Although we're down temporarily as we're experiencing a huge strain on our servers, we hope to see some of you in the near future!

We're all up and ready to go! The bots are excited to play wth *you*! Again, hope to see you there and have a great time!

Define ready. The site still lags on and off. The bots are really good though, they provide a nice challenge.

How do you negotiate with bots?

@champ - you don't. Consider it gunboat.

Again, this game is called Diplomacy, not Tactics.

@champ and @gopher.

The bots are currently only usable in gunboat, not in full press. Using them in press is a further multi-year project. They can correctly read gunboat alliances, and will recognize signals such as support holds. However they only have a one-turn memory. If you stop a support hold for a year, they'll take it as you're no longer an ally.

And?

Meaning you can conduct diplomacy with them, in a fashion. Its not straight tactics

You miss my point. Gunboats are not this game. "Bots" are not playing Diplomacy.

Will a bot stab an ally? When and how?

Also, can the bot perceive its ally turning against them, through troop movements elsewhere, while still doing support holds?

I'm willing to believe for the point of argument that a bot can look for a rock to pick up, but can it say "nice doggie" while doing so?

What a disturbingly acute analogy.

Are people unfamiliar with Will Rogers?

The bots work in 1 v 1’s right?

Can a mod set 7 bots against each other on classic to see which country is actually the best?

@gopher - ah, I understand then. You're of the opinion that gunboat isn't real diplomacy. I used to be like that.

@alex - yes, they can work on any classic map that doesn't have special rules, so they can do 1v1s. And the bots played hundreds of games against each other to train, but I wouldn't give much stock to any results from it.

This is rather definitional. A Diplomacy game with no diplomacy is not Diplomacy. It is Tactics. However, the tactics within this game are purposefully VERY simplistic because the game is Diplomacy.

The 'bots should of course be called Skynet 1, etc. "We're here for your playing pleasure" ;) ... whilst muttering into the ether 'Diplome THIS...' as they launch their attacks.

If a player CDs, the position should be taken by a Skynet... pretty soon there will be vast regions of maps uninhabited by sentient beings or me where Skynet 'bots roam chewing up supply centres...

So I have to say, I'm a bit displeased to see such flippant dismissal of the bot project, especially by someone officially representing this site.

This is the largest project that has happened in the diplomacy space online since it's inception, with a University putting in tens of thousands of developmental dollars into a project that massive benefits to the Diplomacy community as a whole. Their work is groundbreaking as being the first Machine Learning standard for a non-cooperative game and the bot can, with a decent degree of skill and accuracy, understand alliance offers in gunboat games and when to take/reject them. Its offers of support have a higher success rate then human players.

The statement that "the tactics within this game are purposefully VERY simplistic because the game is Diplomacy" is without question the most inaccurate I've seen since the bot's release. Tactics in diplomacy are astonishingly complex and coming from someone with 3 classic game completed, the statement takes on an extra level of absurdity. Becoming skilled in tactics in the game takes a decent amount of work, and the bots will be an amazing tool for people to work up to that point.

The bot work here is setting the stage for future projects which will eventually lead to a press version of the bot being able to communicate with others. See existing press technology like (https://talktotransformer.com/). It also makes diplomacy as a whole competitive with other gaming sites and apps which offer instant gratification in the form of opening a site/app and having immediate available game play options which has resulted in a falling off of new people into the diplomacy hobby.

It's also discouraging and frankly, insulting for the Diplomacy community to be dismissive of the work put in by outside groups taking the time to work with our communities for free, to deliver to us, improvements to the game. That type of collaboration with outside researchers is rare and should be appreciated and encouraged.

Not to mention that it's also insulting to myself and other devs who contribute towards the overall code base that both sites share.

So to sum it up, the bots are a massive technological breakthrough, immensely beneficial to the Diplomacy hobby, lay the groundwork for future press communicating versions, and the largest research collaboration in the Diplomacy hobby to date. So if you don't have anything nice to say about them, kindly be quiet.

One (of many) potentially useful application of the bots would be playtesting variants for balance. The bots would be able to far more quickly go through variants, giving their creators and users additional statistics. For this reason alone, we should be supportive of this project.

Although reading the paper it does seem the current bots were trained on existing human classic diplomacy games, so perhaps the variants would have to be similar for now? I'm curious as to how they would deal with even a simple change like Fleet Rome.

Well said Jmo. Do you know if this project absorbed any of the developments achieved by the Daide Project? <daide.org.uk> Their champion bot, the Albert bot, was not only able to play in any map the were put in, but, after playing or watching a human against six bots several times in both platforms (daide server and webdip) I have to say the Skynet bots are not superior to the Albert bots in many ways, e.g.: ganging up the leader when he threats soloing (Alberts play to stop solos and shift alliance to make it happen); change their set of moves when the chosen set doesnt work (i see skynet bots repeating their whole set of orders for years in a row for no gain while things are happening in the other side of Europe ); realize animosity from moves that don't actually takes centres from them (Albert); adjust their strategy upon what happens around (a sum up of all the above).
I know the bots are not under webdip dev team control, but hey, could you point the Albert level in Daide playform to the actual developers? tx.

Challenging people's credentials is always the mark of someone with a strong case.

Trolling a thread celebrating an incredible development and accomplishment for the online Diplomacy community and speaking out on poorly informed opinions without any credibility is likewise always the mark of someone with a strong case.

Exciting project! Can't wait to try it out. AI development in Diplomacy has always been quite niche and limited (even when it's reasonably good like the DAIDE project), so putting it on the webDip family is nothing short of amazing.

@CCR - These bots were trained against top level Albert, as you can see in section 6.2 of the paper: https://arxiv.org/pdf/1909.02128.pdf

DipNet solos in roughly 30% of games against 6 Albert AIs, whereas 1 Albert AI solos about 6% of games against 6 DipNets. There are definitely some aspects Albert does better at than DipNet, but overall DipNet is far stronger.

This is pretty fascinating. I'd be curious to see how the bots performed in non-gunboat situations. I bet they could be pretty good. I have played against (and with) players that I presumed were not native speakers of English. I bet a bot could be amazing. The tactics a bot would need are not complicated. Identical situations arise time and time again. Combining a competent machine focused upon passing a Turing test with the simplistic (relatively) tactics that they need would be the interesting test.

Those are very interesting points, TCB.

Wow. So, if they are taught what A. already knows about holding lines, teaming up at the right time, silent moves, and the others' goals, then they become our Deepest Blues.

Ye ghods, bo_sox48, an admin! I guess "incredible achievement" and "community" rather give it away... DipNet vs SkyNet would be interesting ... Seriously good to see you in here. A serious use of the bots would be to have them sub in for a CDing player, which would get rid of the kamikazi low-RR stand-in...

I am hoping the tech will be available for me to be uploaded when I die. If so, I will gladly take on a slew of full communolication Dip games at that time!

Going off what Chumbles said; you could have the bots enter orders for players who would have otherwise NMRed. There are several problems with this, but it could be an interesting idea to explore.

"Coming soon will be an option on all games (starting with gunboats but eventually rolling out to all games) called "Allow Bot CD Ordering". Which will allow a bot to submit orders for, but not takeover a position for someone in CD." - jmo1121109

So yeah, this was already an idea they'd been working on. It's definitely a cool idea. I like it.

Cool.