webdiplomacy.net

New DiploStrats video!

https://www.youtube.com/watch?v=AWQFhYSD7h4

Interesting stuff.. were you expert 1, 2 or 3? Going to talk to Noam about whether we can run any of these bots here to mix it up a bit from the DipNet bots, and provide some varying difficulties etc.

I was Expert 3. Interestingly, looking at the games from the outside, Diplodocus-Low seemed stronger to me! I think that was because when you're not actually playing against it, it playing in a very human-like way seems less important (since that's only really relevant if you're trying to spot and counter it).

Really enjoyed the vid. Definitely wonder what 6 Diplodocus-High or Low vs. 1 Human would be like.

Very interesting. Didn't know until this vid and some further digging that yolo5 (a user that I ran into multiple times last summer, in non-tournament games) was the bot. Wish those games were unranked like the tournament ones were...

Aristocrat wrote: ↑

Fri Oct 14, 2022 12:16 am

Very interesting. Didn't know until this vid and some further digging that yolo5 (a user that I ran into multiple times last summer, in non-tournament games) was the bot. Wish those games were unranked like the tournament ones were...

We were asked to keep certain bot accounts under wraps to ensure no bias in people's play, which I know is a bit questionable but I think justifiable. Happy to retroactively alter the scoring if you want to let me know the games to look up

kestasjk wrote: ↑

Fri Oct 14, 2022 3:25 am

Aristocrat wrote: ↑

Fri Oct 14, 2022 12:16 am

Very interesting. Didn't know until this vid and some further digging that yolo5 (a user that I ran into multiple times last summer, in non-tournament games) was the bot. Wish those games were unranked like the tournament ones were...

We were asked to keep certain bot accounts under wraps to ensure no bias in people's play, which I know is a bit questionable but I think justifiable. Happy to retroactively alter the scoring if you want to let me know the games to look up

It doesn't really matter that much. I suppose the best might be to refund games where Yolo5 solo'd, but not where others did (or draws). That said, games like this https://webdiplomacy.net/board.php?gameID=364950 (Yolo5's biggest win) were due to poor orders from humans (Swordsman, Tjappator) more than elite bot play. Still, it stings to lose a few hundred coins to a computer that knows the best technical moves each turn.

Very interesting video, Meme.

Do you think the fact that the AI went hard after Scandinavia is a peculiarity of gunboat or do you think it's revealing a bit of underappreciated tactics that could translate to full press Diplomacy too?

Short answer - I'm not sure.

Long answer - I gained an appreciation for just how strong a secure Scandinavian position is from these games, and that definitely still holds in Full Press. It basically has all the advantages that the UK homeland has, but it's much much more difficult to backdoor - all the provinces are SCs except for Finland, so if you already have Scandi under your control it's tough for any enemy to involve their armies in their attack on you.

It had three strategies it would use pretty frequently (iirc). As Germany, it's the one shown in the video but generally with A Kie instead of F Kie. As Russia, it would play defensively on the southern front and put three on Norway (Swe/Fin/StP). And as England it would fasttrack an attack by going into Ska in 01 as well as convoying into Nwy.

All of these approaches share a key principle - you ally with a certain player (France as E/G, Turkey or Austria as R) while committing so hard to the region where you're looking to pick up SCs for yourself that you basically don't help your ally at all. Moving Nth-Ska in 01 for example means there's no way you can help France with Bel until F02.

I think as some of these powers you can get away with that (Ger isn't really expected to help France early in an FG, for example) but it's very dicey as others, especially in Full Press. Turkey should try to form an AT if Russia is giving them no support. France should stab England if England is not helping them punch through the Bel/Ruh/Mun line. The bot very much takes advantage of the fact that these stabs need to be immediate, and players are hesistant to immediately stab someone who has shown them favour in gunboat.

So I guess conclusion - taking Scandinavia is very much worth it, but doing it in the exact way the bot does in Gunboat is probably not going to work in high level FP.

Also very important to note - the bot played under Sum of Squares scoring in the games in this tournament, so having a power advantage over all other players in the region had a very distinct advantage.

Any strategy in which you sacrifice relationships with all players in your region for an early power advantage is going to be much worse in DSS FP, unless you get significant help from the other side of the board.

Interesting. I'm definitely skeptical of Skag working out for England that early in FP because of the hairiness of that 3-way but maybe there's a lesson in here somewhere about Russia opening north. I think it's generally less fashionable to open north because of the issues you pointed out but maybe it's worth exploring more if the right conditions are there for it.