Do you mean just for ordering units in gunboat?

I have actually thought about something similar (I've studied some AI). One simple solution for the early game would be to just analyze the best moves based on a database of earlier games (similarly as it's done in chess, for instance). You should probably use a mixed strategy though, so other bots or player couldn't predict your move. Especially when there's a guessing situation.

This can maybe take you through the first three or four years or so (especially for Classic as there are more archived games) but later on, you would need to use something more advanced like a path finding algorithm. The number of game states is not as large as in chess, for instance, so it should be doable in reasonable computing time, I'm guessing. But even later on, I would still use a mixed strategy to keep everyone else guessing. Maybe also try to identify the situations where randomizing your move is crucial.

Just off the top of my head, haven't read any articles etc. related to Diplomacy AIs.

Also, there's still some "diplomacy" involved in Gunboat (supporting units of players to want to build an alliance with etc.), so that makes matters more complicated.

And trying to identify when other players are allying up against. Or trying to prevent a player from getting to the victory condition. When to suggest draws etc. So it gets pretty complicated if you want to add too much "human-like" behavior.

It would be fascinating to see the bots play each other. I think bots would be (obviously?) be better than humans if the project were taken as seriously as teaching computers chess.

This has been done already.

It is a lot of fun, but nowhere near as good as human players. There's even a rudimentary form of negotiation available, which I used a bit but found too hard to manage on a long term scale.

I wish the program had a map designer that could port to the DAIDE format.

Some quick google links after googling "DAIDE" :

https://sites.google.com/site/diplomacyai/QuickStart
https://sites.google.com/site/diplomacyai/home/daide
https://sites.google.com/site/diplomacyai/
http://www.daide.org.uk/
http://diplomacyai.blogspot.com/
http://www.ellought.demon.co.uk/dipai/

Maybe you guys who know AI and programming languages could take up the torch and build a frontend/client suited for this decade?

Scuba - Chess is a two player game with no cooperation or negotiation and no need to look at your opponents moves in a potentially friendly light. In short, chess is a simpler game in diplomatic and emotional intellect. Ches has more complex moves but simpler motives and fewer players. This makes the decision engine easier to write.

@YCHTT, then a bot for the two player variants would be much easier then even a gunboat one for the Classic?

Yep. In fact a two player bot would likely be easier than a chess bot due to fewer starting pieces and movements being simpler.

Yeah.

Head to head diplomacy would be pretty easy for a bot. It would be able to make the best move every time.

I suspect that teaching a computer what constituted "friendly" or "aggressive" play wouldn't be too difficult. That is to say, extremely difficult but not as difficult as chess. I have no objective facts or information to support my belief.

Batallion, it already happened. The DipAI project, some ten years ago.
The ground for the bot battles was the daide server
http://www.ellought.demon.co.uk/dipai/
The Albert bots have won. Try them out, following the links above by Teacon.
I challenge anyone to play seven games against Alberts (latest release, max level). If you get one solo, you are a good gunboat player!