AI bots on webDiplomacy: Skynet edition!

[Tom Bombadil]

I'm astonished how good these bots are.

http://webdiplomacy.net/board.php?gameID=247148

In this game I was worried that sending fleets through the med as Turkey to help prevent a solo would "trigger" the bots to think I was attacking them. It did not. The Italian bot allowed me into Adriatic, and to send a fleet all the way to MAO and neither the Italian or French units viewed it as aggressive which I think is pretty remarkable.

[jmo1121109]

On the chance anyone noticed, we had the bots offline for about 10 minutes, Philip was making some adjustments. The adjustments will ideally help out end game play a little bit in keeping powers from being "stuck" in all holds for to long, and will prevent a bot from trying to retreat 2 powers to the same location. We really appreciate him continuing the work with us to make some fine tuning where possible!

[CptMike]

On the chance anyone noticed, we had the bots offline for about 10 minutes, Philip was making some adjustments. The adjustments will ideally help out end game play a little bit in keeping powers from being "stuck" in all holds for to long, and will prevent a bot from trying to retreat 2 powers to the same location. We really appreciate him continuing the work with us to make some fine tuning where possible!

Intelligent design

[goldfinger0303]

So I finally got around to playing my first game, which was a rousing success.

Overall, it was an interesting experience, but there are definitely some odd choices by the AI. Such as Italy choosing the same moveset to attack Austria for 9 turns in a row, Turkey attacking Russia while losing Greece to Italy, Germany continuing to attack Italy while I attacked him, and Italy letting Turkey walk right into his centers.

It seems to me that the AI does an excellent job at defending itself from attack and executing attack. But things like me offering support moves into a territory went completely unnoticed. They also seem to be woefully unprepared for a stab each time it happens. And importantly, they don't seem to recognize that "This game is stalemated. If the human keeps going now, it's because they want to eliminate me"

There were two great moments for me in the game: one when Germany supported Turkey into Norway and I was like "omg the AI knows how to toss a game and ragequit" but while his disbands were to the effect of helping Turkey cross the line, he never followed up on that with the rest of his units. The other was when Italy suicided on my fleet line in the Med. Was pleasantly surprised by that (although it was wholly unnecessary if he had just had any foresight into the Turkish stab.)

All in all - a wonderful addition to the site, and it has a solid base with much room for improvement. However, they thankfully aren't world beaters yet :)

[zultar]

Update: please read!!!

Hi all,

After getting the feedback from many of you and further testing on our end, Philip and I talked and he has worked some more of his voodoo magic. There are now two different versions of the bots. One version is the original bot that you have been playing with for the last few days and another is a more aggressive bot that may take more risks and try other moves. In a classic gb game, half of the bots will be the original and half will be this more aggressive version (all random). Our hope is that you will see more variations in the mid to end games and where the bots will take some more chances to go for the win. :)

In 1v1 game, half of the time you'll get the original and half you'll get the new version.

Please try out several games and keep some notes of where you think you see awesome bot behaviors or some changes in how they play and post them here. It'd be best if you can provide links to the specific game(s), the phase/year where you saw some interesting moves, and why.

[BobMcBob]

I'm already impressed. In my game "Better Bots???????????????????????????????????????" I immediately see not only Austria opening to Tyrolia, but also Italy opening to Piedmont. These new bots look like a ton of fun to play against.
http://www.webdiplomacy.net/board.php?g ... ocache=917

[Flamebere]

I've been enjoying the AI's, its nice to try new strategies with no risk. However, things like this makes me question their intentions.
https://i.imgur.com/I0CCUZ5.png

[Claesar]

I've been enjoying the AI's, its nice to try new strategies with no risk. However, things like this makes me question their intentions.
https://i.imgur.com/I0CCUZ5.png

Why? That'a solid moveset.

[BobMcBob]

I've been enjoying the AI's, its nice to try new strategies with no risk. However, things like this makes me question their intentions.
https://i.imgur.com/I0CCUZ5.png

Why? That'a solid moveset.

I think he's talking about the move to MAO, which had pretty much no hope of succeeding. Although if it had been a human player, England could have supported to Brest leaving MAO empty. It works. It's not a completely stupid move.

[Claesar]

I've been enjoying the AI's, its nice to try new strategies with no risk. However, things like this makes me question their intentions.
https://i.imgur.com/I0CCUZ5.png

Why? That'a solid moveset.

I think he's talking about the move to MAO, which had pretty much no hope of succeeding. Although if it had been a human player, England could have supported to Brest leaving MAO empty. It works. It's not a completely stupid move.

Exactly, and
1) The bot doesn't know whether England is human.
2) I've seen bots support my move to mess up my self-bounce!

I think Spa-MAO is a very defensible move here.

[Mercy]

I think that move is defensible, too. Would any move have been strictly better than Spa - MAO?

[Chaqa]

I think the move is a bit silly. Wouldn't it be better to support MAO into Brest and also have Spain to MAO?

[SpaceDip]

I'm astonished how good these bots are.

http://webdiplomacy.net/board.php?gameID=247148

In this game I was worried that sending fleets through the med as Turkey to help prevent a solo would "trigger" the bots to think I was attacking them. It did not. The Italian bot allowed me into Adriatic, and to send a fleet all the way to MAO and neither the Italian or French units viewed it as aggressive which I think is pretty remarkable.

Tom, here is another evidence of the bots superior intellige. They are capable of planning their moves far far ahead. Just check the three Turkish fleets moves begining in 1907.

http://webdiplomacy.net/board.php?gameID=246858

[Octavious]

Well, it's certainly different. Quite an impressive achievement, although they're a long way short of human players as it stands. They have some strange blind spots in terms of tactics, and don't seem to understand the concept of stabbing. But still, pretty nifty nonetheless.

There is something rather empty about beating bots, though. The emotion just isn't there. No sense of triumph or regret, just a bit of a headache like the ones I got when given a draughts computer for Christmas as an ickle one.

[zultar]

Well, actually the second bot does have some idea of backstabbing. There's a fair chance that it would backstab its "allies" where that probability is next to zero for the first version. And yeah, for many instances, it doesn't do as well as some of top players, but for the majority of our users, this would give them quite a bit of a challenge. So far the feedback we have received has been overwhelmingly positive. I've finished several games with the new bots and they can be more aggressive and show more varied tactics (some of which will of course be bad) but probably about the same as humans would (since they are based on human-played games).

[LeonWalras]

Out of curiosity, were PPSC games included in the training dataset?

[jmo1121109]

I think the move is a bit silly. Wouldn't it be better to support MAO into Brest and also have Spain to MAO?

Am I missing a link to the game, because the move making sense is completely dependent on North Atlantic Ocean being occupied or not. If it is occupied EC support move into Brest from MAO and then move to MAO from North Atlantic Ocean. We also can't see Western Med to establish how devastating that move set would be. We cannot and will not suggest changes based off of screenshots. We need the gameID, the turn it happened, and the country to be able to look into it. Since the bot's also act off the previous turn, a single screenshot just doesn't help.

[jmo1121109]

Out of curiosity, were PPSC games included in the training dataset?

Yes, PPSC was. Of the classic games played on webdip 21,987 were WTA, 32,245 are PPSC, 1,728 are Unranked, and 2,216 are SOS.

[BobMcBob]

I think the move is a bit silly. Wouldn't it be better to support MAO into Brest and also have Spain to MAO?

Am I missing a link to the game, because the move making sense is completely dependent on North Atlantic Ocean being occupied or not. If it is occupied EC support move into Brest from MAO and then move to MAO from North Atlantic Ocean. We also can't see Western Med to establish how devastating that move set would be. We cannot and will not suggest changes based off of screenshots. We need the gameID, the turn it happened, and the country to be able to look into it. Since the bot's also act off the previous turn, a single screenshot just doesn't help.

The game was http://www.webdiplomacy.net/board.php?g ... #gamePanel, the turn was Autumn 1902. North Atlantic Ocean was unoccupied. Move to Western Med would have resulted in either getting Tunis or convincing Italy to abandon the Ionian to Turkey, probably the latter, which would have been bad.

Italy was focussed East and couldn't afford to let up in the East. The biggest threat to France here was England. As far as I can tell, this was the best move for France in this situation, as it could backfill the Atlantic in the case of English support.

[Octavious]

Well, actually the second bot does have some idea of backstabbing. There's a fair chance that it would backstab its "allies" where that probability is next to zero for the first version. And yeah, for many instances, it doesn't do as well as some of top players, but for the majority of our users, this would give them quite a bit of a challenge. So far the feedback we have received has been overwhelmingly positive. I've finished several games with the new bots and they can be more aggressive and show more varied tactics (some of which will of course be bad) but probably about the same as humans would (since they are based on human-played games).

Fair enough. My ally bot must have been in particularly trusting mood. I think it's worth empathising that I'm not a top level gunboater by any stretch of the imagination, but I guess I have a decent feel for the patterns of the game.

One potential weakness of the bots is that they didn't seem to recognise the danger of unit that had sneaked behind the front lines. In my game the French bot was doing a passable job of maintaining a stalemate line against me (Turkey) but allowed a fleet of mine to slip through a gap to the north Atlantic. This fleet was subsequently ignored, and sailed around happily picking up supply centres unchallenged until the game was won. It was a bit like lamping rabbits. The bots were utterly focused on the stalemate line and oblivious to everything else.

But yeah, for the most part they're pretty nifty