An AI named Cicero can beat humans in Diplomacy, a complex alliance-building game. Here’s why it’s big news

In rare good news from Meta, the company’s artificial intelligence researchers have just announced a scientific breakthrough. Their AI program named Cicero can now play the board game Diplomacy on a human level.

Now, before you get too excited, Cicero doesn’t play at a superhuman level. He’s been beaten by about 10% of the humans he’s played against. By comparison, in earlier stages of AI, like AI beating humans at chess or Go, humans have long since been completely overtaken.

DeepMind’s Go-playing AI is, for example, a “Go god” – according to Chinese grandmaster Ke Jie. Even the human go world champion would now lose 100-0 to the computer.

Diplomacy is a simplified and abstract game, involving rival armies and navies either invading or not invading each other’s territories. It is fair to say that it lacks the complexity and subtlety of the type of diplomacy undertaken in the halls of the United Nations.

Nonetheless, news of Cicero’s performance was in the eye of tech rivals such as Google, owner of DeepMind. The CEO and founder of DeepMind, Demis Hassabis, is an expert in diplomacy. He won the World Tag Team Championship in 2004 and finished 4th in the world at the 2006 World Championship.

I expect Hassabis to easily beat Cicero due to some of the limitations I will point out shortly.

The game of diplomacy

Diplomacy is what AI researchers call a “seven-player, zero-sum, deterministic game of imperfect information.” A seven-player game is much harder to solve than a two-player game like chess or go. You must consider the many possible strategies of not one but six other players. This makes it much harder to write an AI to play the game.

Diplomacy is also a game of imperfect information, as players make moves simultaneously. Unlike games such as chess or go, where you know everything about your opponent’s moves, Diplomacy players make moves without knowing what their opponents are about to do. They must therefore predict the next actions of their opponents. It also adds to the challenge of writing an AI to play it.

Finally, Diplomacy is a zero-sum game in which if you win, I lose. And the result is deterministic and does not depend on chance. Nevertheless, before victory or defeat, it always pays off for players to form alliances and team up on each other. Indeed, one of the real challenges of the game is to manage the informal negotiations with the other players before making simultaneous moves.

The main reason Cicero’s performance is a scientific breakthrough is that he can both play the game well and perform these informal negotiations. This combination of natural language processing and strategic reasoning is a first for any gaming AI.

Defeat Cicero

A careful reading of the article Meta published on Cicero in the journal Science offers some clues on how you can beat him.

First, Cicero is almost entirely honest (unlike the famous Roman whose name he bears). On the other hand, Diplomacy is a game of betrayal and dishonesty. Players offer to form alliances but often forfeit these deals instantly. Cicero no. It always plays straight.

Honesty is a surprisingly effective strategy in diplomacy, but not if your opponents know you will never betray them. That’s the problem. Cicero played anonymously, so his human opponents probably wouldn’t have solved this problem. But if you know this fact, it will be easy to take advantage of it.

Second, Cicero (this time like his namesake) is very talkative. Expert diplomacy players exchange twice as many messages with other players as non-expert players. The trick is to form alliances and reassure your opponents of your intention. Cicero also exchanges twice as many messages as the human players he tends to beat.

Of course, being a bot, it’s much easier for Cicero to manage six simultaneous conversations. And that, I would say, is an unfair advantage of being a computer in this scenario.

And then ?

It is unclear how Meta intends to build on this research. A computer that can reason about the beliefs, goals, and intentions of others, as well as persuade and build relationships through dialogue, is a powerful tool. It’s one that could easily be misused. Let’s not forget that several years ago, Facebook (which is owned by Meta) received much justified criticism for an experiment aimed at manipulating user emotions.

Still, it’s hard to say exactly what Cicero’s actual applications might be. After all, real-world diplomacy is neither zero-sum nor deterministic. Two countries can agree not to go to war, and both will win.

Then there are a host of random factors that can alter a result. The Spanish Armada, for example, lost more ships to unexpected summer storms than to enemy fire.

Whatever Meta intends, the breakthrough is another example of how big tech companies are taking control of the AI ​​race with billion-dollar investments that cannot be matched by the industry. audience. Cicero was produced by a team of over 25 researchers. Nobody working in a university has that kind of resources to solve a board game.

As an artificial intelligence researcher at one of these universities, I am conflicted. I remember a famous graffito in Pompeii that said

Suti Ciciiro vapla again

“You will love Cicero, or you will be whipped”.

This article is republished from The Conversation under a Creative Commons license. Read the original article.