Yejin Choi

Yejin Choi (born 1977 in South Korea) is one of the world’s leading natural-language-processing researchers. After a PhD at Cornell and professorships at Stony Brook and the University of Washington — where she also led projects at the Allen Institute for AI, including the ATOMIC machine common-sense knowledge base — she won a 2022 MacArthur “genius” Fellowship for teaching machines to reason about the unstated. In 2025 she was appointed a professor at Stanford and its Institute for Human-Centered AI.

Smart, stupid, and worth building

Choi’s 2023 TED talk, “Why AI is incredibly smart and shockingly stupid,” became a touchstone of the ChatGPT era: language models ace exams yet flub kindergarten logic because common sense arrives only as a byproduct of scale, not as a learning objective. She warned that extreme-scale models are so expensive that “we are now at the mercy of those few tech companies” able to train them, and asked whether AI without robust common sense can be truly safe — quipping that brute-force scaling is like reaching for the Moon by making the tallest building one inch taller at a time.

Crucially, her critique is a research agenda, not a protest: Choi has spent her career building the systems she dissects, and argues for smaller, more transparent, common-sense-aware models — democratized beyond a handful of corporate labs — rather than for slowing the field down.