Do facial-recognition systems have politics? A critical analysis of automated facial recognition through the course's lenses

goalThe analytical goal

Choose one contemporary technology and analyze it using the course's central premise: that technologies are not neutral forces of progress but human creations that reorganize power, and that they are best understood as contested terrain rather than as engines that determine history. The deliverable is an argument, not a report. The reader should be able to disagree with your thesis — which means it must be a claim, not a description.

This worked example takes facial recognition because it lets every major course framework do real work: Winner's question of whether an artifact "has politics," Benjamin's refusal of the progress-only story, Headrick's account of technologies of control, and the historiographical habit of reading primary sources against scholarly ones. (Social-media recommendation algorithms make an equally strong subject; see §7 Extensions.)

sessionsSessions and skills exercised

The project deliberately reaches across the syllabus rather than living in one week. It draws its core lens from Sessions 1–2 (Situating the Machines / "Do Artifacts Have Politics?"), its model of technologies-of-control from Sessions 6–9 ("Tools of Empire" and Empire & Mobility), its institutional account of state-and-research power from Sessions 21–22 (the Military-Industrial-Academic Complex), and its grounding in the digital age from Sessions 25–27 (Dominance of the Digital / Computers: A Human History). It practises the course's signature skills: building a layered argument, sorting primary from secondary sources, and naming the determinist fallacies that simplistic tech writing relies on.

deliverableThe deliverable

A roughly 3,000-word argumentative essay with: a clear thesis stated up front; a multi-part argument in which each section advances a sub-claim backed by named evidence; explicit engagement with at least two serious counter-arguments; a reflection on power and ethics; and a full reference list distinguishing primary from secondary sources. No code, no models — the work is interpretive and historical, exactly as the course requires. The structure below maps one-to-one onto sections 2 through 8 of this page.

The brief

Facial recognition is routinely sold as a neutral convenience: it unlocks a phone, speeds a border queue, "finds" a suspect. The dominant public story is a progress story — better cameras, better algorithms, safer streets. The course trains us to distrust exactly that framing. The brief, then, is to ask of FRT the question Langdon Winner asked of Robert Moses's low bridges over the parkways of Long Island: does this artifact, in its very design and deployment, settle a political question about who is seen, sorted, and controlled — and on whose terms? [1]

The research question

In what sense does facial-recognition technology "have politics," and what does reading it through Winner and Benjamin reveal that the progress narrative conceals? The question is historical as much as it is contemporary: FRT is the latest instance of a long lineage of technologies of identification and control, from the colonial census and the fingerprint file to the punch-card tabulator.

Note how the thesis is falsifiable: a reader could argue FRT is a neutral accelerant, or that its politics are contingent rather than inherent. Those are the counter-arguments §5 must answer.

3.1Move one — Winner's first sense: design settles disputes

Winner's "Do Artifacts Have Politics?" opens with the claim that some technologies embody political choices in their physical arrangement — Moses's bridges, built low so that buses carrying the poor and Black could not reach Jones Beach, encode a social policy in concrete [1]. FRT works the same way at the level of design decisions that look merely technical. The choice of training data, the chosen threshold for a "match," the decision to deploy at a protest versus a passport gate — each settles a dispute about who bears the risk of error.

Test this against the artifacts-have-politics demo: FRT belongs in the "flexible — could have been built otherwise" bin at the level of any specific system. That is move one. Move two argues the second, harder half of Winner's claim.

3.2Move two — Winner's second sense: inherent political requirements

Winner's deeper and more contested claim is that some technologies are "inherently political" — they require, to function at all, a particular arrangement of power [1]. His example is the atom bomb, which demands a centralized, secretive, hierarchical command structure regardless of who owns it. The argument here is that mass facial recognition is similarly structured. It cannot operate without three things: continuous image capture of public space, a centralized database against which faces are matched, and an authority empowered to act on a match. Each of these concentrates power upward and asymmetrically.

3.3Move three — the long history: FRT as the latest tool of control

The course's chronological method insists we resist treating any technology as unprecedented. FRT belongs in a lineage. Daniel Headrick's The Tools of Empire showed how nineteenth-century steamships, the telegraph, quinine and the repeating rifle did not merely accompany imperial conquest but made its scale possible — they let a small center see, reach, and control a vast periphery [4]. Identification technologies did the same work of legibility: the colonial census, the fingerprint archive, and later the Hollerith punch-card tabulator that mechanized national registration all rendered populations readable to a central power.

This is where the project earns its place in a history course. Treating FRT as a brand-new problem is itself a determinist move (the "this changes everything" fallacy the determinist-fallacy demo trains you to catch). Reading it as the newest instrument in a long history of legibility-and-control technologies — the move Headrick and the tools-of-empire demo model — sharpens the analysis rather than dating it.

3.4Move four — Benjamin's dialectic: the document of barbarism

The course's framing epigraph is Benjamin's: "there is no document of civilization which is not at the same time a document of barbarism" [8]. Applied here, the move is to refuse to let the achievement and the cost be narrated separately. The same system marketed as civilizational progress — faster borders, found children, unlocked phones — is, in the same gesture, the infrastructure of a surveillance order whose costs fall on those already most watched and least able to opt out.

Run the case through the civilization/barbarism demo: as you slide attention toward the achievement, the cost does not disappear — it is structurally entailed. That is the dialectical reading the thesis claims.

Method

The argument is built by close reading rather than data collection. Scholarly works (Winner, Zuboff, Headrick, Browne, Benjamin) supply the interpretive frameworks; primary materials (a vendor's marketing claims, the ACLU's litigation record, journalistic documentation of wrongful arrests, corporate transparency reports) supply the contemporary evidence those frameworks are tested against. The method is to let a framework make a prediction — Winner predicts inherent political requirements; Benjamin predicts an inseparable cost — and then ask whether the primary record bears it out.

Reading sources critically

A vendor's claim that a system is "99.9% accurate" is a primary source: it tells us what the company wants believed, not what is true in the field, and its silence — accuracy for whom, under what lighting, against which database — is itself evidence. Buolamwini and Gebru's audit is a secondary, scholarly source that supplies exactly the disaggregation the vendor omits. Holding the two against each other is the analytical engine of the project; neither alone would suffice.

Synthesis