Meta has developed a board game-playing AI that can beat humans at Diplomacy, a strategy game requiring negotiation and conversation.
The AI named CICERO was built by Facebook’s parent company to learn how to play the game through trial and error without human players knowing they are negotiating with a machine.
To play the game, CICERO had to naturally converse simultaneously with six other players online, with hundreds of messages sent during the game.
“What impresses me most about CICERO is its ability to communicate with empathy and build rapport, whilst also tying that back to its strategic objectives. Its strategy informs its communication and its communication informs its strategy,” said Andrew Goff, Diplomacy World Champion.
CICERO has two components, a strategic reasoning planning engine to crunch the numbers and game status, and a natural language processing component for human dialogue and negotiation.
Researchers at Meta began by using a pre-trained AI able to analyse language. They then input data in the form of 12.9 million human messages from more than 40,000 games of Diplomacy.
“A lot of human players will soften their approach or they’ll start getting motivated by revenge and CICERO never does that. It just plays the situation as it sees it. So it’s ruthless in executing its strategy, but it’s not ruthless in a way that annoys or frustrates other players,” added Goff.
Over the course of 40 online Diplomacy games, the AI sent 5277 messages and deceived almost all players in thinking that they are communicating with a real human.
Meta envisions the AI to be able to advance human-to-AI interactions. Imagine a real-life example where voice assistants can maintain longer conversations with humans to teach new skills or provide honest advice. Or perhaps in video games where non-playable characters can plan and adapt conversations with more human emotion.
“It’s really cool that we managed to make AI’s that can beat humans at games, but ultimately, we want AI’s that can cooperate with humans, understand them, and work with them,” said Noam Brown, Research Scientist at Meta.
CICERO is a step forward toward this vision, using Diplomacy as a benchmark for progress.
There are, however, teething problems. According to New Scientist, CICERO ‘had a tendency to assume players were male’, though ‘Meta researchers developed language filters that caught most but not all toxic language’.
“It is important to recognize that CICERO also sometimes generates inconsistent dialogue that can undermine its objectives,” said Meta in a blog post.
To adapt and improve the AI, Meta has open-sourced the CICERO code and models for researchers to responsibly build on top of the current progress.