TECHNOLOGY
SHRDLU
Winograd's 1968–1972 natural language program that conversed in English about a simulated world of colored blocks — the landmark AI demo whose apparent success concealed the structural absence of understanding.
SHRDLU was Terry Winograd's doctoral project at MIT, a program that could interpret English commands about a small universe of geometric objects—red blocks, blue pyramids, boxes on a table—and execute them with apparent comprehension. Users could type 'Pick up the big red block,' ask 'Why did you do that?,' and receive answers that looked like reasoning. To the AI community of the early 1970s, SHRDLU was proof that machines could understand natural language. To Winograd himself, SHRDLU became the most instructive illusion in computing—a system whose success within its closed world demonstrated exactly the conditions under which the absence of understanding becomes undetectable from the outside.
In The You On AI Field Guide
The blocks world contained perhaps two dozen objects. Every object had a shape (block, pyramid, box), a color (red, blue, green, yellow, white, orange), and a size (big, small). The spatial relationships were limited: on, in, behind, to the left of, to the right of. Every word had exactly one meaning. Every sentence