Meta has announced what it calls a "breakthrough" in the specific field of gameplay AI: software called Cicero is the first AI to achieve "human-level performance in the popular strategy game Diplomacy." Diplomacy was originally a board game and has many official and unofficial digital successors. What makes this game an interesting choice is that the core of the game is negotiation, i.e., it is a multiplayer game in which players must constantly bargain with each other.

The article published by "Cicero" acknowledges that AI has defeated humans (fact check: Deep Blue lost to Garry Kasparov, but defeated him a few years later, after which IBM refused to give him a rematch). However, "a truly useful and versatile agent will need to do more than simply move pieces on the board," he said. Thus, Cicero intends to be able to negotiate, persuade, and cooperate with human players to achieve strategic goals, just as humans do.

Diplomacy has long been considered one of AI's grand challenges for precisely this reason. It requires understanding the motivations of other players, adjusting strategies on the fly, and ultimately winning them over to one's side. Cicero played an online version of this game at webDiplomacy.net, where he "achieved more than twice the average score of human players and ranked in the top 10 percent of participants who played multiple games."

In fact, "Cicero was so effective at negotiating with people using natural language in Diplomacy that they often preferred working with Cicero over other human participants."

Betrayal! Vile betrayal.

Part of the achievement is that Cicero was not built by the traditional self-playing reinforcement method (playing against itself and humans millions of times and analyzing the data) by which AI learns games. According to Meta, Cicero incorporates two main elements: "strategic reasoning as used in agents like AlphaGo and Pluribus, and natural language processing as used in models like GPT-3, BlenderBot 3, LaMDA, and OPT-175B"

Of particular importance is the ability of Cicero to recognize which players it needs to win over and to come up with strategies to win them over to its side. The software "runs an iterative planning algorithm that balances consistency and rationality of dialogue," predicts players' future moves based on their dialogue, and makes plans that incorporate those predictions.

Not taking over the world yet: of course, Meta's ambitions for this software go far beyond the old board game. The company sees the potential for this to have a major impact on AI chat assistants. For example, it will be able to conduct learning conversations and dialogues that teach humans new skills. [or imagine a video game where non-player characters (NPCs) can plan and converse like humans.

Now, this is kind of interesting. The Edge magazine may have been right about Doom. You can read more about the technical aspects of Cicero and his research paper here.