Human computation is the idea of deliberately harnessing human effort - usually in small, repeated units, at large scale - to solve problems that computers cannot yet handle reliably. The term was popularized by Luis von Ahn, whose CMU work treats human time and judgment as a computational resource that can be channeled through software. The reCAPTCHA system is the canonical example: the 2008 Science paper showed how millions of routine CAPTCHA solutions could be aggregated to transcribe scanned text at over 99 percent accuracy, doing useful work as a side effect of a security check.
The pattern has several flavors. Some systems, like reCAPTCHA, extract work invisibly from an activity people do for another reason. Others, like von Ahn’s “games with a purpose,” wrap the task in something enjoyable so labeling images feels like play. And some, like Amazon Mechanical Turk, pay people directly to complete small “Human Intelligence Tasks.” All share the same structure: break a hard problem into tiny human-sized pieces, collect many independent answers, and use redundancy to ensure quality.
Human computation is the conceptual backbone of how AI training data gets made. Labeling images, ranking model outputs for reinforcement learning from human feedback, moderating content, and transcribing audio are all human computation at industrial scale - the foundation that makes “automated” systems work.
Why business readers should care: nearly every AI product depends on human computation somewhere in its lifecycle. Recognizing it as a designed system - with costs, quality controls, and labor conditions - helps separate genuine automation from work that has simply been moved out of sight.