I learned backgammon (Tavla, in our household) relatively recently, and was interested for a bit, before I stumbled across
Jellyfish
, a tiny neural-net backgammon game that self-trained by being turned loose on the online servers.
It can
stomp me. (and just about everyone else, for that matter). This isn't a fancy supercomputer, and it doesn't have fancy algorithms and a billion lines of code. Being schooled by a software-AI equivalent of a barely multi-cellular organism takes the shine of a bit. I mean it isn't even close to say, flatworm level... I could handle being compared unfavourably to a flatworm.
Now I realise that refusing to play games that simple computers can beat me at doesn't leave much. There are a handful, usually with enough turn-options that we can overwhelm the computing power
currently available. Interestingly, this forces coders to try more elegant, less mathematical brute-force approaches, which tend to push the envelope of what might one day be true AI...
For now, at least, I recommend GO (Wei-chi, Baduk).
IGOwin
is a decent (simple 9x9) introduction, while
GnuGo
plays the more interesting full sized board. You'll need a gui for GnuGo - follow the links.
Matthias