Timeline
SHRDLU was not an isolated event but the convergence of a science-fiction conceit, a research programme, and a particular machine. The chronology below traces the program from the cultural prehistory of its name, through its construction at the MIT Artificial Intelligence Laboratory between 1968 and 1970, to its publication, the critique that followed, Winograd's own turn from builder to critic, and the reconstructions that keep the dialogue running today.
-
1942
Fredric Brown publishes the science-fiction story "Etaoin Shrdlu", in which a sentient Linotype machine learns language and threatens to take over, only to be pacified by a turn to Buddhism. Decades later it would furnish the probable subliminal source for the program's name. On the name.
-
Mid to late 1960s
Carl Hewitt designs PLANNER, a procedural problem-solving language, at MIT. Marvin Minsky and Seymour Papert co-direct the AI Laboratory and its "microworlds" research programme, which holds that intelligence can be studied by building systems competent within deliberately restricted domains.
-
1966
Terry Winograd takes his BA at Colorado College before moving to MIT for doctoral study.
-
1968–1970
Winograd writes SHRDLU at the MIT Artificial Intelligence Laboratory, supervised by Seymour Papert. The program is written in MacLisp under the ITS operating system, with its deductive and problem-solving components expressed in Micro-Planner. The program.
-
July 1970
The Micro-Planner Reference Manual is issued as MIT AI Memo 203 (a revised 203A follows in December 1971), authored by Gerald Jay Sussman, Terry Winograd, and Eugene Charniak. Micro-Planner is the practical subset of Hewitt's PLANNER on which SHRDLU's reasoning is built.
-
February 1971
Winograd's thesis, Procedures as a Representation for Data in a Computer Program for Understanding Natural Language, is issued as MIT AI Technical Report 235. (The work is sometimes dated to its completion in 1970; the technical report carries a February 1971 issue date.)
-
1972
The thesis reaches a wide audience: as the book Understanding Natural Language (Academic Press) and as a complete issue of Cognitive Psychology 3(1), pp. 1–191. In the same year Hubert Dreyfus publishes What Computers Can't Do, arguing that artificial intelligence succeeds only within artificial micro-worlds and cannot capture the embodied, non-formalisable background of human understanding. SHRDLU becomes a central exhibit. Critique & legacy.
-
1973
Winograd leaves MIT for Stanford, where he will spend the rest of his career.
-
1983
Winograd publishes Language as a Cognitive Process, Vol. 1: Syntax. The planned multi-volume framework stalls, and no further volumes appear.
-
1986
Winograd, now working with Fernando Flores, publishes Understanding Computers and Cognition: A New Foundation for Design. Drawing on Heidegger and the phenomenological tradition, the book repudiates much of the rationalist programme that produced SHRDLU. The builder has become a critic. People.
-
1992
Dreyfus issues a revised and expanded edition, What Computers Still Can't Do (MIT Press), restating the critique against the longer history of symbolic AI.
-
2023
Patrick van Bergen publishes a complete replication of the canonical SHRDLU dialogue, reconstructing the system so that the historic conversation can be run again from end to end. Reconstructions.
-
The present
In the era of large language models, SHRDLU stands as the symbolic counterpoint to statistical and neural approaches: grounded but not general, where today's systems are general but arguably ungrounded. The Winograd Schema Challenge, a coreference test named in his honour, descends directly from the kind of pronoun resolution SHRDLU performed in its blocks world. SHRDLU and the LLMs.