Apple's Robot Struggles Show AI's Demo-to-Reality Gap

Apple's robotics team has spent more than a year trying to ship a home robot, and people close to the project say the basics still aren't working. That's the headline problem, but it points to a wider pattern this week: AI systems keep running into the gap between demo and deployment. Researchers are now borrowing from game theory, the branch of math that predicts how rational actors respond to incentives, to try to keep AI agents from going off script.

The idea, sometimes called reverse game theory, flips the usual question. Instead of asking how an agent will behave under a given set of rules, it asks how to design the rules so the agent behaves the way you want. MIT Technology Review's Amit Katwala lays out the cake-cutting logic behind it: give one party the knife and the other first pick, and suddenly fairness is in everyone's interest.

Whether that logic holds up once you're dealing with a large language model instead of two kids and a cake is the open question. Meanwhile, in Washington, a different kind of control experiment is playing out, this one involving a presidential order and the word "intelligence" itself.