Prophecy

Carissa Véliz argues that predictions have always functioned as instruments of power rather than quests for truth, and that in the age of artificial intelligence, this dynamic has grown dangerously unchecked. The book spans thousands of years, from the Oracle of Delphi, an ancient Greek shrine where people sought prophecies, to modern algorithmic forecasting, building a case that prophecy shapes individual fates, undermines democracy, and demands urgent ethical scrutiny.

Véliz opens by establishing what she calls the wise view of prediction, contrasting it with the naive view that treats forecasts as neutral attempts at knowledge. She identifies five overlooked characteristics of predictions. They are guesses, not facts, because the future has not yet occurred. They are wishful, shaped by the desires of those who make them. They are exercises of power, closer to commands than descriptions. They are sometimes impossible, since the most consequential events tend to be unforeseeable. And they can be harmful, altering reality by changing people's expectations and behavior.

Part One traces the promises of prediction through history. Véliz recounts how the Roman emperor Tiberius tested astrologers by throwing incorrect ones off a cliff until one, the astrologer Thrasyllus, impressed him by demonstrating that predictions are about power, not knowledge. The Oracle of Delphi, she argues, was above all a commercial enterprise, its surrounding economy of games and taverns as profitable as the prophecies themselves. Prophecies consistently served political purposes: The Roman emperor Augustus published his horoscope to legitimize his rule, and cities from Rome to Baghdad were founded on prophetic grounds.

Véliz then traces how prediction shifted from qualitative to quantitative methods through three philosophical developments: the sociologist Max Weber's concept of the disenchantment of the world, in which rational explanations displaced magical ones; the idea of the universe as a clockwork mechanism; and a growing trust in numbers over people. She profiles key figures, from the polymath Girolamo Cardano, who wrote the first analysis of games of chance in the 16th century, to Thomas Bayes, whose work laid the groundwork for modern statistical inference, to the Belgian statistician Adolphe Quetelet, who applied the bell curve to human populations and turned statistics into a tool for judging citizens. Quantification, Véliz argues, killed individuals with names and idiosyncrasies, confidence in free will, and trust in human judgment. The success of mathematical predictions nurtured the belief that all uncertainty could eventually be conquered, driving the creation of AI.

AI is the culmination of this trajectory. Machine learning, in which algorithms analyze large datasets to identify patterns and predict outcomes, fulfills the fantasy that enough data and computing power could yield perfect prediction. Véliz characterizes this triumph as a corporate victory, achieved through mass surveillance and exploitation of vulnerable workers rather than scientific breakthroughs. Predictive algorithms now pervade daily life, from news feeds and navigation apps to hiring decisions and insurance pricing. One algorithm used on more than 200 million Americans halved the number of Black patients identified as needing urgent care because it used health-care spending as a proxy for actual health needs.

Drawing parallels between ancient court astrologers and modern tech executives, Véliz argues that figures like Elon Musk, Jeff Bezos, Peter Thiel, and Sam Altman wield more power than any previous government advisers. They are wealthier than the agencies tasked with regulating them and provide governments with infrastructure that is not easily replaced. After the September 11 attacks, the U.S. government funneled corporate surveillance data to intelligence agencies rather than restricting it, and social media became essential for political campaigns. At Donald Trump's presidential inauguration, tech billionaires were seated in front of the president's own cabinet.

Part Two addresses the perils of prediction. Véliz argues that predictions are deceptively similar to facts. Large language models, AI systems that generate text by predicting likely word sequences, produce output that is persuasive but detached from concern with truth. Invoking the philosopher Harry Frankfurt's concept of bullshit as speech designed to persuade rather than inform, she argues that these models are its ultimate practitioners. She cites chatbots that gave legally dangerous advice: Air Canada's chatbot provided incorrect bereavement policy information, costing the airline a lawsuit, and New York City's official chatbot advised citizens to break multiple laws. The fundamental problem is that machine learning tracks correlations but understands nothing about causes.

The chapter Véliz calls the book's core argument contends that predictions are commands disguised as descriptions: When a tech CEO predicts universal AI adoption, he is prescribing action, not describing the future. Self-fulfilling prophecies, predictions that cause their own realization by changing beliefs and behavior, operate everywhere: in finance, where panic selling creates the crash investors feared; in policing, where algorithms direct officers to already over-policed neighborhoods; and in mortgage approvals, where denial of credit confirms the original assessment. Véliz presents China's mass surveillance apparatus, with roughly 700 million cameras and a social credit system, as the extreme case of where predictive societies lead. Invoking the political theorist Hannah Arendt, she warns that when predictive accuracy about persons increases, it is because we are determining the future rather than discovering it.

Yet the world remains fundamentally unpredictable. Véliz classifies five sources of predictive trouble: bad or misleading data, social phenomena that defy standard models, unforeseeable scientific breakthroughs, historical flukes, and the paradox that overuse of prediction increases systemic risk by eliminating redundancy and destroying diversity.

Part Three asks how prediction should be rethought. Véliz critiques effective altruism, the philosophical movement popular among Silicon Valley elites, arguing that it depends on predictive abilities humans do not possess. Sam Bankman-Fried, the founder of the cryptocurrency exchange FTX and the movement's biggest donor, was convicted of massive fraud after secretly transferring customer deposits to fund bets. Véliz defends deontological ethics, which emphasizes duties and rights, and virtue ethics, which focuses on character, as approaches better suited to a world of irreducible uncertainty.

She argues that society should be designed to allow people to defy their predicted odds, citing the abolitionist Harriet Tubman's escape from slavery and Katalin Karikó's decades of marginalized research on mRNA, the messenger-molecule technology behind some COVID-19 vaccines, which led to a Nobel Prize. Preparation, she contends, is more valuable than prediction: Airplanes are built to withstand unpredictable events, not to predict each one. Creativity, humor, and democratic participation all depend on spaces of indeterminacy that prediction seeks to eliminate, since if destiny is written, there is no purpose in holding elections.

The final chapter draws on the ancient Greek philosopher Epicurus, who denied divination entirely and taught that freedom from fear and desire is the path to happiness. A prophet, Véliz concludes, has nothing to offer someone troubled by neither fear nor desire. An epilogue distills 10 heuristics for living amid uncertainty, including taking predictions as invitations for defiance and preparing rather than predicting. In a postscript, Véliz frames the book as a work of AI ethics, identifying surveillance and prediction as the original sins of digital technology. The good life, she argues, "is not a script to discover or follow, but a blank page to write on" (xvi).