Study Guide

• Know author + title, main arguments, key terms, cases, and the HCE toolkit.
• Essay habit:
  1. What is the question asking for?
  2. Define the concept precisely.
  3. Explain the case with context.
  4. Apply the concept to the case. Don't just list.

Based on his argument in “Do Artifacts Have Politics,” how would Langdon Winner analyze the racial passbooks and race tests used by Apartheid South Africa?

In “Do Artifacts Have Politics,” Langdon Winner seeks to explain how technologies (“artifacts”) can have political qualities. For Winner, “politics” means the ability to shape society, not just in elections or political parties, but in power and “social order.” One example was the bridges in New York that were so low they stopped buses (and therefore the poor) from traveling to the beach. Another was factory automation that eliminated skilled workers and undermined the power of the labor union. Still another was nuclear power, which unlike solar power seems to depend on strong forms of political authority to prevent disastrous meltdowns. Some of these are political in their specific context, like bridges in New York or automation in a fight with the union, and others seem to inherently affect any society that has them, like nuclear power. But the overall point is that technology affects what is possible in society and what is required in society, how power and authority functions in society, and who has that power.

I believe Winner would see the passbooks and racial tests of Apartheid South Africa as deeply bound up with the white supremacy of that society. Apartheid South Africa was built around strict divisions between racial groups (white, indian/asian, colored, and black) and domination of whites over other groups. The passbooks and racial tests were a necessary part of this power structure because they determined what group a person belonged to and where they could legally live or work. The tests were very invasive and included things like appearance, background, or even how curly your hair was (the “pencil test”). People were also judged based on their friends and family, and some people changed associations in order to pass for another race. The passbooks were especially discriminatory because only black people were required to carry them, and if they lost even a few pages they could be put in jail. The passbooks contained everything in a person’s life: jobs, residence, marriage, and more, and basically kept track of people as if they were a threat to society.

While some people might say these racial technologies were only important in the South African context, so they are contextually political, I believe these technologies are inherently political, because any society that is so concerned with “scientifically” testing people’s race and keeping track of them is going to be a deeply racist society based on domination. First, the passbooks and tests required a strong hierarchy to enforce them. The passbooks were deeply resented, and the racial tests often contradicted people’s own lived experiences of their identity. Had the government been any less powerful, no one would have submitted to the evaluations and surveillance necessary to collect the data that was needed for them to “function.” Furthermore, like nuclear power, these tests and passbooks change the society they exist in, making it a society based around racial control. The desire to test and classify races is always going to be “strongly compatible” with separating those races or making hierarchies between them. Ultimately, the tests and passbooks were a necessary part of the infrastructure of apartheid, and would inherently push any society that used them toward apartheid-style thinking.

Based on his arguments in "Do Artifacts Have Politics?", how would Langdon Winner analyze the politics of automated risk assessment systems?

In "Do Artifacts Have Politics?" Langdon Winner argues that technologies are not neutral tools whose politics depend only on how people use them. Their design and arrangement can embody and enforce relations of power. By "politics" he means "arrangements of power and authority in human associations," which reaches well beyond elections and government. He rejects technological determinism, the idea that technology develops on its own logic and molds society. He also rejects pure social determinism, the idea that only the surrounding social system matters and the artifact itself is irrelevant. Winner takes a middle position: the technical details of an artifact are politically meaningful in their own right. He describes two ways technology can be political. First, a technology can settle a social issue through a design choice that could have been otherwise. Robert Moses's low overpasses kept buses, and so poor and Black riders, away from Jones Beach, and the tomato harvester favored large farms over small growers. These effects can be unintended, as with buildings inaccessible to disabled people through neglect. Second, some technologies are "inherently political." They either require a particular social order, as the atom bomb requires a rigid, centralized chain of command, or they are strongly compatible with one, as solar power is compatible with decentralized control.

Automated risk assessment systems take data about a person and produce a score predicting risk, such as reoffending, missing a court date, or greatest need in a homeless services system. The scores then guide decisions about bail, supervision, or housing priority. Winner would first ask what social issue the system settles. The answer is who is trusted, who is watched, and who gets scarce resources. The system answers through design choices: which variables count as "risk," what historical data trains the model, and where the cutoff falls. None of these was inevitable. Like Moses's bridges, the choices become part of the infrastructure and appear neutral, and they can disadvantage groups without anyone intending it. If past policing was uneven, the score reproduces that pattern, much like the neglect behind inaccessible buildings. Like the tomato harvester, the system's purposes go beyond its stated goal of efficiency, since it shifts discretion from judges and caseworkers to whoever builds and configures the tool. It also becomes an obligatory passage point, because people must pass through the score to get what they need.

On the second question, I think Winner would see these systems as strongly compatible with a particular social order, though they don't strictly require one. Risk scoring depends on centralized collection of records about people, standardized categories, and authority to act on the resulting ranks. That fits a society that governs people as risks to be managed from above, and it displaces the local, contextual judgment a more democratic arrangement would use. It does not strictly require that order, since one could imagine similar tools under community oversight, as solar power could be run democratically. Winner would therefore warn that debates about whether the algorithm is "biased" or how judges use it miss the point. The artifact's design already distributes power, so the public should scrutinize the arrangement itself.

How is Bowker and Star's account of South African apartheid relevant to today's datafied world? Use one HCE tool and one case study.

In "Race Classification and Reclassification under Apartheid," Bowker and Star argue that classification systems are not neutral bookkeeping but sites of political struggle hidden beneath a bureaucratic surface. As they put it, each category "valorizes some point of view and silences another." In apartheid South Africa, the Population Registration Act sorted everyone into four racial categories (European, Asiatic, Coloured, Bantu), which determined where a person could live, work, study, and marry. Black South Africans had to carry pass books recording their whole lives. Reclassification Boards judged disputed cases using "appearance and general acceptance and repute," through tests like the pencil test and questions about relatives and lifestyle. Bowker and Star explain the chaos with two concepts. The law wanted Aristotelian categories with clean, rigid boundaries, but enforcement could only rely on prototypical judgments of resemblance and social perception. The gap produces torque, the twisting of a person's life to fit a category that doesn't fit them. Sandra Laing was born to white parents but classified Coloured and expelled from her school, and Vic Wilkinson was reclassified five times as his life changed.

The HCE tool of classification shows why this matters today. Classification is the sorting of people and knowledge into categories governed by identifiable principles, and classification systems shape social order and vice versa. As our lectures added, classifications configure how a person appears ("counting people as..."), and they embed human judgments into infrastructures that are repeated and coordinated until they disappear into everyday life. Bowker and Star show that this work is invisible until someone doesn't fit, which gives us a method for denaturalizing data. My case study is the Coordinated Entry System and its VI-SPDAT score, which local homeless service systems use to rank unhoused people's "vulnerability" and prioritize scarce housing. The parallel is close. Like apartheid's categories, the score reduces a life to a rank that determines access to something material. It presents itself as an objective assessment, though the answers depend on what a person discloses and how a surveyor records it. People whose situations don't fit the survey's questions experience torque: they may be ranked low despite serious need, and the consequences are as real as a reclassification hearing. Critics have argued that such scores can produce racial disparities.

There are differences, which show what carries over. Today's classification is quieter. There is no pass book or board, only a form and a number, so it looks even more natural and is harder to contest. Data also circulates and endures, so a score travels away from the context that produced it. The lesson from Bowker and Star is to ask of any datafied category who made it, what it silences, who bears the torque, and how someone can contest it.

How does James Scott's reading relate to the relationship between states and populations? Use at least one HCE tool.

In Seeing Like a State, James Scott argues that modern states govern by making society legible: readable, measurable, and manageable from a distance by a central authority. Legibility means replacing local, particular, messy knowledge with simplified categories such as maps, grids, and standard units. His central example is scientific forestry in 18th-century Prussia and Saxony. Officials could not see a forest, only "commercial wood," an abstraction of revenue per acre. Fodder, medicine, and habitat were bracketed out. Foresters then reshaped real forests into single-species, same-age rows to fit their tables and their model tree, the Normalbaum. This worked at first, but over time Waldsterben, forest death from depleted soil and pests, showed that the ignored complexity had mattered. Scott pairs this with the metric system, a uniform standard imposed after the French Revolution. It helped commerce and taxation and carried an emancipatory promise of equal measures for equal citizens, but it replaced local practice with state-defined units. Both cases show simplification used as a tool of control, and both reshaped reality to match the standard.

The HCE tool that best connects Scott to states and populations is performativity: descriptions, classifications, and measurements can shape and even create the phenomena they describe. This is Scott's key move. The state does not simply observe a population that already exists. It creates the categories through which people become visible and governable, then reshapes life to fit them. Our lecture on populations and states develops this. Statistics, the "science of the state," and populations were made together through the state's interest in governing territory and increasing its power. John Graunt's tables turned individual deaths in London into aggregate patterns. The 1790 U.S. census sorted people into free white males, free white females, other free persons, and slaves, building race, gender, and freedom into how the population was seen and treated. Quetelet's "Average Man" created the idea of a norm, with deviation as "abnormal." Together these form the administrative grid, and Foucault's biopower names the result, in which managing human life becomes a resource of political power. Power is also relevant, since the abstractions reflect the assumptions and interests of those who hold it.

So the relationship Scott reveals is two-way. Statistics make populations visible to states, and states make populations through statistics. The view from above gains control but loses local knowledge, and the forest shows the cost: when the simplified grid stands in for reality, what was left out can return as failure and harm. Scott's metric system also keeps the story from being purely negative, since standardization delivered real benefits like uniform units and a promise of equal citizenship. His point is not that states should stop simplifying, but that simplification is a choice with winners, losers, and hidden costs.

What is socially robust knowledge? Use the sheep example and the air quality case.

Socially robust knowledge is the idea that scientific knowledge aims less at absolute certainty and more at knowledge that holds up both inside the scientific community and in society, while accepting that uncertainty always persists. It is not about the logical or epistemic strength of scientific claims, but works in tandem with it. This follows from the course's account of science as human and social work. Kuhn showed that data alone do not settle which theory to accept, and that communities decide through persuasion and shared practice. Robust knowledge therefore has an internal dimension, how the scientific community manages disputes and shares information. It also has an external dimension, how the community manages its relationship to the public. Lecture 7 adds three things. Expertise is "judgment in the face of uncertainty" (Jasanoff), so errors are inevitable over time and can erode trust. Scientists have a responsibility to consider how they communicate results. And openness to public participation that respects local knowledge can increase trust and improve the science itself.

The Cumbrian sheep case (Collins and Pinch) shows the external dimension failing. After Chernobyl, government scientists confidently predicted the radioactive caesium in sheep would disappear within weeks. It kept rising for years, because the scientists assumed lowland clay soil, which binds caesium, while Cumbria's peaty upland soil lets grass keep re-absorbing it. Trust collapsed for three reasons. Officials projected false certainty instead of admitting uncertainty, and farmers minded that more than the errors themselves. Scientists dismissed the farmers' knowledge of their land and flocks, and a fenced experiment failed for exactly the reason farmers predicted. And farmers watched the ordinary messiness of science up close, which contradicted the public image of certainty. It is a case of Collins and Evans's problem of extension: whose expertise counts? Collins and Pinch conclude that science is a fallible "golem," valuable but not infallible, and should be combined with local expertise.

The Bay Area Air Quality Management District case from Lecture 3 shows the same issue from another direction. The AQI is a made representation. It converts measurements of six major pollutants into a 0-500 scale with six color-coded health ranges, so the public can act on it. There are signs of internal robustness, as when BAAQMD warns that some data is raw and may contain errors, and posted an advisory that its PM2.5 AQI values were inaccurate after the EPA revised its standard. There are signs of external openness too: it publishes flare emission reports and causal reports for refineries like Chevron Richmond, and schools like Berkeley's use the data to set activity rules. But every representation is selective. Someone decided what counts as a pollutant, what counts as a monitoring site, and what counts as a flare, and readings are grouped into regional zones. Each neighborhood or block may have its own "view from below." Berkeley's schools supplement BAAQMD with AirNow and PurpleAir for that reason. The lecture's point is that grassroots groups near refineries may need to "make data of their own" to be represented at all. That is the Cumbria lesson again: robust knowledge means being open about uncertainty and treating local knowledge as a real input, not just publishing official numbers.

---

1. Artifacts & Politics

Tools: Power, Sociotechnical systems, Identity/Positionality, Narratives

• Technologies/artifacts have politics = their material design/arrangement enacts and enforces power relations. It's not just about good/bad use.
• Case: Robert Moses's overpasses. ~200 parkway overpasses with low clearance. 12-ft buses couldn't pass. Poor and Black residents (bus riders) were physically barred from Jones Beach; wealthier white car-owners weren't.
• How tech creates social order (5 steps): (1) integrates into everyday life → (2) becomes landscape (unnoticed) → (3) embodies values/positions/narratives → (4) engineers relationships between people → (5) becomes an obligatory passage point.

Reading: Winner, "Do Artifacts Have Politics?" (1980)

• Politics (Winner's def.): "arrangements of power and authority in human associations, and the activities within them." Broader than government/elections.
• Technological determinism: tech develops by its own internal logic and then unilaterally shapes society. Winner: naive.
• Social determinism of technology: only the social/economic system matters; the artifact itself is irrelevant. Winner: goes too far. Design details matter.
• Winner's middle position: take technical details seriously and social context.
• Two ways tech is political:
  1. Settling a social issue (contingent design choice): could have been built otherwise, but the chosen design has political effects.
     • Moses's bridges; tomato harvester (favored big farms, ~32,000 jobs lost); McCormick's machines (used to break a union, not for efficiency); inaccessible buildings (unintentional bias from neglect).
     • Key: a tech's purposes go beyond its immediate function.
  2. Inherently political technologies: require or are strongly compatible with a particular social order.
     • Strong (requires): atom bomb needs a rigid, centralized, authoritarian chain of command; ships need one captain.
     • Weak (compatible with): solar power is compatible with decentralized/democratic control but doesn't require it.
  • Winner: both types exist, so judge case by case. Societies accept authoritarianism inside technical systems while resisting it elsewhere.

2. Making the Datafied World

Tools: Power, Sociotechnical systems, Identity/Positionality, Narratives

• Datafied world: everyday life rendered as data and acted on by algorithms.
• Layers (bottom → top): ubiquitous data → reliable analytics → culture of algorithms → autonomous agents.
• Everydayness: datafication is embedded in routine life, so it goes unnoticed (parallels step 2 of L1).
• World-making: tech-society interaction simultaneously shapes power, social order, and meaning.
• Technological determinism & sociotechnical narratives: stories like "tech just advances/AI is inevitable" hide human choices.
• Datafication is selective: it quantifies some things and leaves out others (often qualitative/human). That selection is a political choice.
• Case: RealPage (YieldStar). Software recommends rents using pooled landlord data. DOJ alleges algorithmic price-fixing (illegal if done by humans talking). Sociotechnical system: landlords, tenants, regulators, journalists. Cities (e.g., SF) moving to ban it.

3. Making Data

Tools: Power, Representation, Classification

• Ask of any data: How was it made? Who made it? How? Why? In what context? So what?
• Naturalize / denaturalize: choices in representations and classifications get taken for granted as "just how things are." Denaturalizing = revealing them as choices.
• Representation is 2-sided: it produces knowledge and politics/power.
• Counting vs. "counting": what can be counted "counts" (gets attention/resources); what can't gets ignored.
• Data travels: abstracted from its original context and moved elsewhere (including into ML/AI).
• Key terms: selection (what is measured), abstraction (dropping context/detail), mediation (instruments/methods stand between us and the phenomenon), perspective (data is made from somewhere), view from above (institutional, aggregate, distant) vs. view from below (situated, local, lived).
• Case: AQI and BAAQMD ⚠️
  • AQI: turns pollutant concentrations into one simple, color-coded index (using the EPA scale), so complex measurements become legible to the public. Selection (which pollutants, which thresholds/standards), abstraction (many readings → one number/color), mediation (stations, sensors, formulas).
  • Reading: BAAQMD Public Data Center. Organized into four groups:
    • "What you breathe": current air quality, five-day forecast map (five Bay Area zones, EPA AQI scale), open burn map, annual air quality summaries.
    • "Who we regulate": permit applications, public notices, Title V permits, facility maps, current rules.
    • "What we measure": emissions inventories, standards and attainment status (11 pollutants), GHG data, flare data (refineries), CARE data (areas where pollution most affects health), monitoring maps (incl. monitors near refineries).
    • "How to get involved": complaint forms, notifications, board meetings, budget, public records requests.
  • Presentation: mostly maps, tables, forecast zones, and color scales, all aggregated views from above.
  • How to think about using it: it's designed to inform residents and show regulatory accountability, yet it privileges what instruments count (regulated pollutants, station locations, regional zones). Local/lived experience appears only through channels like complaints. CARE and facility maps let communities connect pollution to place and inequality, but they only cover what's measured. Ask: what is not counted or visible, and who decides thresholds ("attainment")?

4. Making Personal Data

Tools: Vulnerability, Representation, Classification, Power, Identity/Positionality, Narratives

• Key properties of data: representation, classification, circulation, abstraction. They matter more when people are in the data.
• Database: sorts, ranks, and classifies people. Abstraction makes you look away from the person behind the data.
• Circulation: data travels in ways that serve others' needs (not the data subject's) and is durable, so it can be reused against people later.
• Personal data beyond "individual identifiability/privacy":
  • Relationality: data about you is also about family, community, groups (collective, not just individual).
  • Representational harm: being misrepresented/miscategorized is itself a harm.
  • Vulnerability: personal data is entangled with human vulnerability.
• Also (from notes): Berkeley punch cards (1960s; "Do not fold, spindle, or mutilate"), students reduced to categories on a card. Reflexivity: measuring changes behavior.
• Ask: How am I looking for/listening to the data subjects?
• Cases ⚠️:
  • "'Cause we got a nice little database": data assembled and used in ways the subjects didn't foresee or control.
  • Coordinated Entry System + VI-SPDAT: scores unhoused people's "vulnerability" to prioritize housing. Classification/ranking of people. Abstraction reduces life to a score; vulnerability is turned into data with real consequences.
  • DTC DNA testing: data about one person is inherently about relatives (relationality); circulation to companies/law enforcement.

5. Making People out of Data

Tools: Performativity, Power, Identity/Positionality

• The self is made, not just found, co-produced with tech and society (contingent and variable).
• Performativity: classifications, systems, norms, and practices produce ways of being a person (see toolkit for what it is NOT). Marriage "I do" analogy: the words create the marriage.
• Personality as historical product: pre-1900s people were understood by character (moral). From ~1917, personality = measurable traits, tied to psychology as a discipline, testing, and industrial workforce management.
• Self-quantification: "know thyself" needs external tools (writing, counting, mirrors). Examples: Franklin's 13 virtues chart; Santorio's weighing; Fitbit/Apple Watch/Quantified Self; looksmaxxing. What's unique about self-making through numbers?
• Cyborg: self as porous and relational, built from personal, social, biological, and technical parts.
• Cases: "Multiple Personality Disorder" (see Hacking); personality tests (Myers-Briggs/quizzes).

Reading: Bowker & Star, "Race Classification and Reclassification under Apartheid"

• Classification isn't neutral; it's a site of political struggle hidden under bureaucratic surface. Its work is often invisible, until it breaks down (e.g., someone doesn't fit, or a case is contested), which makes its artificiality visible again. That's a route to denaturalizing data.
• Torque: the twisting of a person's biography when a rigid category doesn't fit their lived life.
• Aristotelian categories (rigid, necessary-and-sufficient boundaries; what the law wanted) vs. prototypical categories (fuzzy, best-example, everyday judgment; what enforcement actually relied on). Apartheid's contradiction = demanding the first, enforcing with the second.
• Framework: Population Registration Act + Group Areas Act (4 races: European, Asiatic, Coloured, Bantu/Native), which determined where you lived, worked, studied, married.
• Pass books: 50+ page ID for Black South Africans; total surveillance.
• Reclassification Boards: criteria were "appearance and general acceptance and repute," not descent/one-drop rule.
• Technologies of classifying: pencil test, earlobes/cheekbones, skin-color scales, questioning relatives, lifestyle checks (porridge, bed vs. floor).
• Cases: Vic Wilkinson (reclassified 5 times); Sandra Laing (white parents, expelled from white school, classified coloured); David Wong (white neighbors' affidavits calling him "White" → government crackdown); "Dottie"/Griqua (language/appearance vs. assigned race); Ronnie van der Walt (boxer reclassified before a match; punitive example).

Reading: Hacking, "Making Up People" (2006)

• Claim: human sciences don't just discover kinds of people. They create new ways of being a person.
• Looping effect: classified people change in response to the label (and community forms around it), so the classification must change in turn.
• Statement A ("there were no X in 1955, many in 1985") = existence claim, often unproductive. Statement B ("in 1955 that wasn't a way to be a person; in 1985 it was") = claim about social roles/experience.
  • MPD/DID: A arguably true, B true. High-functioning autism: A false, B true.
• Ten engines: Count, Quantify, Create norms, Correlate, Medicalise, Biologise, Geneticise, Normalise, Bureaucratise, Resist (reclaim identity).
• Natural kinds: real kind (horse; many bundled properties) vs. finite kind (white things; one shared property). Autism ≈ real kind; obesity ≈ finite kind.
• Link: Hacking (performativity of classification) + Bowker & Star (classification's material force) = classification acts on people.

6. Making Robust Knowledge

Tools: Narratives, Power

• Science as human and social activity: real people, judged by a community, not "stick-figure scientists."
• Three philosophies:
  1. Logical positivism: theory = data organized by logic; verify by induction. Problem: black swan (can't prove "all swans are white").
  2. Falsificationism (Popper): good theories are falsifiable and not yet falsified. Problem: real scientists don't drop theories at one contrary result (Planck: science advances funeral by funeral).
  3. Underdetermination (Duhem-Quine): hypotheses are tested with a web of background assumptions, so failure doesn't tell you what to blame; data alone can't force a choice between theories.
• Theory-dependence (theory-ladenness) of observation: what counts as data already reflects theory.
• Conclusion: no simple relationship between data and theory; scientific views can change radically; look at social, intersubjective work (content and context).

Reading: Kuhn, Structure of Scientific Revolutions

• Paradigm = the whole package a community shares: landmark exemplar/achievement (e.g., Newton's Principia), theories/laws/concepts, instruments/techniques, standards for legitimate questions and solutions. Both a worldview and a form of practice.
• Normal science: puzzle-solving within a paradigm; failure blamed on the scientist, not the paradigm; students trained via exemplars.
• Revolution: anomalies accumulate → crisis → new paradigm → community converts.
• Incommensurability: rival paradigms can't be compared directly because they disagree on (a) which problems matter and what counts as solved, (b) meanings of key terms ("mass," "earth"), (c) what's seen as significant ("live in different worlds"). Choice is by persuasion, demonstrated success, aesthetics (elegance), and generational turnover, not pure logic.
• Case: Copernican Revolution ⚠️: shift from earth-centered to sun-centered astronomy as an example of paradigm change; old data didn't decide it, since data alone underdetermined the choice (Ptolemaic vs. Copernican models both fit observations).

7. Science in Society

Tools: Power

• Expertise: skill/knowledge in a domain, gatekept by education and credentials; a source of social power. Jasanoff: expertise = "judgment in the face of uncertainty" exercised in the public interest. Errors are inevitable and erode trust.
• Socially robust knowledge (core concept): science aims less at absolute certainty and more at knowledge that holds up both inside the scientific community and in society, and acknowledges the inevitable persistence of uncertainty. Two layers:
  • Internal: how the community manages disagreement and shares information.
  • External: how it manages its relationship to the public, including whether it respects local knowledge/expertise and needs.
• Implications: scientists must consider how they communicate results (esp. recommendations); openness to public participation raises trust and can improve the science.
• Science as politics by other means: science helps make social order; its authority rests partly on rhetoric/persuasion (echoes Kuhn).
• Scientization of politics / politicization of science: political disputes get recast as technical ones, then each side recruits experts.
• Problem of Legitimation = what justifies giving expertise special authority in political decisions? Problem of Extension = how far does it reach, and whose input (lay/affected communities) counts?
• Autonomy vs. transparency as bases for scientific authority.
• Post-1945 moment: elevated status of science, government funding/direction (WWII, Cold War), new social contract for science (society funds and trusts; science delivers).
• Cases:
  • Climate science ⚠️: expert authority, uncertainty communication, politicization.
  • Aquila trial (2009): experts publicly reassured residents shortly before a quake killed 300+; independent researcher Giuliani had predicted activity; scientists prosecuted (convicted, largely acquitted on appeal) for how they communicated risk.

Reading: Collins & Pinch, "The Science of the Lambs" (The Golem at Large)

• Story: After Chernobyl (1986), fallout hit Cumbrian sheep farms. Government (MAFF) scientists confidently said risk would vanish in days/weeks. Instead caesium in sheep kept rising for years; bans lasted indefinitely and devastated farmers dependent on timed lamb sales.
• What went wrong scientifically: scientists assumed lowland clay soil (caesium locks up). Cumbria's upland peaty soil keeps caesium mobile, so grass re-absorbs it and sheep are re-contaminated. Relevant data existed since the 1960s but was read as a direct exposure question, not a chemical uptake pathway (theory-dependence of observation).
• Why trust broke down:
  1. False certainty, not honest uncertainty: confident predictions reversed; farmers minded the swagger and refusal to admit error, not the mistakes.
  2. Dismissal of local expertise: farmers knew their land's water/radiation pooling, lamb-timing, sheep behavior (won't eat imported straw; fencing makes them waste). Scientists ignored them; a fenced mineral experiment failed for exactly the reason farmers predicted.
  3. Farmers witnessed science's messiness: calibration troubles, inconsistent readings, re-measuring, contradicting the public image of certainty.
• Golem metaphor: science is powerful but clumsy, human-made, fallible; dangerous when we expect certainty.
• Overcorrection: farmers swung from "science = certainty" to "science = political conspiracy." Both wrong. Healthy view: fallible but valuable; value local expertise alongside institutional science.
• Kuhn connection: a clash of ways of seeing/standards of evidence, and a case where incommensurable-like gaps between expert and lay frameworks blocked listening.

8. Populations and States

Tools: Performativity, Power, Classification

• States "see" via simplifying abstractions: the administrative grid (census categories, surveyed grids). Abstractions reflect the assumptions/interests/norms of those in power and are performative (bring into being what they describe).
• Statistics lets society see itself: view of the sovereign/view from above; abstract numbers, tables, graphs; social regularities constitute "society"; interest in norming populations and assessing difference.
• Terms: State; population (created as a concept by state counting); biopower (Foucault; managing human life as a resource of political power); Statistik (German, "science of the state"); political arithmetic(k) (English; Petty, Graunt); normal curve.
• Cases:
  • John Graunt's Table of Casualties (17th-c. London): individual deaths → aggregate patterns.
  • U.S. Census 1790: free white males (16+/<16), free white females, other free persons, slaves. Race/gender/freedom status built in from day one.
  • Quetelet's "Average Man" (1830s): applied bell curve to human traits/crime; created "normality" with deviation = abnormal/pathological.

9. Eugenics and Statistics

Tools: Power, Classification, Identity/Positionality

• Claim: modern statistics (regression, correlation, significance testing) was co-produced with eugenics, an elite/state-driven project to norm and manage human heredity in populations for "social improvement," almost always tied to white supremacist, elitist, ableist visions of a "desirable" order. Part of the longer history of state-population-statistics (biopower). Eugenic ideas linger in powerful institutions today.
• Key figures:
  • Galton: regression/correlation; regression to mediocrity (offspring drift toward the average); popularized bell curve for human traits.
  • Pearson: chi-squared test and template of significance testing, developed to argue races/skull types are distinct and of different worth.
  • Fisher: Statistical Methods for Research Workers (1925) made significance testing standard; pushed sterilization of the "feeble-minded."
  • Lewis Terman (Stanford): IQ testing ⚠️: adapted intelligence tests for large-scale sorting, tied to eugenics.
• Significance testing critique: asks only "is there an effect?" not "how big / why / does it matter?" That "sizeless" yes/no fit eugenics' needs.

10. Taking the Global View

Tools: Narratives, Power, Identity/Positionality, Representation, Performativity

• "The global" is modern (post-WWI/WWII); built through the administrative state and its successors: UN, IMF, World Bank. Global datasets carry the view from above. Ask: who can make global data, how, why, what's left out?
• What made "the global" emerge: Cold War (First/Second/Third World), decolonization (Global South), economic globalization (flows of goods/capital/labor, "development"), new international institutions (mostly created by/for the Global North).
• Other users of these tools: insurers, pollsters, marketers.
• Key terms: development, fertility, demography, "family planning."
• Case: Gapminder.org (global fertility data visualization): whose narrative does it tell? Contrast with narratives centered on people on the ground.
• Population-control narrative: "population explosion" fear framed poor women (esp. women of color in the Global South) as the problem, fertility management as the solution, ignoring capital flows, wages, migration, environment. Reproduces colonial relations (women as experimental subjects/resources). Language shifted from "population control" (1950s-60s) to "women's empowerment/family planning as choice" (1990s-2000s).
• Core tension: simplifying data lets institutions function but can be unfair, incomplete, and agency-denying to those modeled.

Reading: James Scott, Seeing Like a State

• Scientific forestry (18th-c. Prussia/Saxony): officials couldn't see a forest, only commercial wood (revenue yield per acre); everything else (fodder, medicine, habitat, local uses) was bracketed out. Then they re-engineered forests into single-species, same-age rows to fit tables and the Normalbaum (model tree). Simplification became control. Short-term success; long-term Waldsterben (forest death: soil depletion, pests, storms) because complexity was ignored.
• Legibility: making something readable/measurable/manageable from a distance by a central authority via simplified categories (maps, grids, standard units). Legible to the state/officials, for taxation, control, administration. Local, particular knowledge stays illegible to outsiders. Legibility becomes performative when reality is reshaped to fit the grid.
• Metric system (post-French Revolution): parallel case: standardizing tech imposed from above. Helped commerce and taxation, tied to Enlightenment/emancipatory ideal of equal citizenship ("same law, same measures"), took decades to replace local practice.
• Point of the comparison: both impose a single uniform standard on messy local reality to make it legible/controllable, and both reshape reality to match the standard.
• Ties: administrative grid (L8), biopower, performativity, local knowledge (Cumbrian farmers, L7), view from above (L3, L10).

Reading: Clayton, "How Eugenics Shaped Statistics"

• Galton, Pearson, Fisher were eugenicists; eugenics was the motive behind key statistical inventions (see L9).
• Central argument: significance testing's sizeless yes/no question served eugenic aims (prove groups differ, no need for magnitude/cause).
• Takeaway: statistics is a human practice shaped by its makers' interests, not a view from nowhere. Since we won't stop using it: be suspicious of claims to pure objectivity, focus on magnitude, mechanism, context, push for more inclusive/self-aware statistics. "Objectivity" can be a rhetorical strategy.