SHRDLU was Terry Winograd’s 1968 computer program for manipulating objects in a simulated blocks world through natural language.
Winograd developed SHRDLU at MIT as a research demonstration of language understanding, reasoning, and action. Users could type commands and questions about colored blocks, pyramids, and boxes. The program parsed the language, represented the objects and their relationships, and directed a virtual robot arm to perform permitted actions.
SHRDLU impressed researchers because it could answer follow-up questions and use context within its limited world. For example, a command involving “it” could refer to an object mentioned earlier. Its success came partly from the tightly controlled environment, where the vocabulary and possible actions were sharply restricted.
The program is sometimes presented as proof that general language understanding had already been solved. It was not: its abilities did not transfer reliably to the open-ended real world. SHRDLU instead became a landmark example of both the promise and limits of early symbolic AI.