The game is Lasca. It's a strange game in that is deceptively simple to learn but really hard to play well and predict outcome of moves.
It was invented by a chess champion a 100 years ago & I have loved it since finding a newspaper clipping about it when I was a kid. Despite being around a long time, it never really caught on.
When playing with people (I try to teach as many people in local area as possible & take a board with me to the pub!), a frequent response is "Wow.. I never expected that move to work out like that." This happens to me even when having played it for ages.
The rules are here.http://www.lasca.org/
Now, I am frustrated by all the computer versions I can find, as I can beat them all, and so they are just not much fun to play. I decided I would just have to make my own.
My AI already beats the other computer versions I have but still rarely beats a reasonable human who has played it a while. People ironically usually get good quite quickly when you show them.
Thanks for your tips and will try them.
* I will experiment with giving captured pieces a small value & set two AIs against each other (one with and without) but so far I have been constantly surprised by what seems to be a good weight and what isn't. For instance I no longer even score the number of pieces controlled by each player as I discovered doing that just made it play worse!!
* Taking future game states into account by scoring is really hard as the game can change dramatically very quickly. This is largely due to being able to force people to capture you, so sacrifice can be a great move in some circumstances. So I might frequently let someone capture 3 or 4 pieces, just so I can get them back later but this only works if you also make a plan to get them back. In tests I have found situations where an AI needs to look 50 ply ahead to reach a quiet point in the game with no material change within a full move i.e. you really know who came off better from that sequence. This makes it highly unpredictable.
Me and my friends who play Lasca in the pub frequently theorize about what is & isn't important but then find when I get back and adapt the AI we are usually wrong! More frustrating is the human brain seems to adapt to this game quickly but we have little idea really what we are doing.