SHRDLU was the 1968 MIT AI program that understood commands in a simulated blocks world.
Terry Winograd developed SHRDLU at the Massachusetts Institute of Technology. Users typed English instructions such as asking the program to move, stack, or identify colored geometric blocks. The system maintained an internal representation of the small virtual world and translated language into actions.
SHRDLU appeared impressive because it combined language processing, reasoning, and manipulation in one tightly controlled environment. It could answer questions about the blocks and explain some of its actions. However, its understanding depended heavily on the limited vocabulary, objects, and rules of that world.
The program became an influential demonstration of symbolic AI and natural-language interaction. Its success also illustrated a major limitation: performance in a carefully defined micro-world did not automatically transfer to the messy, ambiguous situations of everyday life. SHRDLU is therefore both a landmark and a classic example of the limits of early AI.