Data dignity treats AI training data as compensable labor, not a free byproduct

AI-generated writing.

Jaron Lanier’s “data dignity” (also called “data as labor”) argues that data always originates from someone’s work, so it should be compensated like labor rather than treated as a free, weightless resource. His proof that information isn’t weightless is physical, not moral: a bit that took no work to produce also disperses no heat, yet data centers now consume more energy than almost anything else humans build — so information “is physical or it’s nothing.”

The framing reframes AI itself: instead of a new autonomous entity, a large language model is a collaboration of everyone whose writing, code, or art it was trained on, compressed into something closer to a giant, involuntary Wikipedia. Once you see the model that way, the “black box” problem stops being mysterious — you open it by acknowledging the people inside it, which turns hallucination, bias, and jailbreak problems into questions about which human data sources are missing or overrepresented, rather than unexplainable properties of a black box.

Lanier’s broader point is that culture keeps forcing information pricing into two extremes — free, or so expensive it excludes everyone but the biggest players (he cites 5-axis milling software costing more than an apartment per year) — when a paid, affordable middle tier is what actually funds an ongoing economy of new creative and technical work.