SHRDLU was a natural language understanding program written by Terry Winograd at the MIT Artificial Intelligence Laboratory between 1968 and 1970. It grew out of his PhD thesis, submitted to the MIT Department of Mathematics, which was issued as MIT AI technical report AITR-235, “Procedures as a Representation for Data in a Computer Program for Understanding Natural Language,” in early 1971. This entry uses 1971 as the report date.
SHRDLU let a person carry on a typed English dialog with a computer that controlled a simulated robot arm in a “blocks world” of colored blocks, pyramids, and a box. A user could type commands like “Find a block which is taller than the one you are holding and put it into the box,” and the program would parse the sentence, figure out what it referred to, plan the actions, and carry them out. It could answer questions about what it had done and why, and it tracked context across the conversation.
What made SHRDLU important was that, within its tiny world, it tied together syntax, meaning, and reasoning in a way that looked like genuine understanding. It became the showcase example for the symbolic approach to language: represent the world and the rules explicitly, then reason over them. Its success in a restricted setting also exposed the central difficulty, that this hand-built approach did not scale to the messy open world outside the blocks.
Why business readers should care: SHRDLU is an early ancestor of today’s conversational AI, but it worked in the opposite way to modern systems. It relied on hand-coded rules and a closed world, where today’s large language models learn language statistically from vast text. The contrast explains why modern chatbots are far more general but also less predictable.