These are good enough to mention, but not good enough for me to call "interesting".
| Jay : game learning : less interesting |
Some of them are commercial games; they may have fascinating AI, but the publishers are usually reluctant to release details. Seek more information at Steven Woodcock's current and upcoming games page.
Battlecruiser: 3000AD
A complex commercial game with a neural-net based AI
which controls many details of game play.
Coins Neural Game
Play a simple guess-your-opponent's-move game on this web page.
The sparse implementation is by Geert Jan Bex.
Two neural networks are included in the choices of opponents.
Creatures
A commercial artificial life game in which you breed "creatures"
which are controlled by neural networks.
Dynamic Gin
A shareware gin rummy program which adapts to its opponent
while it is playing, from
CyberSym Technologies.
The web page has no technical information that I found, but
the company wrote to me that the learning engine "is (primarily)
a temporal difference neural net." They say the play is strong
but not top-level, partly because they wanted fast response even
on older computers.
Galapagos
A commercial game in which you have some control over the environment
of a "synthetic organism", Mendel, which can learn on its own.
It is a puzzle game in which one of the puzzles is to teach Mendel
how to behave the way you prefer.
Last One Loses
A simple game in which the computer opponent learns to play perfectly
by rote learning. You can download the Visual Basic program.
Queenbee
Jack van Rijswijck
A hex program which is planning to use unspecified machine learning methods.
Diplomacy Programming Project
A project to write a Diplomacy AI.
Diplomacy is a game that could benefit from machine learning but,
as far as I can tell, doesn't yet. Also see
The Diplomatic Pouch, which has a number of articles about the
DPP.
Xconq
A free strategic wargame framework in which game designers
can create a wide variety new games on a hex board, using a declarative
game design language. As an example of the flexibility, the xconq
distribution includes not only the usual WWII-like games, but also a
Godzilla-Tokyo scenario.
It supports plug-in AI modules; the one included AI can play all
xconq games, though weakly. Writing a smarter AI would be a good metagame
task. There are problems: The human interface is clumsy. The AI interface
is undocumented.