In the late 1960s, Terry Winograd at MIT built SHRDLU, an early natural‑language understanding program that let users converse with a computer to move and query objects in a simplified blocks world. Written in Micro Planner and experimental Lisp in 1968 on a DEC PDP‑6 with a graphics terminal, the system demonstrated interactive AI parsing using only about 50 lexical items.

The program’s world was deliberately tiny: blocks, cones and balls could be described with nouns, verbs like “place on” and adjectives such as “big” or “blue”. SHRDLU stored a short memory of recent actions, allowing pronouns and adjectives to resolve to previously mentioned objects, for example, and could answer feasibility questions about stacking or physics constraints.

SHRDLU’s success sparked optimism in AI research, showing that a limited grammar and world model could produce seemingly intelligent dialogue. Later extensions at the University of Utah added full 3‑D rendering, but attempts to scale the approach to richer, ambiguous environments quickly ran into limits, tempering early significant hype. The system remains a landmark demonstration of language‑grounded reasoning for future.