In my undergrad AI course, we have been tasked with a final project demonstrating some aspect of AI (neural nets, ant algorithms, knowledge bases, etc). The topic we choose is pretty wide open. My first idea, after reading someone else's idea to do this (but who never actually did it), was to create a very basic 2-D fighter type game. 1 player, 1 comp. Comp starts out only knowing very basic concepts (pain). As the player beats him up, the comp will learn to defend/fight back (again, using very basic moves: block, kick, punch, jump). The graphics could easily be put on the backburner and this could be done totally in text if need be. I'm fairly adept in coding, but have never dealt with neural nets before. What is the feasability of doing this in 1.5ish months? Also, do any of you have some suggestions as to possible projects? I am looking to do something that can, IN SOME WAY, tie into game dev, or my current RPG project. Ant algorithm, particle swarm optimization, fuzzy logic, natural language processing, artificial life are some of the topics we have been given, but we're able to pick anything dealing with AI. Pre-Thanks for any constructive feedback ;)

It is feasable, but I fear you will spend more time on the rest of the game (controls, graphics, etc() than the actual A.I.

In my bachelor degree days, I had to choose a similar project, only I had only 2 weeks and needed to code it in Scheme. I did an expert system that learned from user input to increase its knowledge base and provided feedback on how he made its decision. A+. If you want to impress an AI teacher, do something he didnt tell you about.

I had bad experiences with neural networks, which are difficult to train correctly I think. If you want to do learning, maybe you could look into "Support Vector Machines". They are newer, and in my field (computer vision), people use them with great success.

Hope that helps!