Facebook Says It Has Created A ‘Human-Level’ Board Game AI

Image from article titled

Facebook, or as we now call it Meta, announced today that its CICERO artificial intelligence has achieved “human-level performance” in board games. diplomacywhich is notable for the fact that it is a game built on human interaction rather than movement and manipulation. like chess.

This is a frankly disastrous trailer.

Never played diplomacy, and in case you’re wondering what the big deal is, it’s a board game first released in the 1950s that primarily involved sitting around a table (or resting in a room) and bargaining. There are no dice or cards that affect play. Everything is determined by how humans communicate with other humans.

So it’s a pretty bold claim by the creators of the AI ​​to say they’re playing on a “human level” in a game like this! One that handles game progress and status, and one that tries to communicate on a human level in a way that we understand and interact with.

Mehta roped in “Diplomatic World Champion” Andrew Goff to back their claim.he To tell, “Many human players start motivating themselves by softening their approach or getting revenge, but CICERO never does that. Its strategy, but not so ruthless as to annoy or annoy other players. ”

That sounds optimal, but as Goff puts it, perhaps that too Optimal. This reflects that while CICERO is performing well enough to keep up with humans, it is far from perfect.As meta himself says in a blog postCICERO “Sometimes produces inconsistent dialogue that can defeat its purpose.” And my own criticism is that all the examples of communication they provide (like the one below) make the psychopathic office worker look terrified.!!!” You will think you are a terrible person.

Image from article titled

Of course, the ultimate goal of this program is not to win board games.it simply uses diplomacy As a “sandbox” to “facilitate human-AI interaction”:

CICERO can only play diplomacy, but the technology behind this achievement is relevant to many real-world applications. For example, controlling natural language generation via planning and RL can ease communication barriers between humans and AI-powered agents. For example, today’s AI assistants excel at simple question-answering tasks such as forecasting the weather, but what if they could maintain long-term conversations aimed at teaching new skills? Or non-player characters (NPCs). Imagine a video game that plans and converses like a human, understands the player’s motivations, and adapts the conversation accordingly.

I may not be a billionaire Facebook exec, but instead of spending this much time and money improving AI assistants, no one seems to care more than AI research and company spending. Is it not? instead?