AI Newbie

#1 Use a static weight table configuration for general static logic
abilities. It would be paramaterized through polymorphic interheritiance
class objects which could be assessed via the Torque scripts.

Use a dynamic neuro net for training new abilities and also for elvolution.

You can find algrorithms on the internet and/or code from books.

neuro nets are in motion so you'll need to know some calculas if there
going to be advanced.

There's really no other way to do one "Correctly". You could simulate AI
using simple lookup tables and conditional logic but they are very limited
because of size/speed issues.

3 months work minimum unless you have good experiance.

Please forgive my spelling.

good luck.
Areal Person

#2 Thanks areal. that was quite some info!
even though all of this is completely new to me, and i dont even know what a neuro net is or polymorphic , i now have something to google for.
Thank you for ur help,
Gil

#3 You could try decision trees instead where you just branch off from one common point. For example: Is the creature alive? Yes/No. The No branch would lead you into a menu system for recreation, whilst Yes could lead you on to things like: Has it been fed/petted/played with, etc. They would then each branch off.

It all depends on how deep you want to go with "Academic AI" or "Game AI".

Neural Networks are definatley a very powerful way to go as they actually mimic the way a human brain works. There are some good books that will cover these, one I recommend is:

Artificial Intelligence: A Guide to Intelligent Systems by Michael Negnevitsky (ISBN: 0-201-71159-1)

There is a really in-depth book on Neural Networks themselves, and it is:

An Introduction to Neural Networks by Kevin Gurney (ISBN 1-85728-503-4)

That is probably more than you'd need though.

You could also look into:

Finite State Machines
Fuzzy Logic
Genetic Algorithms

The animation would depend on how far you want to take it. Simple 2D graphics would be straight forward.

It certainly sounds interesting to use PHP and VB. Is that under the .NET framework? If you know javascript then you could knock something together with that!

Anyway, good luck with it all.

Oh, you could look at the Artificial Life movement as well, some of that stuff is really cool!!

#4 I like Martin's idea about the decision tree, it would be easier on precessor then a whole Neural Net. after all it is suppose to be a animal/pet. toss in a few random elements and i think you might have something.

As far as it's decision tree you should take a look at the difference between wants and needs as and make some sort of priority system. a dog needs food and wants to get petted but the need wieghs heavier. it needs shelter but wants a nice cosy dog house, some that could effect it's happiness or something, just some ideas. (check out MASLOW'S HIERARCHY OF NEEDS sorry about the caps i pasted it)

#5 I strongly, strongly, advise against using neural nets or genetic algorithms for your problem. They are not artificial intelligence schemes. They are tools for piece-wise parallel regression, and not very good ones IMO.

The decision tree is a good way. If you really want to include learning, check out learning trees, and information theory.

A simple state machine might also work well. Obviously you want something that is mostly data-driven. I remember seeing some articles about how The Sim's AI worked, floating somewhere on the web, you might want to search for that.

Between a Sim and a Tamagotchi, its pretty much the same! ;)

#6 ^Oh, I dunno. I think GAs and ANN's can be used effectively in AI. I don't think either is a total solution in of itself, but with the proper contraints, I think both can be pretty useful.