Facebook says it created a ‘human-level’ AI board game

Facebook, or as we’re supposed to call them now Meta, announced earlier today that their CICERO AI has achieved “human-level performance” in the board game. Diplomacywhich is a game based on human interaction, not movements and maneuvers (like, for example, chess).

Here’s a trailer that’s quite frankly heartbreaking:

If you never play DiplomacyIt’s a board game that was first released in the 1950s and is played mostly by people sitting around a table (or divided into rooms) and negotiating things. There are no dice or cards affecting play; Everything is determined by humans’ communication with other humans.

So for the makers to say that AI plays on a “human level” in a game like this is a very bold claim! One that Meta backs up by saying that CICERO actually works on two different levels, one that breaks down progress and game state, and one that tries to communicate with human levels in a way that we understand and interact with.

Meta impersonates “world champion of diplomacy” Andrew Goff to support their claim, who says “many human players will soften their approach or start motivating them with vengeance and CICERO will never do that. He plays the situation as he sees it. So he is ruthless in executing his strategy, But it’s not harsh in a way that upsets or frustrates other players.”

This sounds ideal, but as Goff says, maybe very better. Which reflects that while CICERO plays well enough to keep up with the humans, he’s far from perfect. As the Mita themselves say in a blog post, CICERO “sometimes generates inconsistent dialogue that can undermine his goals”, and my criticism is that every example they give of his communication (like the one below) makes him sound like a psychopathic office worker terrified that if they don’t finish with every sentence!!! You will think they are a terrible person.

Of course, the ultimate goal of this program is not to win board games. It is simply used Diplomacy As a “sandbox” to “promote human-AI interaction”:

While CICERO is only capable of playing diplomacy, the technology behind this feat is relevant to many real-world applications. Controlling natural language generation by mapping and RL, for example, could ease communication barriers between humans and AI-powered agents. For example, today’s AI assistants excel at simple question-answering tasks, like telling you the weather, but what if they could maintain a long-running conversation with the goal of teaching you a new skill? Instead, imagine a video game where non-player characters (NPCs) can plan and talk as people do – understanding your motives and adapting conversation accordingly – to assist you in your quest to storm the castle.

I may not be a billionaire Facebook CEO, but instead of spending all that time and money improving AI assistants, there seems to be something that no one cares about outside of AI research and company expenses, we can’t just… hire humans I can talk to at while that?