The people

People

SHRDLU was, formally, the work of one person: a doctoral thesis written by Terry Winograd at the Massachusetts Institute of Technology between 1968 and 1970. SHRDLU sat inside a dense and unusually generative milieu, the MIT Artificial Intelligence Laboratory, whose microworlds programme, programming languages and linguistic inheritance are visible throughout the code.

To read SHRDLU critically is therefore also to read the people whose ideas, tools and arguments it carries. Winograd is key, of course, but the program inherits Carl Hewitt's PLANNER, the Micro-Planner implementation he built with Gerald Sussman and Eugene Charniak, M. A. K. Halliday's systemic grammar, and the microworlds research agenda of Marvin Minsky and Seymour Papert. Its later reception is bound up with two critics, Hubert Dreyfus and Fernando Flores, the second of whom Winograd would eventually work with productively.

THE PRINCIPAL

The principal

Terry Winograd

Terry Allen Winograd was born on 24 February 1946. He took his BA at Colorado College in 1966 and went on to a doctorate at MIT, where he was supervised by Seymour Papert. (Papert, not Marvin Minsky, was Winograd's advisor, a point frequently confused in the secondary literature.) It was at the MIT Artificial Intelligence Laboratory, between 1968 and 1970, that he wrote SHRDLU.

The work was issued as MIT AI Technical Report 235 (AI-TR-235) in February 1971, under the title Procedures as a Representation for Data in a Computer Program for Understanding Natural Language. It reached a wider audience in 1972, both as a complete issue of the journal Cognitive Psychology (volume 3, number 1) and as the book Understanding Natural Language (Academic Press).

Winograd moved to Stanford in 1973. A planned multi-volume framework for language largely stalled, with only the first volume, Language as a Cognitive Process, Vol. 1: Syntax, appearing in 1983. By then his confidence in classical artificial intelligence had been shaken, partly by Hubert Dreyfus's phenomenological critique and partly by his encounter with the Chilean philosopher Fernando Flores. Winograd and Flores co-authored Understanding Computers and Cognition: A New Foundation for Design (1986), a book that turned away from the representational assumptions SHRDLU had embodied.

Winograd's later career lay in human-computer interaction and design, at Stanford and its d.school. He advised the doctoral student Larry Page, and joined Google in 2002. The Winograd Schema Challenge, a test of machine intelligence proposed in 2012 by Hector Levesque, a computer scientist at the University of Toronto, is named in his honour. The arc from builder to critic is itself a story worth reading closely. See the critique and legacy and the program for how this bears on the code.

THE MILIEU

The milieu

SHRDLU drew on the people, languages and theories that surrounded it at MIT and beyond. The cards below sketch the principal figures whose work is inscribed in the program or in its reception.

Marvin Minsky & Seymour Papert

Minsky and Papert were co-directors of the MIT Artificial Intelligence Laboratory and the chief architects of the microworlds programme, the research strategy of building artificial intelligence by radically restricting the domain in which a system must operate.

Papert, a mathematician and learning theorist, was Winograd's doctoral advisor. SHRDLU is in many respects the exemplary microworld, a deliberately small and well-behaved tabletop within which understanding could be made to appear. See the blocks world.

Carl Hewitt

Carl Hewitt designed PLANNER, a programming language for automated reasoning developed at the MIT AI Laboratory. PLANNER expressed goals and the procedures for achieving them as part of the language itself.

The deductive component of SHRDLU was written not in full PLANNER but in Micro-Planner, a practical subset of Hewitt's design. The language thus supplied the procedural problem-solving machinery on which the program's apparent reasoning depended.

Gerald Jay Sussman & Eugene Charniak

Micro-Planner, the working subset of Hewitt's PLANNER used in SHRDLU, was implemented by Gerald Jay Sussman, Terry Winograd and Eugene Charniak at the MIT AI Laboratory.

Their implementation provided the deductive substrate on which SHRDLU's planning and inference ran. Both Sussman and Charniak went on to substantial careers in artificial intelligence in their own right. See the program for how Micro-Planner figures in the code.

M. A. K. Halliday

The linguist Michael Alexander Kirkwood Halliday developed systemic grammar, an approach that treats grammatical choices as a network of meaningful options rather than as a fixed set of rules.

Winograd's parser, PROGRAMMAR, took Halliday's systemic grammar as its linguistic basis. SHRDLU is often cited as the first major computational application of systemic theory, embedding a particular account of language directly in procedure.

Hubert Dreyfus

The philosopher Hubert Dreyfus was the most influential phenomenological critic of microworld artificial intelligence. Drawing on Heidegger and Merleau-Ponty, he argued in What Computers Can't Do (1972) that human intelligence rests on an embodied, non-formalisable background that symbolic systems cannot capture.

SHRDLU was a central exhibit in his case: a system that succeeded only by confining itself to an artificial micro-world. See critique and legacy.

Fernando Flores

Fernando Flores is a Chilean philosopher and engineer whose work brought phenomenological and language-action perspectives to bear on the design of computational systems.

Flores was Winograd's co-author on Understanding Computers and Cognition (1986), the book in which the builder of SHRDLU turned decisively against the representational assumptions of classical artificial intelligence.

FURTHER

Further