Facebook, or as we’ll call them now Meta, announced earlier today that their CICERO artificial intelligence has achieved “human-level performance” in the board game Diplomacywhich is known for the fact that it is a game built on human interaction, not moves and maneuvers (such as chess).
Here’s a frankly disturbing trailer:
If you’ve never played Diplomacy, and so you might be wondering what the big deal is, it’s a board game first released in the 1950s that’s mostly played by people just sitting around a table (or breaking into rooms) and negotiating things. There are no dice or cards that affect the game; everything is determined by people communicating with other people.
So for an AI creator to say it plays on a “human level” in a game like this is a pretty bold claim! One that Meta supports by saying that CICERO actually operates on two different levels, one crushing the progress and status of the game, the other trying to communicate on human levels in a way we would understand and interact with.
Meta has called in “Diplomacy World Champion” Andrew Goff to support their claims, who says “Many human players will soften their approach or they’ll start to be motivated by revenge, and CICERO never does. 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.”
It sounds optimal, but as Goff says, maybe also optimal. Which reflects that while CICERO plays well enough to keep up with people, it’s far from perfect. As Meta himself says in a blog post, CICERO “sometimes generates inconsistent dialogue that can undermine its goals”, and my own criticism would be that every example they give of the communication (like the one below) makes it look like a psychopathic office worker who is terrified that if they don’t end every sentence with!!! 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 to use Diplomacy as a “sandbox” for “promoting human-AI interaction”:
While CICERO is only capable of playing Diplomacy, the technology behind this feat is relevant to many real-world applications. For example, controlling natural language generation via scheduling and RL can ease communication barriers between humans and AI-powered agents. For example, today’s AI assistants excel at simple questioning tasks, like telling you the weather, but what if they could maintain a long-term conversation with the goal of teaching you a new skill? Alternatively, imagine a video game where the non-player characters (NPCs) can plan and talk like humans do – understanding your motivations and adapting your conversation accordingly – to help you storm the castle.
I may not be a billionaire Facebook executive, but instead of spending all this time and money on making AI assistants better, which no one outside of AI research and corporate spending seems to care about, we couldn’t just . .. hire people I can talk to instead?