Do Artifacts Have Politics?

In the AI Story

Winner wrote the essay against the prevailing assumption — shared by liberals and conservatives, engineers and policymakers — that technology was a neutral tool whose political character depended entirely on who used it and how. The Moses bridges example, drawn from Robert Caro's The Power Broker, became the canonical illustration because it was undeniable: the low clearances had no engineering justification, only a social one, and they operated silently, automatically, permanently to exclude the populations Moses wanted excluded.

The second category — technologies with inherent political requirements — was more controversial and more consequential. Winner argued that some technologies, by the nature of their operation, require particular political arrangements. Nuclear power's failure modes are civilizationally catastrophic, which means it requires centralized control, security hierarchies, expert authority structures, and limitations on democratic participation. The reactor does not negotiate with democratic theory. The physics imposes the politics.

Applied to AI, the essay's framework is uncomfortable. Large language models are amplifiers with architectural preferences — training data choices, optimization targets, deployment channels — each operating as Winner described: silently, automatically, without requiring user awareness. Luke Fernandez's 2025 paper Do AIs Have Politics? documented this directly: citation failure in LLMs is 'not so much a bug as a feature,' a political architecture that undermines intellectual accountability by design.

The essay's most durable insight is that depoliticization is itself political. When a technology is presented as a neutral tool whose effects depend on use, the politics embedded in its design are rendered invisible and therefore invulnerable to democratic contestation. Making the politics visible is the prerequisite for governing them.

Origin

Winner wrote the essay during the late 1970s at MIT, where he was engaging with the emerging field of science and technology studies and with political theorists concerned about the democratic implications of technological change. The essay appeared in the Winter 1980 issue of Daedalus devoted to 'Modern Technology: Problem or Opportunity?' alongside contributions from Lewis Thomas, Stephen Toulmin, and others.

It was reprinted in 1986 as the opening chapter of The Whale and the Reactor, where it framed the book's broader argument about the political philosophy of technology. Subsequent citation counts have placed it among the most-cited works in the social sciences.

Key Ideas

Two categories of political artifact. Deliberately political designs (Moses's bridges, McCormick's molding machines) and structurally political technologies (nuclear power, industrial agriculture) operate through different mechanisms but both embed power arrangements in material form.

Politics survive intention. The political arrangements encoded in an artifact's design persist long after the original designers are gone, long after the political context that produced them has been forgotten. The bridge has a clearance whether the bridge cares or not.

Depoliticization is a political operation. Presenting technology as neutral is itself a political act — one that protects existing power arrangements from democratic scrutiny.

The amplifier is not innocent. Winner's framework applied to AI reveals that the metaphor of the neutral amplifier (see Do Amplifiers Have Politics) performs exactly the depoliticization Winner spent his career contesting.

Politics are in the concrete. Moses's bridges did not require enforcement, guards, or signs. They operated physically, automatically, through the clearance height. AI's political architecture operates similarly: through pricing, training data, optimization targets, API access — infrastructure that cannot be appealed.

Debates & Critiques

The essay has been challenged on two fronts. Bernard Joerges's 1999 paper questioned the historical accuracy of the Moses bridges example, arguing that public transit restrictions on parkways were more complex than Winner claimed. Defenders, including Steve Woolgar and Geoff Cooper, pointed out that even if the historical details were imperfect, the conceptual point — that artifacts embed political arrangements — remained intact and had been confirmed across countless subsequent case studies. A second debate concerns the second category: critics argue Winner overstates the necessity of particular political arrangements for particular technologies, suggesting more configurational flexibility than Winner allows.