what happens to you when you go into the 3D world. And from AI-joe you can use physics and algorithms to create the next step AI-GI JOE!
-Ian Landis
Artificial Intelligence began as a theory that one day humans would interact with a man-made "brain". As of today we have achieved that. Every man woman and child has the ability to sign on to their OS and find some form of computer game that interacts with them. A couple of years ago I downloaded a program called "Cherie". Cherie was a freeware AI. I could ask her questions and she would respond. She was had a astro-physiological master''s from oxford (or something of the sort) and when she didn''t understand my question I could ask questions closely related and soon she would be able to awnser my origional question.
This is AI, a hardware representation of this would be a "robo-maid". Give it a complete layout of your house and tell it how to clean what with what cleaning material and how much of that material to use. After several testruns the maid will be able to clean your house perfectly. Withing days however this will not prove sufficient. Say perhaps you wish to remodel your living room. The pathway the "robo-maid" made is now disrupted by your change. Thus the robo-maid must be reprogrammed and re-trained.
The final product however is just the same as today''s latest games. Say a Half-life enemy learns something new each time you play... Is this really AI? Or is this a class where a program entity is learning today''s lesson? I would go with the later.
There are somethings about us that are different than a Half-Life bot. However I believe there is a way to create an AI that is almost a real person.
The main reason we know why we are doing something is because of our history. We all have been raised by brains, brains that have learned from their parenting mistakes, who in turn have been raised by other brains. The way you act and socialize have been someone controlled by your parents. And the way your parents have raised you has been influenced by whoever raised them. Thus begins step one.
+How you have been taught to act?
Another thing we must study is our own personal experiences. Why is it that your best friend is never embarrassed, and you cannot dare to drop papers at work? Perhaps your best friend has been embarrased so much he doesn''t care, or perhaps he used to embarrass people constantly. You of course might have had childhood trauma when your suit flew off at the public pool. These are hypothetical incidents none the less and only surface the topic of...
+What have you taught yourself?
Another thing we must realise is that our environment plays a crucial role in how we act. I go to thrift stores often to pick up clothing only because it is popular. I play first-person shooter games because they too are popular. And I listen to Electronica... Now, imagine me fourty years ago. Electronica?? Computers??? Going to thrift stores when your making tons of money??? What is this nonsense? -I would be out of place.
+Where are you and what are your exo-motivations(what outside of yourself motivates you to do what you do)?
Most of these topics are just broad, but could a dog awnser these questions? Better yet could you awnser these questions for a dog?(I use the dog as an example because AI programmers believe that they can program an AI pup. Based on what I have read from the Game Developers Conference) I doubt it. Hence the reason no one is getting anywhere and why AI seems so difficult. No one has ever been a Army Man at the Black-Mesa Compound in Half-Life. And if they have, I beat that game, and I killed lots of people so you probably wouldn''t live to tell about it. No one has ever actually seen an alien, no one has ever fired a gun at bank robbers, and no one has learned from being killed how to fight better next time. Start with yourself. Simple Joe Programmer. How would joe programmer handle this situation? Sure no one wants to play a game where they program the next hit game for 2002, but once you have created an environment, and put AI-Joe into action you can learn from your AI-self what happens to you when you go into the 3D world. And from AI-joe you can use physics and algorithms to create the next step AI-GI JOE!
-Ian Landis
Well, true AI will come within my lifetime. I''ll make sure that happens 
But for now, I think the next great step would be a "Talking" AI. With this, I mean an AI you can talk to, in such a way that your AI actually *understands* you. Right now, all that there pseudo-AI''s can do, is put your question through an algorithm and give an anwser. Without *understanding*.
The understanding part will be hard to do... But possible. And then, we can really have fun with AI :p
But for now, I think the next great step would be a "Talking" AI. With this, I mean an AI you can talk to, in such a way that your AI actually *understands* you. Right now, all that there pseudo-AI''s can do, is put your question through an algorithm and give an anwser. Without *understanding*.
The understanding part will be hard to do... But possible. And then, we can really have fun with AI :p
Talking AI that understands you?
Okay, my problem with this is that communication is a form of interaction based on semantic information transfer (this is an abstraction of the truth, but a useful one). This information transfer is based on symbol manipulation, each symbol being an abstraction of experience. The understanding occurs when a symbol made by another being makes sense to you based on your experiences. The only way this will happen is if you are a being that has experienced reality through direct interaction of the world.
If you have not then any communication will be pure symbol manipulation and no understanding will occur.
So to have a talking bot that understands you need an embodied, situated machine that has learnt about the world through interaction with it.
Not only that, but it''s interactions must be similar enough to yours to make sense. Communication cannot occur between a person and a dog except along the lines of abstracted similar experience. Food makes sense, walk makes sense, playing the piano does not. This suggests you''d need a physiology and a brain that work in a similar way, not necessarily but quite probably.
This does not leave creating understanding in a communicating bot as such a trivial task as might first appear.
Mike
Okay, my problem with this is that communication is a form of interaction based on semantic information transfer (this is an abstraction of the truth, but a useful one). This information transfer is based on symbol manipulation, each symbol being an abstraction of experience. The understanding occurs when a symbol made by another being makes sense to you based on your experiences. The only way this will happen is if you are a being that has experienced reality through direct interaction of the world.
If you have not then any communication will be pure symbol manipulation and no understanding will occur.
So to have a talking bot that understands you need an embodied, situated machine that has learnt about the world through interaction with it.
Not only that, but it''s interactions must be similar enough to yours to make sense. Communication cannot occur between a person and a dog except along the lines of abstracted similar experience. Food makes sense, walk makes sense, playing the piano does not. This suggests you''d need a physiology and a brain that work in a similar way, not necessarily but quite probably.
This does not leave creating understanding in a communicating bot as such a trivial task as might first appear.
Mike
quote: Original post by MikeD
Okay, my problem with this is that communication is a form of interaction based on semantic information transfer (this is an abstraction of the truth, but a useful one). This information transfer is based on symbol manipulation, each symbol being an abstraction of experience. The understanding occurs when a symbol made by another being makes sense to you based on your experiences. The only way this will happen is if you are a being that has experienced reality through direct interaction of the world.
If you have not then any communication will be pure symbol manipulation and no understanding will occur.
Mike is talking about the symbol grounding problem which could be described as the holy grail of philosophical AI at present. For an excellent coverage of the problem read Stevan Harnad''s works.
On the issue of realisable AI, you have to first ask yourself what you define intelligence to be before you can stipulate whether something artificial is intelligent or merely displays intelligent behaviour. Is there a difference? Is it possible to have intelligent behaviour without being intelligent?
Currently, behavioural AIs are at the level of roughly a human 2 year old. Some VERY interesting results have been achieved in this area. Here''s an example.
Several years ago I was visiting the US to attend the massive AAAI conference. It''s one of the largest Ai conferences in the world and draws researchers from all corners of the globe. During an evening display session I came across a small stand where a Japanese researcher sat behind a table. On the table was a small robotic cat, complete with fur. It was purring. I stroked the cat and scratched it and it made suitable noises and moved in ways that reminded me of my own cat back home. It behaved as I expected given my actions. I spoke to the researcher and he told me that the cat was an amalgam of tactile sensors, actuators and an emotional model and it''s purpose was for disabled people to interact with. Because it was artificial it needed less looking after than a real cat.
I played with the cat a little more and then decided to test the boundaries of its behaviour. I smacked it quite sharply on the head. It made a sharp irritating sound that could easily be interpreted as a scream. The researcher looked astonished. He explained to me that he had not specifically programmed such a response in the cat. It''s good behaviours had been calibrated so as to present realistic feedback to the user but he had never specifically designed the system to respond in this manner. It was clear that the model he had designed was capable of emergent behaviour. Something that is certainly a pre-requisite for intelligence!
It certainly freaked me out!
Cheers,
Tim
One of the problems with discussions like these is that people start talking about it without defining what they are talking about.
What do you mean by AI? If you want the computer to figure out a solution to a very specific problem, we can do that in nearly every case. The more general you make the problem, the more difficult it becomes, until it becomes intractable.
A few definitions of AI:
Getting computers to do what they do in the movies.
Anything we haven''t figured out how to make the computer do yet.
That said, I think there have been enormous strides in AI. Hand writing technology and speech recognition are fast becoming an everyday occurrence. AI in games get more and more of the clock cycles every year. AI is beginning to get the same billing as the graphics engine (look at how lauded the Unreal Tournament AI bots were).
I still think we are far away from a general purpose AI that can clean your house, drive your car, have deep philisophical conversations with you, is self-aware and speaks 13 million forms of communication.
Ut
What do you mean by AI? If you want the computer to figure out a solution to a very specific problem, we can do that in nearly every case. The more general you make the problem, the more difficult it becomes, until it becomes intractable.
A few definitions of AI:
Getting computers to do what they do in the movies.
Anything we haven''t figured out how to make the computer do yet.
That said, I think there have been enormous strides in AI. Hand writing technology and speech recognition are fast becoming an everyday occurrence. AI in games get more and more of the clock cycles every year. AI is beginning to get the same billing as the graphics engine (look at how lauded the Unreal Tournament AI bots were).
I still think we are far away from a general purpose AI that can clean your house, drive your car, have deep philisophical conversations with you, is self-aware and speaks 13 million forms of communication.
Ut
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
