Quote: Original post by lucky_monkey
[...]So you're suggesting that any problem solving process that doesn't involve inspiration is not intelligent?

No, that was merely an example.
Another example: How do you identify objects that you see? If you could reduce that to an algorithm, you could have a perfect(in the sense that it would work in any situation where a person would) image recognition system and you'd be rich. Despite your (and everybody elses, currently) inability to do that(though many have tried and good approximations exist), you can still identify objects you see. Beyond that, you can put them in contexts and guess (generally correctly) other information not given by the current inputs.

If you see a stack of paper bound on the two larger faces by a thicker material that also extends over one of the smaller faces (which takes a lot of work to recognize in the first place), you can guess that it is a book, and that information is spread out over the papers in a serial fashion with most of the papers being numbered on both sides, possibly the pages on one side contain an outline of the information (table of contents) and the papers on the other side indicate where certain more granular topics are contained in the book (index), etc.

Quote: Original post by ToohrVyk

Quote: Original post by lucky_monkey
So you're suggesting that any problem solving process that doesn't involve inspiration is not intelligent?

If you can explain how to solve a problem without needing inspiration, intuition or other unexplained human abilities, then what you describe is an algorithm that solves the problem, and can create a machine that implements that algorithm. That problem is then similar to any other algorithmic problem, such as adding together two integers.

I'm a mechanist: our brain is such a machine.

Quote: Original post by Extrarius

Quote: Original post by lucky_monkey
[...]So you're suggesting that any problem solving process that doesn't involve inspiration is not intelligent?

No, that was merely an example.
Another example: How do you identify objects that you see? If you could reduce that to an algorithm, you could have a perfect(in the sense that it would work in any situation where a person would) image recognition system and you'd be rich. Despite your (and everybody elses, currently) inability to do that(though many have tried and good approximations exist), you can still identify objects you see. Beyond that, you can put them in contexts and guess (generally correctly) other information not given by the current inputs.

If you see a stack of paper bound on the two larger faces by a thicker material that also extends over one of the smaller faces (which takes a lot of work to recognize in the first place), you can guess that it is a book, and that information is spread out over the papers in a serial fashion with most of the papers being numbered on both sides, possibly the pages on one side contain an outline of the information (table of contents) and the papers on the other side indicate where certain more granular topics are contained in the book (index), etc.

You're underestimating the vast amount of experience we accumulate before we can make these differentiations and inferences ourselves. Give a sufficiently complicated machine enough experience to draw upon and it'll do the same thing. Our capability for abstract symbolic reasoning is developed, not a magical property we are born with.


Problem solving is a perfect example of the mechanisms of intelligence. We deal with problems by breaking them into smaller pieces and comparing them to other problems that we know how to solve. This displays a capacity for generalisation, pattern recognition, logical reasoning, and the ability to utilise experience and previously formed connections.

Intuition is the result of a subconscious leap in this process, whereby we can't identify the experience(s)/connection(s) that led us to make a particular connection.

Inspiration is a process whereby a link is formed while the experience you are drawing upon is actually happening.



p.s. This is all my opinion, and I approach this discussion with an open mind, so feel free to brutally shoot my crackpot ideas down :)

I'd like to point out that Real AI is an oxymoron.
Real Artificial Intelligence?

so yeah, of course there isn't any Real AI. It's either AI or Intelligence.
With that out of the way, what was called 'evolved scripting' could also be called 'AI method of choosing appropiate scripts'

But that's just nitpicking. It's AI. What character-driving AI wouldn't involve a script? would you have AI algorithms that manipulate bones so you wouldn't have to stick to animations? that's kind of using raytracing instead of polygons. Cool, but not really a good solution for the original problem. Would you use AI to generate sentences and then have it output sound? Sure, if you can pull it off.
Instead, we use pre-recorded animations, audio and simple behaviors even. Its a limited AI, yes... but things are always limited in a simulation. The limit just moves with technology.

in the 80's, this would have been called "life-like characters"

Quote: Original post by Madster
I'd like to point out that Real AI is an oxymoron.

I hope you ment "A rhetorical figure in which an epigrammatic effect is created by the conjunction of incongruous or contradictory terms". Also note that meaning of a sentence, or a phrase is not always possible to determine from an atomical meaning of each world.
Other interesting oxymorons are: "accurate estimate" "absolutely unsure" "American education" "fair trial"

Quote: Real Artificial Intelligence?

The correct term is a strong AI.

The problem with strong AI in games is an amount of AIs need by a RPG. Alas lack of memory is a big concern, thus even if developers are not wasting the memory by abuse of standard containers, they are still short of it by 15x- 300x.


----
That scripting debade is rather strange. Of course we can call some of actions, done by AI, methods of choosing AI state. Or we could pretend AI is creating scripts, or machine code, at runtime. Will choice of nicknaming matter?


---

Quote: Original post by lucky_monkey
We deal with problems by breaking them into smaller pieces and comparing them to other problems that we know how to solve.

Not everyone developes new things this way. Localy optimal pieces of problem don't need to create globaly optimal solution.

Quote: Original post by Raghar

Quote: Original post by lucky_monkey
We deal with problems by breaking them into smaller pieces and comparing them to other problems that we know how to solve.

Not everyone developes new things this way. Localy optimal pieces of problem don't need to create globaly optimal solution.

Sorry, I don't quite follow you.

I can't think of a problem I wouldn't solve using this method. Give me an example and I'll reconsider my opinion.

Quote: Original post by Raghar
I hope you ment "A rhetorical figure in which an epigrammatic effect is created by the conjunction of incongruous or contradictory terms".

Uh... is there another meaning?
add to the list: "Real Artificial"

And I agree, the nicknaming doesn't really matter.
As for AI in large groups, I always thought clustering of some kind was used. Is this true?

Quote: Original post by lucky_monkey

Quote: Original post by Raghar
I can't think of a problem I wouldn't solve using this method. Give me an example and I'll reconsider my opinion.

I can't think of a problem I wouldn't solve using this method. Give me an example and I'll reconsider my opinion.

Travelling salesman problem.
Hamilton circle.
Assault problem.
And just any problem where feedback function would drive researcher from optimal solution. And also any problem where reseracher would have all or nothing.

[Edited by - Raghar on January 23, 2006 1:08:39 PM]

I'm pretty sure that the optimizations for those problems, if found, will be due to breaking the problems into discrete steps and deriving answers to those steps from past knowledge.

In fact, in analysing those problems you can solve them (though not in an optimized fasion) by breaking them into smaller pieces and figuring out what needs to be done. Your examples are invalid. Problem solving is, by nature, a discrete process. The simple fact that someone could write a program that attempts to solve one of those problems (even if it would take a million years) means that it has been broken into steps and been solved... Just that it's been solved poorly.

What you are talking about is sometimes called reductionism. There are a lots of cases when so called reductionism is either useles, or harmful.

As for above problems.
It could be shown that Delaunay triangulation isn't optimal solution of Travelling salesman problem. Hovewer Delaunay triangualation could be viewed as a ruductio ad absurdum for the Travelling salesman problem.
Now if we will look for suboptimal solutions, I remember on an algorithm that created a very nice aproximation, however it used all input data, it didn't need to partition them. From my experience solutions type global - local - global are superior to local - local - global.

Hamilton circle is just a more difficult type of a Travalling salesman problem, that is possible to convert to TSP.

Assault problem is particulary nasty because all input set isn't correctly known when output should be ready, so it could be best described as an interaction with random function.

Note I said the reductionalism can't guarantee an optimal solution.