Ken Williams presented AI::Categorize recently.

I just read an article on the Semantic Web recently and the following quote tickledme:

it is a curious phenomenon that the AI label tends to get dropped once the problem AI researchers were studying becomes tractable to some degree and yields practical systems. Voice recognition and text-to-speech, expert systems, machine vision, text summarizers, and theorem provers are just a few examples of classic AI tech that has become part of the standard bag of tricks. The AI label tends to mark things which aren't yet implemented in a generally useful manner, often because hardware or general practices haven't yet caught up.

I recently authored Array::PatternMatcher which is a Perl transliteration of the Pattern Matcher presented in "Paradigms of Artificial Intelligence Programming: Case Studies in Common Lisp" by Peter Norvig. I was going to create some sort of CPAN module after that that offered Prolog semantics with Perl syntax but thought it wiser to create perl bindings for SWI-Prolog instead.

for example, I am attempting to develop a rules-based system that can be used as a possible backend of a socket connection for business logic.

I think the hardest part of mating logical formalism with Perl is that rules systems (ie, Prolog) have a built in search engine and a different means of using variables.

This allows for more interesting uses of perl and AI; for example, I am attempting to develop a rules-based system that can be used as a possible backend of a socket connection for business logic

If you consider functional languages rule-based, then Erlang is the ticket for networked functional programming. The phone switching software for country Sweden was written entirely in Erlang.

• Genetics Algorithms.
  Each time I read 'survival of the fittest' it comes to my mind.
  Each time I've to tweak/optimize complex function that I can "rate" the result I choose this techniques too...
  Although transcoding your parameters for efficient crossing-over may be a REAL task...
  (see Masem's work : Algorithm::Genetic or the infamous Genetic Programming or breeding Perls)
  
  
• Neural networks.
  To classify complex objects (picture,sound).
  To create simple semi-adaptative behaviour.
  (tons of links on the web, here is start, CPAN has some modules too...)
  
  
• Computationnal Optimization(Simplex, simulated annealing...)
  Not really IA, but I like to combine with others techniques.
  (See Math::Amoeba for the simplex)
  
  
• "Inferential Theory of Learning" from Michalsky
  He did a wonderful job in formalizing the "learning process"...
  Every symbolic learning engine could take benefit of it...
  
  
• Sorted Alpha-Beta.
  For turn by turn process, with only one action at each step (chess play...)
  

Please tell us informed about what you do...

I presume you are talking to me? . . .

Well, I only ask because I am thinking about programming a game. I've read up on some topics--Tricks of the Game Programming Gurus, and 3D Game Programming With C++--quite a while ago, and I'm renewing my interest with it.

I'm trying to organize in my mind how I'm going to implement the brains, i.e., AI, of the characters, and I was hoping to get some more input onto the subject (thanks to princepawn about the AI perl newsgroup). For one example, if a character needs to find the player then I have a variety of searching and chasing algorithms. However, I need to choose the one most suitable for the character; for example, a primitive, easy, character would probably get the 'guess and check' method. However, that primitive character would probably be allowed a simple ability to learn bits and pieces of more advanced methods if he happened to see it, somewhat taught it to him, etc.

Right now I'm thinking about how smart to make certain characters, and once I've decided that how to make them that smart. The challance is, of course, judging how smart to make them, because it's easy to make an impossible to beat AI, making a challanging one, and one with an unique personality, however, is the challange.

I'm looking, now, at different people's advice on the subject, and at different implementations of it. A challange I've arrived at is the actual data implementation of the 'brain'. If any one has any experience with this, I'd be grateful.

Well, I guess I got somewhat of the subject, but I hope I've answered the question suitably enough for you.

Years back I did game development. What passes for AI will really depend on what type of game it is. For instance, I was working on a driving game, so we came up with ADI (Ass**** Driving Intelligence). The goal was to make people drive as badly as real people. (I'm from Massachusetts btw) So our goals were to mimic how people really drive. An RPG or a Quake style shooter have very different ideas of intelligence. Also, unless it's turn based, speed will be an issue as well.

As far as the searching method in your example, I would think the best way would be to define this "attribute" with a range, say 1..5.

Then I would define 5 (or however many) search algorithims as functions and call the appropriate one based on the entity's level. When you decide he's getting smarter, bump up his search "attribute" level.

You could also take an opposite approach and only have one ideal search and have a clutter factor that adds randomness to the result. This clutter factor would get reduced as the character gets "smartah"

-Lee

"To be civilized is to deny one's nature."