Terry Winograd

Some people change the course of history without the world noticing right away. Terry Winograd is one of them. In 1970, while finishing his doctorate at MIT under Seymour Papert, he created SHRDLU: a program that could receive instructions in everyday English, move blocks in a simulated world, and respond with coherent explanations. That sounds modest today, but at the time it was a seismic event. For the first time, a machine demonstrated something close to genuine understanding of human language, and the world of artificial intelligence was electrified.

His achievement was so monumental that in 1971 he became the first-ever winner of the IJCAI Computers and Thought Award, the most prestigious recognition in artificial intelligence for young researchers. He was 25 years old. But Winograd did something rare after that triumph: he stopped to doubt himself. In 1986, together with Chilean philosopher Fernando Flores, he published "Understanding Computers and Cognition," a book that questioned the very foundations of the representational AI he had helped build. It argued that machines don't "understand" anything in the human sense, and that technology design must revolve around people. That self-critique made him a respected voice in the philosophy of technology and human-centered design.

Winograd spent four decades as a professor at Stanford, where he founded the Human-Computer Interaction Group and was a founding faculty member of the Hasso Plattner Institute of Design (the famous "d.school"). In the 1990s, as co-principal investigator of Stanford's Digital Libraries Project, he advised a PhD student named Larry Page, and told him to follow his instinct to explore the link structure of the web. Page described that conversation in Fortune magazine as "the best advice I ever got." That idea became PageRank, Winograd is a co-author of the original 1998 paper, and PageRank became Google. One more detail worth savoring: the pronoun-resolution puzzles Winograd included in his 1972 doctoral dissertation are so subtle that in 2012 they inspired the "Winograd Schema Challenge," a benchmark for machine intelligence that large language models (like Claude) finally passed in 2019.

What moves me most about Winograd's story is that combination of boldness and intellectual honesty. He built something extraordinary and was then among the first to point out its limits. He founded disciplines, trained generations of researchers, and helped launch one of the most influential companies in the world, yet he kept asking the uncomfortable question: do machines truly understand, or do they just imitate very well? Every time you chat with Claude, SHRDLU sits somewhere in that family tree. And Winograd's question, whether understanding is real or just appearance, is the one that keeps us honest today.