Carissa Veliz on the Ethics of Prediction

Veliz argues predictions sound like factual claims about the future but often work as verdicts or orders, especially in the social domain. When a tech executive predicts everyone will use AI, he is not forecasting but trying to get you to act, buying into his vision.

Social predictions can become self-fulfilling even without the speaker intending it, as when telling a student they will fail affects their performance. Veliz calls this the perfect crime, since no error signal is produced: we never see the counterfactual world where the prediction was not made.

Veliz claims virtually all social predictions exert a self-fulfilling force, unlike predictions about the weather, but this force can still fail to overcome reality, as shown by the Fyre Festival and Theranos, where outlandish claims eventually crashed against unmet substance.

Veliz applies J.L. Austin's idea that some sentences do things rather than describe the world, comparing social predictions to performative utterances like marriage vows or ship christenings, since they implicitly prescribe action while appearing to describe fact.

Journalists report predictions from figures like Elon Musk or Bill Gates as though they were established facts, which grants those figures outsized power. Veliz compares this to historical deference to oracles, astrologers, and Rasputin.

When decisions about loans, jobs, or insurance rest on unfalsifiable predictions rather than clear criteria, the rejected person cannot contest the decision, unlike a factual criterion they could challenge. Veliz argues this erodes principles of merit and fairness and can create vicious cycles that validate biased predictions.

Veliz criticizes effective altruism's practice of projecting trillions of future people to outweigh present populations, arguing that once near-infinite quantities enter moral calculation, all sense of proportion is lost and almost any action can be justified.

Her interviewer suggests a tech executive's prediction could simply be a sincere true belief that happens to also benefit them financially, rather than a manipulative command. Veliz responds that even sincere predictions can become self-fulfilling and manipulative in effect regardless of intent.

Her interviewer pushes back that using probability to assess loan risk is simply sound business practice, comparable to insurance. Veliz counters that what works for an individual company can be unjust for society, since unfalsifiable predictions deny people due process and can create self-confirming cycles of disadvantage.