The Anthropomorphic Trap

Edsger Dijkstra waged a sustained campaign against what he considered one of the field’s most damaging intellectual habits: the use of human-sounding language to describe what computers do. He did not merely find it sloppy. He found it dangerous—an active source of error that shaped how practitioners reasoned about their systems, what they expected from them, and how they responded when those systems failed in ways no genuine mind would fail. When you say a program “knows” a fact, you import the entire human concept of knowing—with its attendant ideas of belief, justification, awareness, understanding—and attach it to a system that has none of those things, merely a pattern of bits. The metaphor then does your thinking for you, and it thinks badly. It leads you to expect behavior appropriate to a knower from a system that manipulates symbols without comprehending them, and to be surprised when it fails in ways obvious to any genuine knower. The discipline