Chelsea Finn wants robots to get better at learning

Robots and AI agents are still limited when it comes to learning new things. That’s exactly the problem that fascinates Chelsea Finn, a Stanford computer science and engineering professor. Her research focuses on giving autonomous systems frameworks to reason through complex tasks. Robots, for example, can be trained to break large problems into smaller steps. Even when they know the subtask, “translating that into low-level motor commands is incredibly difficult to achieve in robots,” Finn says. “I’ve actually spent more time on the low-level motor control because in many ways it’s harder.” Finn’s research, for example, has focused on developing hierarchical machine learning models that use chain-of-thought reasoning and language models to think through problems step by step. While agents and robots will be expected to take over many mental and physical work tasks, additional research is needed to make them more reliable, Finn says. “I think that’s the number one area where we need robots and any sort of autonomous agent to be improved before we see them in important workplace situations,” she says. “If these systems aren’t operating at a high level of reliability, it’s really hard to trust them in important situations.” Finn acknowledges that researchers will need to achieve some sort of breakthrough to create truly reliable agents. But she’s optimistic. She points to autonomous cars as having progressed to a high level of reliability. “Since I’m frequently in San Francisco, I’ve been in Waymos, and it seems like the reliability right now is actually really high, to the point that it seems like it’s higher than people [driving] in many scenarios,” she says. “So that gives me some optimism that we can get to really high degrees of reliability with machine learning systems.” This profile is part of Fast Company’s AI 20 for 2026, our roundup spotlighting 20 of AI’s most influential technologists,