Véliz: Beware the Power of Prediction (TED 2026)
Summary: Carissa Véliz argues that predictions — especially AI-driven ones — are not neutral forecasts but speech acts with political power: they shape the futures they claim to describe, invite manipulation, and create uncontestable injustice.
Sources: Transcripts/Véliz_2026_04_transcript.md
Last updated: 2026-05-07
Core argument
We treat predictions as knowledge claims — descriptions of the probable future. Véliz argues most social predictions are better understood as power moves: they generate self-fulfilling prophecies, justify ideologically loaded decisions under the guise of facts, and foreclose contestation.
“Predictions are often power plays in disguise. They justify value-laden decisions under the pretense of facts.”
Predictions as speech acts
Véliz draws explicitly on the philosophy of language: predictions about people function as what Austin called speech acts — language that does something rather than merely describes.
“Social predictions are veiled commands. They implicitly tell us how to act.”
When a tech executive declares that AI will be used for everything, everywhere, the prediction is not neutral: it tells audiences how to act, funds the infrastructure that makes the prediction come true, and profits from the compliance. Believing the prediction is a form of obedience.
“Predictions about the weather don’t influence the weather. Predictions about people influence people. Social predictions tend to act like magnets. They bend reality towards themselves.”
Compare austin-speech-acts: Austin’s performatives change the world by being uttered under the right conditions. Véliz’s social predictions work similarly — they constitute social reality rather than merely reporting it.
Self-fulfilling prophecies supercharged by AI
“An algorithmic prediction about future disease can make someone’s insurance premiums go up, leading to worse health outcomes from stress alone.”
The mechanism is a feedback loop: prediction → changed behavior/conditions → outcome that confirms prediction. AI supercharges this because predictions are now made at scale, at speed, and with an appearance of scientific objectivity that suppresses challenge.
The Oedipus story: had Oedipus dismissed the prophecy as noise, he would never have fled and triggered the chain of events. The prophecy was causal, not merely foreknowing.
Predictions and contestability
“Algorithmic predictions are building this Kafkaesque world in which we can no longer contest decisions because they’re not based on clearly defined criteria.”
“Predictions are never facts. Facts belong to the past. Predictions are unverifiable, unfalsifiable. Since they are about the future, they cannot be challenged for being false, thereby creating the perfect recipe for hidden injustice.”
If a loan is refused because the applicant fails a verifiable criterion, the refusal can be challenged. If it is refused on the basis of a statistical prediction, there is no foothold for contest: the prediction concerns a future that has not happened yet. This forecloses redress.
“Predictions are often unfair because they’re not based on who people are, but on who we think they will become.”
This connects to algorithmic-fairness: the incompatibility theorems show that no fairness metric resolves the underlying value conflicts. Véliz adds a deeper problem — predictions are unchallengeable by design.
Prediction markets and manipulation
Prediction markets claim to aggregate knowledge via financial incentives. Véliz exposes the flaw: if prediction is a power game, large actors can manipulate markets by betting heavily to shift public perception. In February 2026, six anonymous accounts earned $1.2M betting on the attack on Iran, with wallets funded hours before the event.
Surveillance and authoritarianism
“The illusion of a world without crime is a world filled with a very different kind of crime. Authoritarianism.”
Larry Ellison (Oracle) predicted a modern surveillance state in which citizens behave because they are always watched. Véliz reframes this: the world without visible crime is not safe — it is controlled. The cost is freedom. Compare algorithmic-fairness on how policing data encodes and amplifies societal bias.
What to do
“Uncertainty is good news. It means that the future is unwritten, that it’s ours to write.”
“It’s only when we acknowledge that we don’t know what the future holds and act accordingly that we can be sure to live in a free society.”
“Predictions can be weapons of power. But they only work if we believe them.”
The practical response is to resist treating predictions as facts — to choose privacy-respecting products, to demand contestable criteria over statistical pattern matching in high-stakes decisions, and to push for public debate about acceptable uses of predictive AI.
Hannah Arendt: it is pointless to argue with a murderer about whether their future victim is dead or alive; the only appropriate response is to rescue the person whose death is predicted.