Appendix F — Workshop Outline

This workshop introduces ideas that will help people with backgrounds in tech think clearly about the harms of social media and AI, and about how they should be regulated. The material can be used as a two-day workshop with exercises and breaks. I would be very grateful for feedback; in particular, my perspective is often parochial, and many of my references are probably out of date.

F.1 At a glance

  • Length: two days (roughly twelve to fourteen hours of teaching, plus breaks and lunch).
  • Audience: people with programming backgrounds, but little previous exposure to research in psychology, sociology, or politics.
  • Format: 7 lessons of roughly 60-90 minutes, each followed by one or two 10-15 minute exercises taken from those listed with the lesson.

F.2 Stance

  1. Power in society is distributed very unequally, and much of what looks “neutral” or “technical” is actually a rule that favors some people over others.
  2. Inequality is substantially the product of policy choices and inherited advantage, not the natural sorting of individual talent.
  3. Markets are human institutions, not laws of nature. There is no such thing as a market without rules; the only question is whose interests the rules serve.
  4. What counts as “harm”, and therefore what gets regulated, is also the result of decisions made by people. Alcohol, cannabis, sex, and guns are not regulated in proportion to the damage they do; they are regulated according to whose interests and whose anxieties dominate.
  5. Regulation of food, medicine, air, water, cars, and tobacco has repeatedly protected the public. Failures of regulation usually come from regulatory capture and weak enforcement, not from the impossibility of governing markets.
  6. Change is possible. It is usually won by organized collective action rather than by persuasion alone, and technologists have unusual leverage in the fights now under way.

F.3 Learning outcomes

By the end of the workshop, participants should be able to:

  • Distinguish structural explanations from individual ones, and explain why individual explanations are the default in tech culture.
  • Identify who benefits and who bears the cost of specific arrangements, especially when the benefits are concentrated and the costs are diffuse.
  • Recognize market failures (externalities, public goods, information asymmetry, market power) and name the regulatory instruments available to correct them.
  • Describe how cognitive biases, identity, and group belonging shape belief, and how design can amplify or dampen them.
  • Explain how a society decides what counts as harm, why regulation is often not proportional to harm (e.g., alcohol versus cannabis), and the difference between rare-dramatic-attributable and diffuse-pollution models of harm.
  • Give examples of regulation that worked, examples regulation that failed, and the mechanisms (especially regulatory capture) that explain the difference.
  • Explain how media and platforms frame issues, and shape the range of what is treated as debatable.
  • Connect specific harms of social media and AI to each lesson, and sketch realistic strategies for responding to them.

F.4 How to run this

Pick one or two exercises per lesson.
Four or more are provided so you can match the room, the time available, and the participants’ interests. Do not try to run them all.
Exercises are the assessment.
Exercises are where participants convert a claim into something they can use. Do not skip the debriefs: they are where the learning is consolidated.
Use groupwork and think-pair-share.
One or two minutes alone, a few minutes in pairs or larger groups, then report back. This gets more voices into the room than any other structure in the time available.
Keep it concrete.
Every abstract concept in this plan has a named example. When discussion drifts into abstraction, ask, “What does that look like in a product or a law?”
Do not manufacture consensus.
Like-minded people can disagree about ends and means. “What would change your mind?” is usually a better question than “who’s right?”
Separate the descriptive from the normative.
“This is how X works” and “this is what we should do about X” are different claims with different burdens of proof. Keep them distinct.

F.5 Materials

  • A whiteboard or shared document for each group.
  • Sticky notes.
  • Breakout spaces.

F.6 1) Power and institutions

Why do some people’s preferences become policy and other people’s don’t?

F.6.1 Learning objectives

  • Describe the three faces of power (decision, agenda, preference-shaping).
  • Define an institution as “the rules of the game” (formal and informal).
  • Explain why concentrated interests usually defeat diffuse ones.

F.6.2 Key concepts

  • Three faces of power (Lukes 2021):
    1. Who wins when there is a visible decision
    2. Who controls what gets onto the agenda in the first place
    3. Who shapes what people even want, so that some options never occur to anyone
  • Institutions (North 1990):
    • The formal rules (laws, contracts) and informal constraints (norms, conventions, taboos) that structure interaction
    • Institutions always allocate advantage; they are never neutral
  • Collective action problem (Olson 1965):
    • It’s easier to organize a small group with a large stake than a large group with a small stake
    • This is why producers beat consumers and platforms beat users
  • Path dependence:
    • Early choices lock in later outcomes, even after the original reasons disappear
  • Selectorate theory: leaders distribute resources to the small “winning coalition” that keeps them in power, not to the whole population (or the whole “family”).
  • Legibility (Scott 1998): surnames, house numbers, passports, and ID registers exist so the state can tax, draft, and police; standardization is an instrument of administration, not a neutral convenience.

F.6.3 Content

  1. Opening question (3 min): “Think of a product decision at your company. Who actually decided it, and whose interests won?”
  2. What is power? (12 min): Walk through Lukes’ three faces with concrete examples. Visible: who wins the vote. Hidden: who writes the agenda, the roadmap, the API. Invisible: who shapes what feels “natural”, “obvious”, or “inevitable”. Point out that a “neutral” default or a “technical” constraint is usually an act of agenda-setting. “We don’t do politics here” is the third face’s catchphrase: it usually means the winners already have what they want and can describe their arrangement as natural.
  3. Institutions (15 min): North’s definition. Formal vs. informal rules. Show that software itself is an institution: default settings, permission models, and rate limits are rules that allocate advantage. A design decision is a governance decision. Make the point concrete:
  • The 1886 US railroad gauge change: an “incompatibility” was a market weapon—southern carriers profited from unloading and reloading freight at the gauge boundary, so the “technical” choice was political.
  • Tithing and excommunication: the App Store’s 30% cut is a tithe; deplatforming is excommunication (removal from the infrastructure needed to do business).
  • Incentives produce the wrongdoing: Wells Fargo fired 5,300 staff over fake accounts (“eight is great”)—mass wrongdoing flows from impossible quotas, not from bad individuals.
  • Self-fulfilling design: terms of service written for adversaries and moderation that treats everyone as a suspect produce the distrust they assume (contrast with Ostrom’s governed commons).
  1. Platforms as quasi-governments (10 min): eBay adjudicates disputes, App Store review is the only appeal, and Facebook makes de facto regulatory decisions in other countries. The yakuza supplied 2011 Tohoku disaster relief faster than the state—a reminder that “governance” is whoever can actually get things done.
  2. Legibility and institutional money (10 min): James C. Scott’s point that surnames, house numbers, and passports exist so the state can tax, draft, and police; South Africa’s pass laws became the world’s first biometric register (1986). Philanthrocapitalism is the private-sector twin: donor-advised funds (~$230B) and the Giving Pledge convert taxable wealth into permanently controlled endowments, and the Gates Foundation was at one point the WHO’s second-largest funder.
  3. Collective action (10 min): Olson’s concentrated-vs-diffuse logic. Example: ad-tech firms (few, organized, rich) versus individual users (many, unorganized, each with a small stake). This single asymmetry predicts much of tech policy. Selectorate theory sharpens the point: leaders distribute to a small winning coalition, not the whole “family”, and Pfeffer found political skill predicts advancement better than technical competence.
  4. Recap (5 min).

F.6.4 Exercises

Map the decision.
Pick a real decision, such as an API change, a layoff, a content-moderation rule. On a whiteboard, list (a) the visible decision-makers, (b) who set the agenda that made this the decision, and (c) whose preferences shaped what seemed “natural”. Which face of power is hardest to see, and why?
Three-faces audit.
Take a news story about a tech-policy fight. Find one example of each face of power, or explain which face is absent from the public account and why that absence matters.
Concentrated vs. diffuse.
List the stakeholders in a policy question (e.g., banning facial recognition in public spaces). Classify each as concentrated or diffuse, predict which will organize most effectively, and check your prediction against what actually happened.
Design the institution.
A team is choosing default privacy settings for a product. Write three different “rules of the game” (a default, a policy, a norm) that produce three different outcomes, and say who wins under each.
Path dependence hunt.
Find a feature, protocol, or company practice whose current form is explained more by history (“it’s always been done this way”) than by present need. Trace the lock-in and name who benefits from keeping it.
Quasi-government audit.
Pick one platform decision (a ban, a commission change, an API restriction) and compare it to a government decision. Who is the constituency? Where is the appeal? Which face of power is doing the work?
Incentive autopsy.
Take a case like Wells Fargo’s fake accounts. Trace the quota, the measurement, and the reward structure that produced the wrongdoing. Then redesign the institution so the bad behavior is no longer the rational response.

F.6.5 Takeaway

Power is not only about who wins visible fights; it is about who sets the agenda and who shapes what people think they want. Institutions are the mechanism, and they are never neutral.

F.6.6 Further reading

F.6.6.1 Big Tech is Like a Company Town

George Pullman’s 1894 company town is the single best illustration of the workshop’s key claim that a design decision is a governance decision.

When Pullman cut wages during the depression of 1893 while holding rents fixed, workers could not cushion the blow by cutting other expenses, because the company controlled those too. They had no recourse and nothing left to lose. The 1894 Pullman Strike paralyzed rail traffic across the country and required federal troops to suppress, which tells you something about what total control eventually produces.

The payoff line to bring it into tech:

An employer controls your income; an infrastructure provider controls the conditions under which you can earn, spend, and participate in social life.

Other useful references are the danwei/WeChat parallel (“A user whose account is suspended loses not just a communication tool but the infrastructure through which they conduct their civic and commercial life”), and Foxconn’s company-town campuses.

F.6.6.2 We’re All Family Here

The Musk “hardcore” email illustrates the lesson’s question (“who actually decided it, and whose interests won?”):

In November 2022, after laying off about a third of its original workforce, Elon Musk sent an email to the remaining Twitter employees asking them to click a button to confirm that they were committed to working ‘hardcore’ for the company’s next phase. Those who did not click by the deadline would be treated as having resigned. A few months earlier, Twitter’s former leadership had described it as a family.

Lasswell’s definition is a clean, quotable anchor for the whole lesson:

The political scientist Harold Lasswell defined politics in 1936 as the study of ‘who gets what, when, how.’

F.6.6.3 We’re All Family Here

Extends the “concentrated vs. diffuse” point from Olson. The “winning coalition” idea explains why everyone else is “interchangeable” despite the family language. Also useful: Basecamp’s 2021 ban on “societal and political discussions” and the immediate resignation of a third of staff: “the decision to ban discussion of politics was itself a political decision, made unilaterally by the winning coalition.”

F.6.6.4 The Invention of the Corporation

The workshop says software is an institution. This shows the corporation is too, and that its “neutral” form is a choice with distributional consequences:

The usual excuse for why tech companies aren’t worker cooperatives or other equitable structures is that cooperative governance is too slow and cooperative financing too limited. This explanation is conveniently incomplete. It omits the fact that the founders, lawyers, and venture capitalists making the choice benefit most from the conventional structure.

F.6.6.5 Big Tech is Like the Enclosure Movement

The cleanest “institutions allocate advantage” case, and a bridge to Lesson 4:

The process was not neutral arbitration between competing claims: the landowners who stood to benefit were the same class that controlled Parliament. The commoners whose rights were extinguished had no equivalent political representation.

F.6.6.6 Privacy, Power, and the Self

Scott’s legibility argument is the through-line for the “passports, house numbers, and biometrics” examples: the state standardizes people so they can be taxed, drafted, and policed.

F.7 2) Where markets come from

What are markets, and when do they fail?

F.7.1 Learning objectives

  • Explain why “the free market” is a description of a rule-bound institution, not an absence of rules.
  • Name and give examples of four kinds of market failure.
  • Say who bears the cost and who captures the benefit in a negative externality.

F.7.2 Key concepts

  • Embeddedness (Polanyi 2001): markets are always embedded in a web of legal, social, and political rules. Property rights, enforceable contracts, currency, and liability all rest on the state. “Deregulation” is not the removal of rules; it is the substitution of one set of rules for another.
  • Externalities: costs or benefits imposed on third parties that the price does not capture. Pollution is the classic negative externality; attention extraction and misinformation are its digital descendants.
  • Public goods: non-excludable and non-rivalrous goods like trust, safety, and an informed electorate that markets under-produce because no one can be made to pay.
  • Information asymmetry (Akerlof 1978): when one party knows much more than the other, the “market for lemons” result is that bad options drive out good ones.
  • Market power: network effects and switching costs produce concentration, sometimes “natural monopoly,” and the ability to set terms rather than take them.
  • Market failure: a situation in which an unadjusted market produces an outcome that is neither efficient nor fair.
  • The commons (Ostrom 2015): Hardin described an unmanaged commons he never studied; Ostrom’s eight design principles show communities can govern shared resources without markets or the state.
  • Enshittification (Doctorow 2025): platforms first attract users, then extract value from users and from their suppliers.

F.7.3 Content

  1. Opening question (3 min): “Which product in this room is unregulated?” Let the room discover that the answer is “none.”
  2. Markets are constructed (15 min): Every marketplace has rules about who may sell, what counts as fraud, who is liable, and how disputes are resolved. Those rules are choices with distributional consequences. Show that “deregulation” is re-regulation. Fill in the history:
  • The barter myth (Graeber 2011): credit and debt predate money; money emerged to settle debts, not to simplify barter.
  • Limited liability is a grant from the state (Dutch East India Company, 1602), not a natural feature of commerce.
  • The Statute of Anne (1710) created copyright to break the Stationers’ monopoly, not to protect authors; the 1998 term extension was timed to keep Mickey Mouse out of the public domain.
  • “Homo economicus” is a political argument dressed as math: from Hobbes to Spencer’s “survival of the fittest” to the rational agent.
  1. Market failure (15 min): Four mechanisms, each with a tech example:
    1. Externalities (misinformation, polarization, teen anxiety as costs priced at zero)
    2. Public goods (an informed public, institutional trust)
    3. Information asymmetry (you do not know what the recommender is doing, or why)
    4. Market power (network effects make dominant social networks hard to exit)
  2. Add the commons as a fifth idea. Apple’s green-bubble lock-in is a market-power case study—internal documents admit RCS would reduce the social cost of switching, and the DMA forced support in 2024.
  3. Public subsidy, private profit (8 min): every technology in the iPhone was government-funded (DARPA internet, GPS, Siri/SRI); the mRNA vaccine produced Moderna’s ~$200B while the NIH’s IP claim went unenforced; Apple paid 0.005% tax on Irish profits.
  4. Enshittification (8 min): Doctorow’s ladder with concrete rungs—Booking.com’s commission from 12% to 25-30%, Columbia House’s penny records, JanSport’s backpacks, YouTube’s 2018 monetization threshold. Platforms first attract users, then extract from them and their suppliers.
  5. Institutional variety (8 min): beyond the shareholder firm—the Islamic waqf (roughly a third of Ottoman farmland), keiretsu, the Mittelstand, the Hindu Undivided Family (Tata/Birla). The firm we treat as “natural” is one choice among many.
  6. The MLM warning (5 min): the FTC found <1% of participants earned a net profit; Amazon Basics competes with its own sellers using their sales data; Uber pitched drivers as “entrepreneurs.”
  7. Who pays, who benefits (5 min): “Efficiency” is not the same as fairness. A market can be efficient and still concentrate benefits and diffuse harms. (Lesson 6 uses this distinction.)

F.7.4 Exercises

Externality inventory.
Pick one social-media harm (polarization, teen anxiety, election misinformation). Identify the negative externality, who bears the cost, who captures the benefit, and why the price does not reflect the cost.
Public-good puzzle.
Classify “an informed electorate,” “trust in institutions,” and “online safety” as public goods. Explain the free-rider problem and why no single company has an incentive to produce them. What follows for who should produce them?
Information asymmetry.
Choose a feature such as a recommender, a “black box” credit or hiring model, or an ad auction. Who knows more, the operator or the user? What could close the gap: disclosure, audit, a right to explanation, or something else?
Natural monopoly.
Argue whether a dominant social network is a natural monopoly due to network effects, high switching costs, and/or data advantages). If it is, what follows for regulation?
Free market autopsy.
Find a product described as “the free market at work.” List every government rule (and private rule) it depends on related to property, contract, fraud, liability, and standards. Discuss: is “deregulation” ever actually the absence of rules?
Subsidy autopsy.
Pick a consumer product and list every government-funded technology inside it (the internet, GPS, voice recognition, vaccines). Then discuss who captured the profit and who paid the research bill.
Commons design.
Take a shared resource (an open-source project, a Wikipedia article, a neighborhood) and draft the design principles that keep it from turning into Hardin’s tragedy. Compare your list with Ostrom’s eight.

F.7.5 Takeaway

There is no market without rules. The interesting questions are which rules exist, who wrote them, and who benefits. When markets fail, regulation is the standard correction, not an alien intrusion.

F.7.6 Further reading

F.7.6.1 Big Tech is Like a Drug Cartel

An opening hook for the workshop as a whole:

In order to show how the free market really works, he went and studied it in its pure, unconstrained form: the cocaine cartels.

The killer detail is brand protection without trademark law:

A cocaine cartel can’t use trademark law. It can, however, use violence against competitors who sell adulterated product under the same name or who operate in territory the cartel has claimed. The aim—maintaining exclusivity and quality signals—is the same, it’s just the mechanism that differs.

F.7.6.2 The Creation of Money

The strongest possible “markets rest on rules” illustration is the 1660s goldsmith story:

This is just as ridiculous as it sounds, but very useful. When you repay a loan, the deposit disappears from your account and the corresponding debt disappears from the bank’s books. Money is destroyed.

F.7.6.3 Big Tech is Like Professional Wrestling

The workshop discusses “market power” but has no example of fake competition:

What the kayfabe obscures is that Apple earns approximately $20 billion a year from a deal that makes Google the default search engine on every iPhone sold. The two rivals are financially interdependent, and each needs the other to play its designated role.

F.7.6.4 The Demand Problem

A market-failure mechanism not in the workshop’s four-part list:

Every firm therefore has a rational-as-in-psychopathic incentive to automate beyond the socially optimal level, because the gain from cutting labor costs outweighs the diffuse shared consequence of eliminating consumer spending.

Contrast this with Henry Ford’s opposite logic (“his employees needed to earn enough to buy his cars”) and Jack Dorsey/Block’s 2025 layoffs being rewarded with a +25% stock bump.

F.7.6.5 Fads and Bubbles

Tulip mania (“a single bulb of the Semper Augustus tulip sold in Amsterdam for the price of a house on a canal”) and Isaac Newton in the South Sea Bubble (“I can calculate the motions of heavenly bodies, but not the madness of men”). The “greater fool theory” line “A bubble collapses when no greater fool can be found” is a quotable takeaway.

F.7.6.6 Big Tech is Like the Sharecropping System

The chapter’s most quotable metaphor for platform lock-in and information asymmetry is:

The audience is the crop; the algorithm change is the landlord raising the rent after the harvest is in.

F.7.6.7 Control and Alternatives

The source for the “institutional variety” point: the shareholder firm is not natural or inevitable, and alternatives such as cooperatives and the Mittelstand are durable institutions, not seminar hypotheticals.

F.8 3) How people actually think

How do real humans make decisions, and how can that be exploited?

F.8.1 Learning objectives

  • Distinguish fast/automatic from slow/deliberate processing.
  • Name several biases and give an example of each in an online setting.
  • Explain how identity and group membership shape belief, and what that implies for “just show people the facts”.

F.8.2 Key concepts

  • Dual-process theory (Kahneman 2011): System 1 is fast, automatic, and associative; System 2 is slow, effortful, and deliberative. Most online behavior is System 1, which is why it is so predictable and so exploitable.
  • Heuristics and biases: availability (what comes easily to mind), anchoring (the first number or frame sticks), confirmation (we seek and credit what we already believe), representativeness (we judge by resemblance, not base rates).
  • Motivated reasoning: when identity is at stake, reasoning becomes a lawyer hired to defend a conclusion rather than a judge weighing evidence.
  • Social identity [Tajfel1979]: the mere assignment of people to groups produces in-group favoritism. People hold beliefs partly to belong, not only to be right.
  • Conformity and obedience: Asch’s line-length studies (Asch 1956) and Milgram’s obedience experiments show how powerfully social context shapes behavior.
  • Moral foundations (Haidt 2012): different audiences weight values such as care/harm, fairness, loyalty, authority, and purity differently. Persuasion that ignores this fails.
  • Attention as a scarce resource: variable rewards, infinite scroll, and dark patterns are engineered to exploit known psychology.
  • Loss aversion and hyperbolic discounting: losses loom larger than gains, and distant costs are heavily discounted—taxi drivers work longer on bad days, and “our future selves are strangers” (gym memberships, quitting smoking).
  • Status and relative position: people care about where they stand relative to others, not just about absolute level; this is why public status comparisons are so powerful.
  • Moral disengagement (Bandura 1999): euphemisms and relabeling let good people do harmful work without feeling like bad people.

F.8.3 Content

  1. Opening question (3 min): “Why did you click the last thing you clicked?”
  2. Two systems and biases (15 min): System 1/System 2, then the biases with everyday online examples. Emphasize: these are features, usually adaptive, but they are predictable, and predictability is exploitability. Add loss aversion and hyperbolic discounting: taxi drivers work longer on bad days, and “our future selves are strangers” (gym memberships, quitting smoking).
  3. Identity, belonging, and status (15 min): Minimal-group results; identity-protective cognition; Asch and Milgram as demonstrations that social context can overpower private judgment (including critique of Milgram’s findings). Moral foundations: why two reasonable people talk past each other about content moderation. Then status: it beats income. Norway’s public tax records widened the well-being gap 29% (Perez-Truglia), and Robert Frank’s local-rank finding shows people care about position, not level.
  4. People want data but believe stories (5 min): public-health campaigns use named people, not statistics; industries fund data-heavy think tanks for the appearance of rigor.
  5. Nudges, defaults, and the attention economy (12 min): Map the psychological mechanisms to specific design choices: variable rewards, manufactured urgency, disguised ads, social proof, default nudges. The nudge proof is strong: pension auto-enrollment took participation from 50-60% to 80-90% (UK 2012, ~10M workers), and the Behavioural Insights Team’s “nine out of ten pay on time” tax letter. A design choice is a psychological intervention, whether the designer acknowledges it or not.
  6. Moral disengagement (5 min): Bandura’s point that euphemisms let good people do harmful work—“user data as exhaust” and “engagement optimization” are the tech versions.

F.8.4 Exercises

Bias autopsy.
Take a recent online argument or viral post. Identify at least two specific biases at work (confirmation, availability, in-group signaling, anchoring) and say how the platform’s design amplified them.
Dark-pattern hunt.
In a product you use, find one dark pattern (forced continuity, disguised ads, manufactured urgency, infinite scroll, confirshaming). Name the psychological mechanism it exploits.
Identity check.
Think of a belief you hold partly because of group membership. Describe, honestly, what would happen to your social life if you changed it. Discuss what this implies for “if we just showed people the facts.”
Moral-foundations translation.
Pick a policy debate (e.g., whether to moderate misinformation). Argue it once from a care/fairness frame and once from a loyalty/authority/purity frame. Which audiences respond to which?
Design a nudge.
Choose a desirable online behavior (e.g., reading before sharing). Design one System 1 intervention (default, friction, social proof) and one System 2 intervention (education, disclosure). Predict which works better and why, then discuss the ethics of each.
Status check.
Identify one place where you compare yourself to others online (follower counts, salary, shipping speed). What does the comparison do to your satisfaction, and who benefits from keeping you in that comparison?
Nudge autopsy.
Find a default in a product you use (auto-renew, pre-checked boxes, a default privacy setting). Trace who set it, who benefits, and what the opt-out rate likely is.

F.8.5 Takeaway

People are not broken computers: they are social animals with fast, fallible, identity-driven cognition. Design that respects this is possible; design that exploits it is the business model of the attention economy.

F.8.6 Further reading

F.8.6.1 People Don’t Maximize Utility

The workshop says “dual-process” and “framing”; this is the canonical, 90-second demonstration to run in the room:

Program A would save exactly 200 people. Program B had a one-in-three chance of saving all 600 and a two-in-three chance of saving none. Most people chose A… Statistically, the two programs are identical, but this time, most people chose D. Nothing changed except how the outcomes were described.

Further:

Participants spun a wheel rigged to land on either 10 or 65, then estimated the percentage of African countries in the United Nations. Those who had seen 65 guessed about 45 percent higher than those who had seen 10.

The follow-through (“prosecutors set high anchor charges,” “retailers display high ‘original’ prices,” “speak first” in salary negotiation) connects it to participants’ own lives.

F.8.6.2 Why Do You Want What You Want?

The workshop treats preferences as a given, so briefly mention Bernays:

Women smoking in public was a social taboo; framing the act as feminist defiance was designed to dissolve that taboo and open the female market to tobacco sales. It worked.

This sets up the “recommendation systems are also preference-construction systems” claim, and pairs naturally with Veblen (“keeping up with the Joneses”) and the beauty-industry point that “the customer is never meant to arrive, because that would end the business model.”

F.8.6.3 Why Don’t People Just Say No?

The workshop mentions Milgram, but not the most useful finding for a workshop about change, which is the effect of a single dissenter:

Sixty-five percent of subjects administered what they had been told was the maximum voltage… because a person in a lab coat told them to.

F.8.6.4 Corporations are Psychopaths

The Broadmoor finding that corporate executives out-score the patients of a hospital for the criminally insane on psychopathy measures will stick with learners. Pair it with the Challenger “normalization of deviance” callout and the “we just build hammers” reframing in More Psychology.

F.8.6.5 Facts Alone Don’t Change Minds

This one line reframes the workshop’s “just show people the facts” discussion, and the Facebook Papers/Frances Haugen callout gives it a current, concrete anchor.

F.9 4) Inequality and stratification

Why are some people consistently better off, and is that “natural”?

F.9.1 Learning objectives

  • Contrast structural and individual explanations of social outcomes.
  • Explain cumulative advantage and opportunity hoarding.
  • Describe intersectionality and why single-axis analyses miss the worst harms.

F.9.2 Key concepts

  • Stratification (Weber 1946): three analytically distinct dimensions that overlap but do not coincide:
    1. Class (economic position)
    2. Status (prestige and honor)
    3. Power (capacity to realize one’s will)
  • Structural vs. individual explanation: attributing outcomes to systems and positions versus attributing them to personal traits and effort.
  • Cumulative advantage (the Matthew effect): advantage begets advantage; early differences compound over a lifetime.
  • Opportunity hoarding (Tilly 1998): groups use credentials, networks, and gatekeeping to keep advantages for themselves while appearing to reward merit.
  • Intersectionality (Crenshaw 1989): overlapping systems of disadvantage interact; they do not simply add.
  • Social mobility: intergenerational mobility is far lower and stickier than the “American Dream” story implies; “meritocracy” often functions as a legitimation story for inherited advantage.
  • Meritocracy (Young 1958): the word was coined in a 1958 satire; merit-pay organizations show larger gender gaps (moral licensing), and believing in meritocracy predicts harsher judgment of the poor.
  • Poverty and bandwidth (Mullainathan and Shafir 2013): scarcity depletes cognitive capacity the way sleep deprivation does, which is why “just try harder” is not a serious explanation.

F.9.3 Content

  1. Opening question (3 min): “When you explain why someone succeeded or failed, which verbs do you reach for?”
  2. Three dimensions (10 min): Weber’s class/status/power, with tech examples: a staff engineer may have high class but limited power; a founder has power; a moderator has neither. Mismatches matter. Wilkinson & Pickett add the cross-national finding: more unequal societies do worse on nearly every social indicator, and the mechanism is status anxiety.
  3. Structure vs. individual (15 min): The fundamental attribution error applied to society: we explain others’ outcomes by their traits and our own by circumstance. Cumulative advantage: early advantage compounds. Opportunity hoarding: how “merit” filters are built to reproduce advantage. Intersectionality: compounding and interacting disadvantage, and why systems (e.g., a facial-recognition model, a hiring algorithm) distribute harm unevenly across overlapping identities. Make discrimination concrete:
  • Audit studies: identical resumes with white-coded names get consistently more callbacks; Rivera’s “polished” versus “rough around the edges” is class, not competence.
  • Amazon’s internal recruiting tool (abandoned 2018) penalized resumes containing “women’s.”
  • Women’s share of computing was higher in the 1980s than today, so the “pipeline” explanation is contradicted by the evidence.
  • Becker’s 1957 prediction that competition eliminates discrimination was falsified; favored-group workers benefit from exclusion (LBJ: “give him somebody to look down on”).
  1. Poverty is a tax on cognition (8 min): scarcity depletes cognitive bandwidth like sleep deprivation (Mullainathan and Shafir 2013); McIntosh’s invisible knapsack explains why “just try harder” fails (McIntosh 1989). Credit bureaus collect without consent and profit whether the data is accurate or not (Equifax 2017: ~147M exposed); accuracy is not fairness.
  2. Meritocracy, examined (8 min): “Meritocracy” was coined by Michael Young in a 1958 satire. Merit-pay organizations show larger gender pay gaps (moral licensing (Castilla and Benard 2010)), and believing in meritocracy predicts harsher judgment of the poor.
  3. Mobility and measurement (5 min): Mobility is low and sticky across generations. Treat “meritocracy” as an empirical claim to test, not an assumption.

F.9.4 Exercises

Explain the gap.
Take a real inequality (who gets into a selective program, who receives venture funding). Produce two explanations (one individual, one structural) then discuss which better fits the evidence and why the individual one is the cultural default.
Three-dimension map.
For four tech roles (e.g., startup founder, staff engineer, content moderator, gig worker), rate class/status/power as high or low. Discuss the mismatches (e.g., high status but low power; high income but low security).
Cumulative advantage.
Trace how one early advantage like a wealthy school, a first internship, seed funding, or an early follower count compounds across a career. Identify the points where policy or design could interrupt the compounding.
Opportunity-hoarding audit.
Find a credential, interview process, or referral network that functions to hoard opportunity. Distinguish its stated purpose from its actual effect.
Intersectional case.
Analyze how a policy or product harms people differently across overlapping identities (a facial-recognition system, a hiring algorithm, an automated welfare screen). Why does a single-axis analysis miss the worst harms?
Merit audit.
Take a supposedly merit-based process (a hiring funnel, an awards program, a promotion ladder). Check whether it produces the outcomes meritocracy predicts, and identify where credentials and networks do the actual sorting.
Bandwidth check.
List five decisions you make easily when rested that become hard when you are tired or under financial pressure. What would a product look like if it assumed its users were often in that state?

F.9.5 Takeaway

Inequality is not a sorting of individuals by merit; it is the accumulated effect of rules, networks, and compounding advantage. Meritocracy is a claim to test, and often a story that legitimizes inheritance.

F.9.6 Further reading

F.9.6.1 The Hidden Ledger

This quote drives home how arbitrary classifications are:

In 1950, South Africa’s apartheid government created the Race Classification Board… The examiners used what they called the pencil test: if a pencil inserted into someone’s hair stayed in place without falling, the person might be classified as ‘Coloured’ rather than ‘White.’

The follow-through on siblings classified differently, with different rights, makes intersectionality concrete. The “Hispanic” category being invented by the Nixon administration in 1970 is the same point in a US key.

F.9.6.2 The Hidden Ledger

Cumulative advantage across generations, made measurable:

Neighborhoods that were redlined in the twentieth century still show lower average credit scores today: not because their residents are less creditworthy, but because they were systematically excluded.

F.9.6.3 Why Don’t You Just…

A scientific explanation of why “just try harder” is wrong, and why individual explanations are the default. The “Universal Credit/just go to the library” loop is the human-scale companion story.

F.9.6.4 Who Gets to Decide

Carnegie is an exemplary paradox: ten workers killed by Pinkertons in 1892, then a lifetime funding libraries “on his terms.” The Flexner Report is the sharper institutional example: it raised medical standards and “closed most of the schools that had trained Black physicians or admitted women in significant numbers.” Both are opportunity hoarding wearing a merit badge.

F.9.6.5 Women’s Work

The workshop doesn’t talk about gender as much as it should. Some points to highlight:

  • Iceland 1975: “ninety percent of Icelandic women refused to work at paid jobs, in the home, or anywhere else… The economy of Iceland effectively stopped.”
  • Waring on GDP (Waring 1988): a woman cooking for her family adds nothing to GDP, but if she hires a cook and works, GDP rises twice.
  • The “second shift” (Hochschild and Machung 1989): women in dual-income couples work “roughly a full additional month” per year, and rationalize it via “family myths.”

This also connects directly to the platform economy (care-work apps formalize undervalued labor “while keeping the wages and protections of the informal economy”).

F.9.6.6 More Psychology

The source for the meritocracy and audit-study material: “meritocracy” was coined in a 1958 satire, merit-pay organizations show larger gender gaps, and identical resumes get different callbacks depending on the name at the top.

F.10 5) Media, ideology, and public opinion

How do people come to believe what they believe about the world?

F.10.1 Learning objectives

  • Distinguish agenda-setting from framing.
  • Describe the shift from human gatekeeping to algorithmic engagement and its consequences.
  • Apply a “manufacturing consent” lens to platform-era media.

F.10.2 Key concepts

  • Agenda-setting: media may not tell you what to think, but they tell you what to think about. Absence from the agenda is a form of power (recall Lesson 1’s second face).
  • Framing: how an issue is defined shapes which answers seem reasonable. “Screen time” versus “attention extraction”; “AI safety” versus “corporate accountability.”
  • Gatekeeping and its collapse: for much of the twentieth century a small number of editors and broadcasters decided what was newsworthy. That concentration had its own distortions: platforms replaced it with engagement-maximizing algorithms, which are a different and arguably worse concentration.
  • The propaganda model (Herman and Chomsky 1988): five filters (ownership, advertising, sourcing, flak, and ideology) shape what gets through. As a result, the range of acceptable debate is narrower than the range of actual opinion.
  • Epistemic fragmentation: filter bubbles and polarization, with the caveat that the empirical “echo chamber” story is more mixed than popular accounts claim; the larger problem may be a shared but distorted diet, not total separation.
  • Moral panic: a disproportionate, often exaggerated public alarm that can coexist with a real, evidence-backed harm, and that often targets the wrong harm while leaving the real one unaddressed.
  • The dependence effect (Galbraith 1998): advertising manufactures the wants that production then satisfies.
  • Narrative economics (Shiller 2015): markets and beliefs move on contagious stories as much as on fundamentals.
  • Privacy as framing: who gets to be private and who is watched is itself a contest; “privacy for persons, transparency for institutions” is one attempt to settle it.

F.10.3 Content

  1. Opening question (3 min): “What did you learn about the world this week, and who decided it was worth your attention?”
  2. From gatekeeping to platforms (10 min): The twentieth-century gatekeeper and its problems; the platform replacement and its problems. Concentration did not go away; it changed shape and added a profit motive tuned to engagement.
  3. Framing and agenda (12 min): Agenda-setting and framing with concrete before/after pairs. Show that a reframe changes the feasible policy menu. Add Galbraith’s dependence effect: advertising manufactures the wants that production then satisfies. And representation changes what is thinkable: Nichelle Nichols’s Uhura (MLK, Whoopi Goldberg, Mae Jemison), and the Rooney Rule as a minimal intervention.
  4. The propaganda model (12 min): Herman and Chomsky’s five filters, updated for platforms (engagement, algorithmic amplification, influencer economies). Manufactured consent: the center of gravity of “respectable” debate is not where public opinion actually is. Add narrative economics (Shiller): markets move on contagious stories. And note the ideology of the future: TESCREAL/longtermism (Torres) lets the already-powerful frame themselves as “trustees of civilization” and justify present costs for an unknowable future.
  5. Privacy and surveillance as framing (12 min): who gets to be private, and who is watched.
  • Goffman’s front/back stage (Goffman 1959): “a society without back stages is one where performance is continuous and exhausting”; Snowden’s measurable chilling effect on Google searches.
  • “Paying for the privilege”: the Stasi had to coerce informants; we pay subscription fees to be surveilled.
  • The first privacy right (Warren & Brandeis, 1890) protected the prominent from gossip, not the powerless from the state; “nothing to hide” is a privileged argument (Browne 2015).
  • Financial secrecy protects power: Swiss banking law 1934 (Jewish depositors to Nazi loot), Zucman’s ~$7.6T offshore, the Panama/Pandora Papers.
  • The balance: “privacy for persons, transparency for institutions” (India’s RTI Act 2005, South Africa’s TRC, and the double edge of the right-to-be-forgotten).
  1. Epistemic fragmentation (8 min): Polarization and filter bubbles, with the empirical caveat. Acknowledge uncertainty rather than overclaiming; this is what a rigorous progressive account does. Disasters reveal mutual aid, not barbarism (Quarantelli/Solnit): “elite panic” versus ordinary generosity, and media selection bias manufactures mistrust.

F.10.4 Exercises

Framing audit.
Find one story covered two ways (e.g., “misinformation” versus “platform accountability”). List the frames and metaphors, and note who is cast as the agent and who as the victim.
Agenda check.
Compare what a news outlet and a social-media feed surfaced about the same event. What was on the agenda, and what was conspicuously absent? Who benefits from that absence?
Propaganda filters today.
Take a tech-policy controversy. Walk through Herman and Chomsky’s five filters as updated for platforms, and identify which are most active in this case.
Manufacturing a panic.
Take a moral panic (e.g., kids and screens, “AI will end humanity”). Trace how news coverage and platform feeds frame it: which experts get quoted, what metaphors recur, who is cast as the villain. How does the framing amplify the alarm while suppressing the evidence?
Redesign the gatekeeper.
If you had to replace engagement maximization with a different editorial principle, what would it be? Name the new distortions it would introduce.
Privacy frame swap.
Take a surveillance story (ad targeting, workplace monitoring, a data breach) and retell it once from the “personal privacy” frame and once from the “power and institutions” frame. Who is the agent and who is the victim in each version?
Representation audit.
Pick a domain where one kind of person is overwhelmingly represented (an executive suite, a cast list, a dataset of “successful” people). What becomes thinkable when that changes, and what is the minimal intervention (like the Rooney Rule) that gets the change started?

F.10.5 Takeaway

Belief is produced, not just discovered. Whoever controls the agenda and the frame controls a large part of what is thinkable. Platforms changed the economics of that production, and with it, the politics.

F.10.6 Further reading

F.10.6.1 Why Do You Want What You Want?

The propaganda model’s origin story connects back to lesson 3:

Women smoking in public was a social taboo; framing the act as feminist defiance was designed to dissolve that taboo and open the female market to tobacco sales. It worked.

F.10.6.2 Big Tech is Like the Penny Press

Reframes “media got worse” as a market structure, not a conspiracy:

Neither set out to destabilize democratic discourse; they just wanted to sell newspapers. However, the market provided a clear signal: crime, scandal, nationalist outrage, and stories written to produce emotional responses drove circulation, and each escalation by one paper forced a matching response from the other.

F.10.6.3 The Eight-Hour Day

A concrete example of agenda-setting done via a calendar:

The Haymarket affair became a rallying point for the international labor movement; May Day was adopted as a workers’ holiday across most of the world (with the notable exception of the United States and Canada, which moved their Labor Day to September specifically to avoid the association).

F.10.6.4 The Psychology of Private Space

For the “manufacturing consent” discussion, the Stasi numbers are unforgettable (~90,000 officers, ~180,000 informants surveilling ~16 million people) and support the claim that surveillance rewires the people it surveils.

F.10.6.5 What We Owe the Future

TESCREAL/longtermism (Torres and Gebru 2024) is the ideology to watch in this lesson: framing AI’s builders as “trustees of civilization” lets the already-powerful justify present costs for an unknowable future.

F.11 6) Deciding what is harmful and how to regulate it

How does a society decide what counts as harm, and what should it do about the harms it recognizes?

F.11.1 Learning objectives

  • Explain why what gets labeled “harmful” and therefore gets regulated does not track a neutral measurement of damage.
  • Contrast the rare-dramatic-attributable model of harm with the diffuse-pollution model, and say which one describes social media and AI.
  • Distinguish a moral panic from an evidence-backed harm, and say what each implies for regulation.
  • Name the instruments by which harm is regulated and the failure mode of capture.

F.11.2 Key concepts

Harm is not self-evident.
What counts as harm, and which harms get regulated, is decided through contests of power, moral panic, racial hierarchy, and commercial interest, not by first measuring damage and then writing a law. Rare, dramatic, and attributable vs. diffuse, cumulative, and statistical.
Engineers are trained on the first model: the faulty valve that causes a boiler explosion. The worst industrial harms (leaded gasoline, tobacco, asbestos, opioids, and now social media) are the second: real and massive, but spread across millions of small exposures so that no single decision can be shown to have caused a single injury. The pollution model.
Once a society accepts that harm can be diffuse and cumulative, responsibility shifts from proving individual causation to holding the emitter accountable for the aggregate. Engineers now study pollution not out of conscience but because liability eventually forced them to. The differential legalization of pleasure.
Alcohol kills hundreds of thousands a year and sits in supermarkets; cannabis, with a lower harm profile, was criminalized for most of a century, with enforcement falling on Black and Latino communities. What gets legalized tracks political economy and racial hierarchy far more reliably than it tracks harm. Precautionary principle and product safety.
Where harm is plausible and irreversible, the burden of proof may rest on those introducing the risk. Pre-market drug approval is the model. Regulation as a toolkit and regulatory capture.
recap of standards, licensing, disclosure, liability, taxation, and structural remedies, and the recurring failure mode in which regulators come to serve the regulated. The uncertainty playbook and the industry response.
When a harm is alleged, the standard playbook is to fund scientists, insist correlation is not causation, call regulation an attack on freedom, create neutral-sounding front groups, and delay. The industry response then moves through four phases: denial, uncertainty, the market argument, and finally acceptance— where compliance reliably turns out cheaper than predicted. Non-identity.
Some harms, especially those to “future people,” cannot name an individual victim; this is the philosophical twin of diffuse, cumulative harm, and it explains why such harms are so easy to postpone.

F.11.3 Content

  1. Opening question (3 min): “Which is more dangerous: alcohol or cannabis? How do you know, and does the law match your answer?”
  2. Sex, drugs, guns, and privacy (20 min): four cases that show harm is contested, not measured.
  • Sex. What was “private” versus what was “criminal” has been redrawn case by case, in someone’s interests, with no consistent underlying principle. Sexual behavior between consenting adults was legally public (in the sense of being criminally regulated) well into the twentieth century; the expansion of privacy was won, not discovered. Sodomy laws are a Victorian legal export (1861/1885, Oscar Wilde) spread through colonial law, not an ancient universal.
  • Drugs. Alcohol kills hundreds of thousands a year and is sold in supermarkets; cannabis, with a lower measured harm profile, was a criminal offense whose enforcement fell almost entirely on Black and Latino communities. Portugal’s 2001 decriminalization of personal possession, paired with treatment and harm reduction, cut HIV transmission and overdose deaths without raising use.
  • Guns. After Columbine, legislators went after violent video games; the evidence never supported the hypothesis, and cross-national comparisons point to gun availability, inequality, and the history of racially organized violence as the real drivers. Motivated reasoning (Lesson 3) means even a trained scientist who owns a gun scrutinizes gun-violence studies more skeptically than climate studies.
  • Privacy. The category “privacy” is contested in exactly the same way: its historical unit was the household, not the person (“the personal is political”), and marital rape was legal in most US states until the 1970s-80s. What counts as a protected, private sphere is won and redrawn, not discovered.
  • The point is that we do not first measure harm and then regulate. We first decide who and what matters, then find or manufacture the harm that justifies the decision.
  1. Two models of harm (10 min): rare/dramatic/attributable versus diffuse/cumulative. The engineer’s default is the first; the worst damage from industry has been the second. Leaded gasoline lowered a generation’s IQs without any single tank of fuel causing a measurable injury; no particular cigarette caused any particular cancer death. Social media’s harms are this kind (so-called “cognitive pollution”), so the causal chain is long, probabilistic, and hard to attribute, and therefore hard to regulate. Parfit’s non-identity problem is the philosophical twin: harms to “future people” cannot name an individual victim, which is why they are so easy to postpone and so hard to regulate.
  2. The uncertainty playbook (12 min): when a harm is alleged, the standard industry playbook runs in five steps—fund scientists, insist correlation is not causation, call regulation an attack on freedom, create neutral-sounding front groups, and delay. The response then moves through four phases: denial, uncertainty, the market argument, and finally acceptance—and compliance reliably turns out cheaper than predicted. The tobacco industry ran this script for decades after Hill and Doll’s 1950 study; the same script ran for leaded gasoline, asbestos, and oxycontin, and is running now for social media and AI.
  3. Pollution and dangerous pharmaceuticals (20 min): two cases where regulation worked, and why.
  • Pollution. The remedy was to hold the emitter liable for the aggregate, not for a specific injury. The ozone layer was protected before it cost lives: proof we need not wait for disaster. Minamata is the same lesson read backwards—Chisso’s mercury poisoning, with cats as the ignored early warning. The Sandoz/Rhine spill (1986) produced the binding Rhine Action Programme, showing that a cross-border disaster can be converted into enforceable rules.
  • Dangerous pharmaceuticals. Thalidomide was approved in West Germany in 1957; Frances Kelsey, a pharmacologist at the FDA, refused to approve it in the US because the safety evidence was insufficient, and the birth defects that followed where it was approved vindicated her. The result was pre-market approval and the precautionary principle. Kefauver-Harris
    1. is the legislative landmark: before it, you could sell a drug in the US without proving it did anything.
  • The limits of a settlement. In the 1998 Master Settlement Agreement the states sued to recover $206 billion in Medicaid costs from the tobacco companies, but this was a settlement, not a solution: it monetized the harm rather than removing it.
  • What worked. Accurate diagnosis of the actual harm; the right instrument; and a regulator that was not captured.
  1. When regulators fail (10 min): capture is the recurring enemy. Wirecard shows the extreme version: Germany’s regulator attacked the journalist instead of the fraud. Korean chaebol executives were convicted and then pardoned as too important to jail: “too big to jail” is not a tech invention.
  2. Moral panic (8 min): recall the moral-panic concept from Lesson 5; it has a lineage—dime novels, then Wertham’s comics, then rock and roll, then D&D, then video games—and cross-national checks reveal how fragile most claims are (Japan and South Korea have high game use and low violence). The real harm in the video-game panic was loot boxes, not gunfire.
  3. Recap and bridge (3 min): Social-media and AI harms are pollution-model harms. The trap is moral-panic framing (e.g., “ban the kids’ phones”) instead of the actual harm: attention extraction, addictive design, discrimination, surveillance. Lessons 6 and 7 apply this to media and to change.

F.11.4 Exercises

Rank the harms, then the laws.
Consider the list “alcohol, cannabis, heroin, gambling, social media, AI-generated misinformation”. Rank the items by measured harm, then by how the law actually treats them. Discuss the gaps and what explains them.
Two models of harm.
Pick a harm (e.g., a bridge collapse, leaded gasoline, a data breach, or teen anxiety from Instagram). Classify it as rare-dramatic-attributable or diffuse-cumulative, and say which regulatory instrument each model naturally suggests. Which model were you trained to think in?
Moral panic vs. real harm.
Choose a concern (violent video games, kids and phones, “AI will end humanity”). List the features that make it look like a moral panic and the features that make it an evidence-backed harm. What is the actual harm the panic may be distracting from?
Precautionary vs. wait-and-see.
Replay the thalidomide decision: you are the regulator with evidence that is suggestive but not conclusive. Do you approve or wait? What does your answer imply for “move fast and break things” and for releasing open-weight AI models?
Redesign the law.
Pick a harm currently regulated on a moral or panic basis rather than an evidence basis (e.g., cannabis, a social-media feature). Propose a regulation that tracks the actual harm, and name the political obstacles to passing it.
Beat the playbook.
You are the regulator and the industry is running the five-step uncertainty playbook. For each step, name one counter-move (independent funding, burden-of-proof rules, naming the front groups, deadlines).
Settlement or solution.
Study the 1998 Master Settlement Agreement. Why did a $206 billion settlement fail to end the tobacco problem? What would a solution have required, and why was a settlement the outcome the parties preferred?

F.11.5 Takeaway

Harm is not a fact that precedes regulation; it is a claim made by the powerful, the panicked, and the organized. Regulating well means diagnosing the actual harm rather than the one that generates the best testimony, using the pollution and product-safety models where they fit, and defending the regulator from capture.

F.11.6 Further reading

F.11.6.1 Unsafe at Any Speed

Nader’s story can be told in just a few seconds:

GM’s response was not to fix the car; it was to hire private detectives to dig up dirt on Nader.” … “GM’s president was summoned to testify before the United States Senate and had to apologize on national television.

F.11.6.2 How the Rivers Ran Again

A good example of how framing a harm precedes regulating it, and about regulation working:

On June 22, 1969, the Cuyahoga River in Cleveland, Ohio, caught fire. This sounds dramatic, but it was also the thirteenth time the river had caught fire since 1868.

Two more “regulation that worked” cases with vivid sensory detail: the 1952 London smog (“so thick that people could not see their own feet,” with a spokesman blaming “influenza”), and going from the Great Stinnk of 1858 to “In 1983, a salmon was caught in the Thames for the first time since the 1820s.”

F.11.6.3 Forty Years to an Agreement

The ISDS callout is the best single example of regulatory capture’s international cousin, and it has a punchline:

a move so transparently opportunistic that even the arbitrators were not impressed.

F.11.6.4 Unsafe at Any Speed

A rare example of a firm choosing public good that keeps Lesson 6 from being uniformly grim:

Nils Bohlin, an engineer at Volvo, had developed the three-point belt the year before, and Volvo had made the patent freely available to every manufacturer in the world.

F.11.6.5 The Tontine

The tontine is a one-paragraph warning about how market structures can build in harm that doubles as the opening image for “the monopolist’s playbook”.

F.11.6.6 The Eight-Hour Day

A good example of reframing a harm to make it regulable:

The managers had locked the stairwell doors to prevent workers from taking unauthorized breaks and to stop theft of fabric. 146 workers died, most of them young immigrant women, many of them Jewish and Italian. Some jumped from the windows. The owners were acquitted of manslaughter charges and collected the insurance.

F.11.6.7 The Monopolist’s Playbook

For the “regulation can fail” part of the lesson, this punchline is hard to beat:

What is less often noted is that several successor companies immediately reconsolidated, and that John D. Rockefeller’s personal fortune increased after the breakup as the stock prices of the subsidiaries rose.

F.11.6.8 What We Owe the Future

Parfit’s non-identity problem is the philosophical twin of diffuse, cumulative harm: if our choices change who is born, no particular future person can say we made them worse off, yet the choice still has victims in the aggregate.

F.11.6.9 Privacy, Power, and the Self

The “privacy is contested” point extends Lesson 6’s opening claim about harm: the historical unit of privacy was the household, not the person, and marital rape was legal in most US states until the 1970s-80s.

F.12 7) How change happens

Given all of this, what actually produces change, and what should we do about social media and AI?

F.12.1 Learning objectives

  • Describe the conditions under which social movements succeed.
  • Explain a “policy window” and assess whether one is open.
  • Connect a specific social-media or AI harm to each of the day’s six prior lessons.
  • Sketch a realistic reform strategy and name their own leverage as a technologist.

F.12.2 Key concepts

  • Social movements and collective action: movements succeed when they combine organization, opportunity, and framing (resource mobilization and political-process traditions).
  • Policy window (Kingdon 1995): when the problem stream (a defined crisis), the politics stream (a constituency and mood), and the policy stream (a ready-made solution) align, change becomes possible. Windows open rarely and close quickly.
  • Coalitions: durable change usually requires unusual alliances across groups that do not otherwise agree.
  • Exit, voice, and loyalty (Hirschman 1970): the three responses to a failing institution, and why voice is usually the only one that changes the institution itself.
  • Technologists’ leverage: scarce skills, access to systems, and inside knowledge vs. constraints (NDAs, golden handcuffs, the myth of the “neutral engineer”).
  • Worker power: institutions such as codetermination, sectoral bargaining, and arbitration have delivered durable gains; they are won, not conceded.
  • The five tools: collective organizing, regulatory pressure, alternative structures, strategic communication, and sustained participation are an alternative taxonomy of the reform menu.

F.12.3 Content

  1. Opening question (3 min): “Name one thing that is better now than it was fifty years ago. Who made it better, and how?”
  2. How change happens (15 min): Organization + opportunity + framing. Examples: civil rights, environmental regulation, seat belts, tobacco, marriage equality, worker safety. Emphasize: change is slow, non-linear, and won by pressure, not persuasion alone. Kingdon’s three streams and the policy window:
    • Problem: is it defined as a crisis?
    • Politics: who is the constituency?
    • Policy: is there a ready solution? Add three more mechanisms: ACT UP (1987) used direct action to accelerate drug approval and changed FDA timelines; the expanding circle distributed slave narratives (Equiano, Douglass) as strategy and ran the 1790s sugar boycott as the first mass consumer boycott; Kerala’s 1969 land reform was won by election, not occupation, and paid off in literacy and life-expectancy gains once the landlord class’s veto was broken.
  3. Worker power (10 min): union wins were not conceded—the Combination Acts, the Tolpuddle Martyrs, the 1926 General Strike, Thatcher, and the eight-hour day that came only after WWI’s fear of revolution. The institutions that made them stick: German codetermination (workers hold half the board), Australia’s arbitration and “basic wage” (Harvester 1907), and COSATU’s role in ending apartheid. Rana Plaza (1,134 dead, 2013) led to the Bangladesh Accord, a real gain, then weakened, then re-won (two steps forward, one step back).
  4. The technologist’s role (10 min): Insider/outsider power. Engineers can refuse, whistleblow, organize, build alternatives, or work on policy. Their leverage is unusual (scarcity, access) and their constraints are real (contracts, career risk, the ideology of neutrality). Two warnings: the passion principle (Cech 2021) turns “do what you love” into a way for employers to extract—karoshi (“death from overwork”) is its Japanese endpoint—and the software-engineering literature covers architecture and management but not workers’ rights or cooperatives, the way medicine once hid its error rate (44k-98k preventable deaths/year).
  5. Applying the toolkit (5 min): Name the harms (attention extraction, misinformation, polarization, discrimination, labor deskilling, surveillance, concentration, environmental cost). Map each to the day’s lessons.
  6. Outline a realistic menu (8 min): transparency and audits, liability, structural remedies (interoperability, breakups), worker power, public alternatives, global coordination. Be honest about capture risk for each. DMA 2024 (gatekeepers, fines up to 20% of global revenue) and India’s UPI / Brazil’s Pix are “public rails before dominance.”
  7. Synthesis and close (5 min): The arc of the day: “how society works” is knowable, and knowing it is the prerequisite for changing it. The finale’s five tools are an alternative reform-menu taxonomy: collective organizing, regulatory pressure, alternative structures, strategic communication, and sustained participation (the Federalist Society’s 40-year pipeline). “Refusing to vote or join a union is a political act”; refusing to engage simply delegates the decision to whoever will.

F.12.4 Exercises

Policy-window map.
Pick one concrete harm. For Kingdon’s three streams: assess whether a window is open and what would open it.
Coalition design.
Choose a reform (e.g., mandatory algorithmic audits). List likely supporters, likely opponents, and the unexpected allies (parents, small business, civil-liberty groups, religious communities) that could make it a winning coalition.
Leverage inventory.
As a technologist, list your specific sources of leverage (skills, access, capital, community) and one concrete action you could take in each of the next 1, 6, and 12 months.
Harm-to-lesson mapping.
Take one harm (e.g., AI-driven hiring discrimination) and write one sentence connecting it to each lesson. The point is to demonstrate that every real harm has structural, economic, psychological, stratificational, regulatory, and ideological dimensions.
Reform-menu debate.
Assign groups different positions (e.g., audits only, liability, breakups, public options, worker power, “it cannot be fixed”). Each group argues its case; then the room discusses which combination is realistic and coherent.
Backward plan.
Pick a target outcome five years out (e.g., “audits of high-impact models are mandatory and meaningful”). Work backward to name the milestones, coalitions, and political conditions required at each stage.
Worker-power audit.
Choose one workplace harm (surveillance, deskilling, on-call exhaustion, pay opacity). Which of the worker-power institutions (codetermination, sectoral bargaining, arbitration, a union) would address it, and who would have to move first?
Five tools.
Take a single reform (e.g., algorithmic audits) and map it onto the five tools—collective organizing, regulatory pressure, alternative structures, strategic communication, sustained participation. Which tool is missing from most tech-industry versions of this plan?

F.12.5 Takeaway

The harms are not mysterious and the responses are not untried. Every tool discussed in the workshop is available to people who organize. Technologists hold real cards; the question is whether they play them.

F.12.6 Further reading

F.12.6.1 The 3.5% Threshold

The 3.5% threshold of (Chenoweth and Stephan 2011) made concrete:

in a city of half a million people, it is seventeen thousand people doing something instead of just complaining.

F.12.6.2 A Paradise Built in Hell

The workshop’s change stories are all about pressure; this adds the capacity side (what gets built between crises). The Halifax Explosion (1917) and the Hurricane Katrina media-lie case are both ready to tell:

Subsequent investigation found that the reported gang violence did not happen, the murder rate in the city did not spike, and most of the ‘looting’ was people taking food and water to survive. But these lies had consequences.

F.12.6.3 The Ozone Hole That Closed

The workshop mentions the ozone layer in passing; this is the full, hopeful case study of why Montreal worked:

The mechanism was a binding international agreement with differentiated obligations, a technology transfer fund, and trade sanctions against non-participants.

F.12.6.4 The Eight-Hour Day

The workshop lists worker safety among “how change happens” examples; this is the origin story with a takeaway:

It was the first time anywhere in the world that workers had achieved the eight-hour day; they did it not through moral persuasion but by applying collective leverage at a moment of their opponents’ weakness.

F.12.6.5 Big Tech is Like the Yakuza

A hopeful, non-obvious change story: a structurally powerful industry shrunk not by becoming less useful but by “a sustained political decision… to make the cost of association with them prohibitive”:

Registered yakuza membership fell from roughly 180,000 in the 1960s to under 20,000 by the early 2020s. None of this happened because the yakuza became less useful.

F.12.6.6 Alternatives That Scaled

The workshop’s “public alternatives” bullet has no examples. These are the existence proofs: Red Vienna’s Gemeindebau (still ~220,000 municipally owned apartments), Amul (3.6 million farmer-members), Desjardins (~8 million members, survived 2008 “in considerably better shape” than the big banks), and Mondragón (“running for nearly seventy years”) are not arguments in a seminar: they are institutions that issue annual reports.

F.12.6.7 What Collective Action Has Achieved

A recent proof that technologists’ leverage is real, for the “technologist’s role” section:

When the strike ended, the WGA had won explicit contract language: AI cannot write or rewrite scripts, and scripts cannot be used to train AI systems.

F.12.6.8 Land to the Tiller

The workshop’s change stories are mostly bottom-up; this is the rare top-down case that reframes what “structural remedy” can mean:

Under American military occupation, Japan’s agricultural land was seized from landlords and sold to the tenant farmers who had been working it, at prices set well below market value, paid in bonds that inflation promptly turned into confetti. This was expropriation, and it worked.

F.12.6.9 Conclusion

The “five tools” taxonomy and the ACT UP / expanding-circle stories: the reform menu is not just regulation—it is organizing, alternatives, communication, and sustained participation.

F.12.6.10 What We Owe the Future

Dario Amodei’s own warning is the sober close to the “technologist’s role” section: AI severs “the average person’s leverage through economic value”—the material basis of democracy itself.

F.13 Closing synthesis (25 min)

If there is time left, use it to make the day’s through-line explicit, not to introduce new material.

  1. Power and institutions set the rules; markets are made, not natural; people are social and fallible; inequality is produced, not earned; media manufacture the thinkable; harm is contested, not measured; regulation can and does work when it is not captured; change comes from organized pressure.
  2. Ask each participant to write one sentence naming a specific social-media or AI harm, the lesson that best explains it, and one realistic response. Go around the room; collect the sentences.
  3. Close on the future: Berlin’s “crooked timber of humanity” warns that monism (one supreme value) is the root of tyranny—the trap the longtermist frame triggers—and “don’t foreclose options” is the most defensible version of what we owe the future.
  4. Point to the reading list below and, if running this with colleagues, to a follow-up session or reading group.

F.14 Reading list

  1. Power and institutions
  2. Markets are made, not natural
  3. How people actually think
  4. Inequality and stratification
  5. Media, ideology, and public opinion
  6. Deciding what is harmful and how we regulate it
  7. How change happens, and applying it all

On social media and AI specifically:

Miscellaneous:

F.15 Appendix: harm-to-lesson map

A cheat sheet for the synthesis in Lesson 7.

  • Attention extraction and addiction
    • Power: engagement design is agenda-setting for a billion minds.
    • Markets: the cost of degraded attention is an unpriced externality.
    • Psychology: variable rewards exploit System 1 and in-group signaling.
    • Inequality: the cost falls hardest on those with the fewest resources to opt out.
    • Media: framing it as “screen time” blames users rather than the extractor.
    • Harm & regulation: a diffuse, cumulative harm (the pollution model), not a dramatic one; regulate the emitter, not the user, and beware the “screen time” moral panic.
  • Algorithmic discrimination
    • Power: the “neutral model” hides a rule that allocates advantage.
    • Markets: information asymmetry (the applicant cannot see the decision rule).
    • Psychology: automation bias leads humans to defer to a confident number.
    • Inequality: training data encode and compound historical stratification.
    • Media: coverage frames it as a “glitch” rather than a design choice.
    • Harm & regulation: diagnose the actual harm (discrimination), then pick the instrument and guard against capture.
  • Election misinformation and polarization
    • Power: whoever sets the feed’s agenda shapes the thinkable.
    • Markets: misinformation is a negative externality; trust is a public good.
    • Psychology: confirmation bias and identity-protective cognition.
    • Inequality: targeting is cheap precisely where people are most isolated.
    • Media: the collapse of gatekeeping replaced editors with engagement.
    • Harm & regulation: a pollution-model harm; the pacing problem and global-versus-national jurisdiction complicate the remedy.
  • Surveillance and labor deskilling
    • Power: surveillance shifts the balance of power between worker and firm.
    • Markets: data is a barrier to entry and a source of market power.
    • Psychology: ambient surveillance changes behavior even without explicit threat.
    • Inequality: monitoring falls heaviest on the least powerful workers.
    • Media: “productivity” framing hides the redistribution of power.
    • Harm & regulation: worker protection and privacy are proven regulatory domains; the challenge is diagnosis and capture, not feasibility.
  • Concentration of the industry
    • Power: a handful of firms set the rules of the game for everyone else.
    • Markets: network effects produce natural-monopoly tendencies.
    • Psychology: switching costs exploit inertia and default bias.
    • Inequality: returns accrue to owners, not to the workers who produce value.
    • Media: owned media and advertiser dependence shape the coverage of tech itself.
    • Harm & regulation: antitrust is the classic structural remedy, weakened and due for revival; self-preferencing is payola with a new interface.