Facebook, or as we should now call them Meta, announced earlier today that their artificial intelligence from CICERO has achieved “human-level performance” in the board game Diplomacywhich is remarkable for the fact that it is a game built on human interaction, not movements and maneuvers, like chess for example.
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 1950’s that’s mostly played by people just sitting around a table (or breaking down into rooms) and negotiating about things. There are no dice or cards that affect the game; everything is determined by people communicating with other people.
So for the creators of an AI to say it plays on a “human level” in a game like this is quite a bold claim! One that Meta backs up by saying that CICERO actually operates on two different levels, one analyzing game progress and status, the other trying to interact with human levels in a way that we would understand and with whom we would communicate.
Meta has enlisted “Diplomacy World Champion” Andrew Goff to support their claims. He say, “Many human players will soften their approach or become motivated by revenge and CICERO never does. It just plays the situation as it sees it. So it is relentless when running his strategy, but it is not ruthless in a way that irritates or frustrates other players.”
That sounds optimal, but as Goff says, maybe at optimal. Which indicates that while CICERO plays well enough to keep up with people, it’s far from perfect. As Meta itself says in a blog postCICERO “sometimes generates inconsistent dialogue that can undermine its objectives,” and my own criticism would be that any example they give of the communication (as 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 horrible person.
Of course, the ultimate goal of this program is not to win board games. It’s just use Diplomacy as a “sandbox” for “enhancing human-AI interaction”:
While CICERO can only play Diplomacy, the technology behind this feat is relevant to many real-world applications. For example, controlling natural language generation through scheduling and RL could reduce communication barriers between humans and AI-powered agents. For example, today’s AI assistants excel at answering simple questions like telling you the weather forecast, but what if they could carry on a lengthy conversation with the goal of teaching you a new skill? You can also imagine a video game where the non-player characters (NPCs) can plan and converse like humans do – understand your motivations and adjust the conversation accordingly – to aid you in your quest to storm the castle.
I may not be a billionaire Facebook executive, but instead of spending all that time and money improving AI assistants, which no one seems to care about outside of AI research and corporate spending, we can’t just…people hire who i can talk to instead of?