Meta researchers create AI that masters Diplomacy, tricking human players

Cicero is the first artificial intelligence to achieve human-level performance in Diplomacy. Cicero has obtained a certain mastery of language necessary to win the game in order to be a notable achievement.

Before Deep Blue beat Kasparov at chess, board games were used as a measure of success. AlphaGo defeated Lee Sedol in 2015. Both of those games follow a set of rules that are easy to understand.

Social skills are involved in a large part of Diplomacy. It's a difficult task for a computer player to win a game. Meta asked, "Can we build more effective and flexible agents that can use language to negotiate, persuade, and work with people to achieve strategic goals like humans do?"

The answer is yes. Cicero learned how to play Diplomacy on the internet. It became a master at the game and was ranked among the top 10 percent of people who played more than one game.

Meta created Cicero by pulling together artificial intelligence models for strategic reasoning and natural language processing and rolling them into one agent. Cicero looks at the state of the game board and the conversation history in order to predict how other players will act. A language model that can generate human-like dialog allows it to coordinate with other players.

Diplomacy-playing bot, provided by Meta.">

Cicero's natural language skills are called a controllable dialog model by Meta. Cicero pulls from a lot of internet text. Over 40,000 human games on webDiplomacy.net were pre-trained on text from the internet in order to fine tune the dialogue model.

The model was able to understand the ins and outs of a complicated game. "Cicero can deduce, for example, that later in the game it will need the support of one particular player," says Meta, "and then craft a strategy to win that person's favor--and even recognize the risks and opportunities that that player sees from their particular point of

The title of Meta's research was "human-level play in the game of Diplomacy by combining language models with strategic reasoning."

Meta suggests that Cicero's research could be used to "ease communication barriers" between humans and artificial intelligence, such as teaching someone a new skill. It could be used to power a video game where the person in charge of the game can talk like a human.

Depending on the context, this technology could be used to impersonate people and trick them into thinking they are dead. Meta hopes that other researchers can build on its code in a responsible manner and that it has taken steps to detect and remove toxic messages in this new domain.

Meta gave a detailed site to explain how Cicero works, as well as open-sourced Cicero's code. Fans of online diplomacy may need to be careful.