PERSON

Terry Winograd

The computer scientist who wrote SHRDLU, the celebrated early AI program that appeared to understand English, and then spent the next half-century explaining—with more care and less rancor than anyone else in the field—why it was an illusion, and what genuine understanding requires.

There is a particular kind of authority that belongs only to the person who built the thing and then walked away from it. Terry Winograd has it. As a graduate student at MIT between 1968 and 1970 he wrote SHRDLU, a program that held a conversation in plain English about a tabletop of colored blocks, took instructions, asked clarifying questions, and remembered what it had done—and was received as one of the great early triumphs of artificial intelligence. It convinced a generation of serious people that machine understanding was close. Then the man who wrote it spent the next half-century explaining, with more care and less rancor than anyone in the field, why he had come to believe it was an illusion—not a fraud, but an illusion about where the understanding actually lived. His deepest question was never “Can a machine produce the outputs of understanding?” but “Where does the understanding come from,

Keep reading with YOU ON AI