Yejin Choi

There are things a five-year-old knows without being taught: you can't fit an elephant in a car, if it's raining outside you'll get wet, people feel sad when they lose something they love. For AI, those things were a mystery for decades. Yejin Choi decided that had to change. This Korean-American researcher, educated at Seoul National University and with a PhD from Cornell, specialized in something the AI community was almost entirely ignoring: commonsense reasoning, that capacity to understand the world that humans take for granted.

Her most well-known contribution came with ATOMIC (2019), a commonsense knowledge base with over 300,000 assertions like "if someone bakes a cake, they probably want to celebrate something." Then came WinoGrande (2019), a benchmark of 44,000 questions designed to test whether AI models genuinely understand context or just memorize patterns. And COMET (Commonsense Transformers), a model that learned to reason about everyday situations using those databases. These works earned Best Paper and Outstanding Paper awards at ACL, EMNLP, NeurIPS, ICML, ICCV, CVPR, and AAAI, a collection of eight recognitions that few researchers achieve in an entire career.

In 2022, the MacArthur Foundation named her a Fellow (popularly known as the "genius grant"), an $800,000 award given without conditions to people whose work promises to transform a field. She was also named one of Time's 100 Most Influential People in AI in 2023 and an AI2050 Senior Fellow in 2024. She was the Wissner-Slivka Professor at the Allen School at the University of Washington and Senior Director at the Allen Institute for AI (Ai2), where she led the MOSAIC team. Today she is a Professor at Stanford and Distinguished Scientist at NVIDIA Research.

To me, Yejin Choi's lesson is about what we overlook because it seems obvious. Everyone assumed that getting AI to reason with common sense was secondary, something that would come on its own. She asked: what if it doesn't come on its own? What if we have to build it explicitly? She was right. Every time an AI assistant responds in a way that's coherent with your actual situation, not just with the words you typed, there's something of her work in that.