The Future of Diplomacy May Be in the Hands of Bots (Kubrick Exhibition/Design Museum/Columbia Pictures)

Artificial Intelligence (AI) has proven time and time again that given enough data on a particular subject it can outperform even the best human players. great chess, go And even Starcraft 2 In recent years, players have increasingly found DeepMind to be wrong, suggesting that strategy is the AI’s strength.

However, some games require more than strategy. They demand soft skills, such as the ability to be diplomatic or duplicitous — skills that it’s easy to assume AI can’t easily mimic. This idea may also be human arrogance, as Meta has created a new AI bot called Cicero, who has become one of the world’s top 10 percent players in the popular online game Diplomacy – without ever using it. The human face. Meta recently spilled the beans on how this all happened in a research paper.

This begs the question, could Cicero promote more than strategic prowess? Can this new AI inform real-life diplomacy, even in war? Or at least build smarter customer-service bots that do more than guide us through FAQs on websites. That would be a good start.

How did a bot master the diplomatic arts?

Diplomacy, as the name suggests, is not just about European conquest, but about negotiating with other players as necessary to meet your own goals. To win, it is necessary to make temporary alliances with other players, coordinate movements and attacks.

Cicero's Logic (Meta)

Cicero’s Logic (Meta)

In other words, Meta wanted to teach Cicero not just the rules of the game, but the rules of human engagement: how to communicate clearly and draw humans into alliances. To do this, Cicero was trained on 12.9 million messages from more than 40,000 diplomacy games, so that it could understand how words affect actions on the board.

“Cicero can infer, for example, that he will need a certain player’s support later in the game, and then devise a strategy to win that person’s side—and also identify the risks and opportunities that player sees from their particular point of view. ,” says Meta.

With this training under his virtual belt, Cicero entered the 40 Games of Diplomacy hosted by Despite the bot sending 5,277 messages to humans, in 72 hours, Cicero achieved “more than double the average score” of the players, with only one player expressing concern that the bot was among their numbers after the match. Sometimes, he was able to explain the strategy to his flesh-and-blood friends, as captured in the second example below.

Example of Cicero's in-game chat (meta)

Example of Cicero’s in-game chat (meta)

Although duplicity was desirable in diplomacy, Cicero generally achieved his goals by being honest and helpful in his dealings with other players. This partly reflects how Cicero was modeled: dialogue was based only on the upcoming turn, not reflecting how it might change over the long-term course of the game.

The authors of the study acknowledged that the bot’s lack of exit might be partly due to the nature of the game Cicero entered, where moves were limited to five minutes to keep things calm. “occasionally sending messages that contained grounding errors, contradicted his plans, or were otherwise strategically subpar”, the authors believe were not due to the “time pressures imposed by the game, as well as the fact of humans”. Make the same mistakes occasionally.

Could bots soon be running the world?

So what does this mean for humans, other than the possibility that we’ll start losing to machines in another whole strand of games in the near future? Well, Meta believes this research could seriously improve chatbots in the real world.

“For example, today’s AI assistants excel at answering simple questions, like telling you the weather, but what if they could maintain a long-term conversation aimed at teaching you a new skill?” Meta asks in A Blog post Along with research.

“Alternatively, imagine a video game in which non-player characters (NPCs) can plan and converse like people — understand your motivations and adapt the conversation accordingly — to help you in your quest to get to the castle.”

So is this the end of human customer-service on Facebook or Amazon – and will we be able to tell the difference if we’re chatting with a next-generation banking bot?

It is a positive spin. The downside, of course, is that if this AI can trick gamers into thinking they’re playing with a fellow human, it can be used to manipulate people in other ways. Perhaps wary of such abusive use, while Meta has open-sourced the Cicero code, the company hopes that “researchers can continue our work in a responsible manner”.

In the same way that AI bots have adopted radical strategies for chess and Go (which have changed the way these games are played), perhaps Cicero could change the nature of diplomacy or war games in the real world? If the secret of Cicero’s success was the use of manners and positive politics, perhaps this is something humans can learn. Our smartest move may be to deploy the ultimate weapon: common courtesy.