Albert is not predictable, and it responds to requests you send. This exists already.

We'll start that vote soon.

omg shit, I guess that means I might have to let a name change be up to the forum...

but awesome work Claeser :)

You'll be lucky if all they can come up with is Cougar McSkimask.

true, its not a very creative forum ay? :P

Also, why Albert? at least come up with something creative :)

I would find a Diplomacy AI very interesting, but I see some problems in making one as well.

One of the problems is that there are many different Nash equilibria in strategies in Diplomacy, with which I mean that there is no best strategy; everything is dependent on the psychology of the other players.

For example, if none of the other 6 players on the board put any effort in trying to keep allies or hold their word, then it is best for you to act the same way, and if you do, everyone on the board is then following the best strategy they can against the other players. However, if the other players do try to be somewhat honest, and punish liars, then the best strategy is something different. These are just examples; the bottom line is that the best strategy differs depending on who your opponents are, and that there does not exist any strategy that beats all other strategies.

This means that it is possible to build an AI that is extremely difficult to play against if all the players on the board are AI as well, but is easy to play against if only some of them are, and likewise, it is possible to build an AI that is very strong, yet does not prepare you for how a match with real top player humans would look like.

If we therefore want an AI to reflect (top) human play, we should build an AI that is best against 6 other humans, and NOT against 6 other AI, since the latter could result in strategies that would not work if most opponents are human (but would be very hard to beat by a human if most opponents are other AI).

This means that playtesting is difficult, since you can't just let the AI play thousands of games against itself and let it learn that way. The AI has to learn from playing against humans if you want the end result to actually be a match in a human environment.

That's right Mercy. Which is why we're already training an AI Diplomacy caper software package here at WebDiplomacy under the assumed player name ....
We had a promising new prototype with new "social skills" & political interests designed to engage in Forum threads as a learning tool, but regrettably KansasBoy....(oops, nearly said too much) was recently banned from the site. . The next version will be better

Austria-Italy doesn’t take too much training, does it?

I mean, that’s what makes it a fascinating problem. If we can build AI that can play video games, chess, and limit hold ‘em, we should be able to do the same for Diplomacy.

It probably will not be able to beat the best players, but in gunboat which does not allow for any communication I would suspect it could do quite well using current tech.

In gunboat, would it learn how to interpret those subtle failed support moves meant to indicate intention to ally.

Yes

A computer can't interpret language easily, but it can take signals and process them. The beauty of it is that to the computer, it wouldn't "know" that those moves signaled something. it would simply be that "in this situation, this type of move typically leads to success down the road with this other type of move"

Depends on if they trust each other for a Key Lepanto.

@datapolitical
"If we can build AI that can play video games, chess, and limit hold ‘em, we should be able to do the same for Diplomacy."
Like I mentioned in my previous post, there is a huge difference, though. While in games like chess and even in 1v1 Diplomacy there is a best strategy, in Diplomacy there is not a best strategy.

There isn't a best strategy in a lot of games for which we've built AIs

There are many worse strategies for games. In Diplomacy, a huge factor is the existence of stalemate lines. Great powers which sit on the main stalemate line are often at a big disadvantage because everyone else wants a piece of them.