🔮 The prediction business sells obedience as foresight
Carissa Véliz on surveillance, democracy, and predictions as speech acts
What you probably do not know yet
- Digital surveillance in democracies now collects thousands of data points per person, at a scale East Germany could only afford for about a third of its population.
- AI scores for hiring, loans, and health look like neutral forecasts; but they are bets on your behavior that institutions act on.
- Headlines about existential AI doom distract from immediate threats to privacy, accountability, and anonymous protest.
What you will know after
Spot when a “prediction” is really a prescription, why models trained on the past can blind institutions to systemic risk, and what resistance looks like (privacy tools, skepticism, time offline).TL;DW
Surveillance got cheaper than East Germany
East Germany’s Stasi was expensive and human-heavy: it could only watch roughly a third of the population. Today’s stack is different. Phones, browsers, and platforms collect thousands of signals per person, cheaply, across whole populations.
“Nothing to hide” protects you from power, not guilt
Véliz flips the comfortable line (if you did nothing wrong, you have nothing to hide). Privacy limits what powerful actors can infer and do to you and to people connected to you.
You already hide your PIN without feeling guilty. The same logic applies at scale: data is often collected far beyond what a service needs (sexual preference data to sell handbags is her blunt example). The risk is abuse of power later, not proof of wrongdoing today.
Data is collected for power, not only profit
Mass data buys prediction and influence: who will click, vote, default, or protest. That is power over your next move, not just ad revenue.
Companies use data to blur who is responsible and to steer political outcomes. Palantir-style state tooling and everyday apps that nudge your next click build soft authoritarian infrastructure where convenience hides who steers you.
Predictions are speech acts, not weather reports
A hiring model that “predicts” you will underperform is not a passive description. It reads like fact and acts like instruction: a move that shapes what happens next (you do not get the interview). Executive prophecy (“you will all use AI tomorrow”) works the same way: it recruits you to fulfill someone’s preferred timeline. That combo produces obedience in advance.
Medicine and chatbots show where “prediction” dodges blame
COVID saw dozens of medical AI tools deployed without proper validation; meta-analyses later found many useless or harmful. Early cancer detection sounds great until overdiagnosis and false positives dominate the tradeoff. Véliz argues for predicting populations and things, not individuals, when individual labels can become self-fulfilling.
The Air Canada chatbot case: wrong grief-fee advice, then the company tried to treat the bot as a separate legal speaker. Courts pushed back, but the pattern is the point: predictions and bots become shields when institutions want to deny responsibility.
Chapter Guide
| Time | Chapter |
|---|---|
| 0:00 | Intro AI surveillance and democracy |
| 2:11 | Origin Véliz’s path into privacy ethics |
| 4:10 | Scale Mass surveillance vs East Germany |
| 6:32 | Fallacy “Nothing to hide” |
| 8:10 | Power Data collection beyond profit |
| 9:23 | Democracy Privacy for law, press, protest |
| 11:23 | Oracle AI as future-shaping prophecy |
| 14:34 | Speech acts Predictions as commands |
| 17:36 | Turkey Past data and systemic risk |
| 22:23 | Panic Existential hype vs real threats |
| 26:57 | Medicine COVID AI and overdiagnosis |
| 30:23 | Accountability Air Canada chatbot case |
| 36:30 | Bullshit Frankfurt and plausible AI text |
| 43:21 | Warfare Responsibility diffusion |
| 51:11 | Sentience Flattery, harm, human bonds |
| 59:03 | Resistance Analog life and privacy tools |
About 69 minutes with Myriam François (The Tea) and Carissa Véliz (author of Privacy Is Power and Prophecy).