To what extent do optimal strategies exist?

[Guest]

This sounds similar to what has been attempted in poker for years. I think that someone can get

good

but not

great

. The AI that we see today just hasn't gotten even close to that level.
That being said, I think that to program AI to get

good

at mafia would be much more difficult than poker. This is simply due to having so many more possibilities to analyze.
Maybe you should assume that there is an optimal strategy and do a cost-surface search using techniques such as neural networks or genetic algorithms. With enough test-cases or generations, respectively, you should be able to get pretty close.