Diplomacy and Artificial Intelligence

[edmgrim]

Human-level play in the game of Diplomacy by combining language models with strategic reasoning

An interesting article from Science.org applying Diplomacy to Artificial Intelligence. Perhaps there is an argument to be found here that puts the art of the back stab on its heels a bit?

[MasterGR]

Perhaps there is an argument to be found here that puts the art of the back stab on its heels a bit?

I was reading about this, but what do you mean that it puts the art of the back stab on its heels a bit? No contradiction between diplomacy being "an art" and AI being able to learn it in my view. After all, AI has been used recently to make literal art, like paintings, and has even won such competitions against humans: https://impakter.com/art-made-by-ai-wins-fine-arts-competition/. Turns out AI can learn to do most things that humans do pretty well.

[edmgrim]

....but what do you mean that it puts the art of the back stab on its heels a bit? ...

I was observing that, aside from some computational errors which were largely interpreted by human players as being indicative of typical mistakes other human players make in an on-line environment, the AI was programmatically unable to plan or execute a stab. Nevertheless, the AI outperformed the human players.

[Pootleflump]

Human-level play in the game of Diplomacy by combining language models with strategic reasoning

An interesting article from Science.org applying Diplomacy to Artificial Intelligence. Perhaps there is an argument to be found here that puts the art of the back stab on its heels a bit?

You realise this was written by our very own Markus aka Captain Meme, who did this research in collaboration with some of the top Diplomacy players in the world.

Not only the one of the smartest people I know, but also one of the nicest

GREAT ARTICLE!!

[Latanst1]

Very interesting!

If available, it would be fascinating to know the game titles and/or player name(s) used by CIcero in order to better observe the real life examples.

Latanst1

Human-level play in the game of Diplomacy by combining language models with strategic reasoning

An interesting article from Science.org applying Diplomacy to Artificial Intelligence. Perhaps there is an argument to be found here that puts the art of the back stab on its heels a bit?

[Perygl]

Markus, do you want Skynet? Because this is how you get Skynet!
.
.
.
.
(good job dude)

[MasterGR]

....but what do you mean that it puts the art of the back stab on its heels a bit? ...

I was observing that, aside from some computational errors which were largely interpreted by human players as being indicative of typical mistakes other human players make in an on-line environment, the AI was programmatically unable to plan or execute a stab. Nevertheless, the AI outperformed the human players.

Ah interesting, did the AI actually never stab? Did it win any solos?

[Captainmeme]

Human-level play in the game of Diplomacy by combining language models with strategic reasoning

An interesting article from Science.org applying Diplomacy to Artificial Intelligence. Perhaps there is an argument to be found here that puts the art of the back stab on its heels a bit?

You realise this was written by our very own Markus aka Captain Meme, who did this research in collaboration with some of the top Diplomacy players in the world.

Not only the one of the smartest people I know, but also one of the nicest

GREAT ARTICLE!!

Need to clarify here that I contributed to the project through playtesting and analysis, which is why I'm a co-author. I did not actually make the AI!

If anyone is interested in seeing gameplay, I recorded a game against 6 of them here: https://youtu.be/u5192bvUS7k

[Hindu Warrior]

....but what do you mean that it puts the art of the back stab on its heels a bit? ...

I was observing that, aside from some computational errors which were largely interpreted by human players as being indicative of typical mistakes other human players make in an on-line environment, the AI was programmatically unable to plan or execute a stab. Nevertheless, the AI outperformed the human players.

If it didn't stab then what is the point of winning and playing the game .....?

[Latanst1]

Though some may dispute the point, this was never meant to be a liars' game (Kissinger and other professional negotiators were fans of this game).

True diplomacy in this game or in real life requires skills of negotiation, mutually beneficial appeal, and at times persuasive misdirection. Betrayal of trust/outright lying (stabbing) is not needed to win and usually shows a poor player, IMHO, unless done in reaction to someone else's lies and betrayal. Once lied to, all rules of decorum in reaction are suspended!

Latanst1

....but what do you mean that it puts the art of the back stab on its heels a bit? ...

I was observing that, aside from some computational errors which were largely interpreted by human players as being indicative of typical mistakes other human players make in an on-line environment, the AI was programmatically unable to plan or execute a stab. Nevertheless, the AI outperformed the human players.

{B}If it didn't stab then what is the point of winning and playing the game .....?{/B}