Self-certified Geniuses

These days, labeling—sorry, branding—yourself is increasing in popularity, especially as claimed expertise starts paying better than demonstrated expertise. I should declare an interest: I am a professional futurist, which is a title nobody issued me and nobody can take away. Some people come to it via the educational route, studying foresight. Some, like me, are given the label by working professionally in foresight—street accreditation. Either way there is no exam, no central register and nothing to revoke. I've been working under this banner for over twenty years. A banner is all it is.

The same arrangement is now running in prediction markets, with a better word: superforecaster. This one sounds like a discipline rather than an ambition, and initially it was earned through monitoring and results. From 2011, the US intelligence research agency IARPA ran open forecasting tournaments, having participants forecast real outcomes with fixed resolution dates that could be checked. Philip Tetlock’s team of amateurs, most of them working from home, beat rival research groups and the professional analysts who had access to classified material they didn’t. The word meant something because there was always a resolution date, after which you were demonstrably right or wrong.

But who needs rigor when you can just appropriate the mantle, and upgrade your own users? Tarek Mansour, who runs the prediction market Kalshi, was asked by the New York Times in August how concentrated his trading volume was. Referring to his biggest whales, Mansour was generous. “They are super-forecasters,” he told the Times. “Those guys are not like 2 percent of volume; they’re 70 to 80 percent of volume.” He means his heaviest users. The word was appropriated from a research program backed by a methodology and granted to whoever spent the most on short-term bets, without the measurement that was the whole reason the label existed.

And the questions they bet on (sorry, predict) keep getting smaller and stupider, because they have to keep the system running without enough bets to go around. A market needs volume, volume needs contracts, and the world doesn’t supply enough questions within the grasp of most participants to generate that volume: elections come around every couple of years, wars start rarely and end slowly, much else is outside their knowledge.

So the focus drifts toward whatever can be settled quickly and listed in quantity: a syllable muttered in an online tournament, a word in a press conference, a margin in a game. Congrats, you’ve re-invented sports betting, with a dash of financial market trading. (In fact, sports were 95 percent of Kalshi’s business last year, and on Mansour’s own account are still around two-thirds. In a consumer-finance survey this year, 27 percent of American users said they were there for entertainment and 53 percent for income.)

Bayeswatch

The migrating prediction jargon did not arrive with the prediction markets, of course. Rationalist and effective-altruist forums spent years using Bayesian terms as an identity marker rather than as a meaningful practice—updating, priors, posteriors, calibration—and it spread through poker boards, startup Slacks and AI labs into general Valley conversation long before anyone was trading noises and phrases.

The hottest bit of rationalist jargon at the moment is p(doom), a person’s stated probability that artificial intelligence wipes out humanity, given as a number between zero and one. Max Read, writing in the New York Times Magazine recently, cataloged how far this linguistic showmanship has spread: people “updating their priors” on Thai food, calling a decision “+EV,” exchanging superintelligence estimates over IPAs.

Bullshit and math

Attaching a decimal to a guess adds nothing to the validity of the guess in reality. It’s still a guess, whether professionally educated or homeschooled. It does flag something about the guesser, however. It displays a skill at upcycling a hunch into a model, a loss into a data point, a compulsion into a practice, and a person with a smartphone into someone with a sharply-honed intellect. They have granted themselves license to predict, and have it called forecasting.

Niamh Rowe spent months reporting from inside Alpha, an invitation-only group of about thirty of Polymarket’s most successful traders, and one of them shows what the license actually comes with. This trader left Oxford once his side hustle put him three-quarters of a million dollars up. His method was not modeling: most of his profit came from stale orders other people had forgotten to cancel, and his deepest research was checking whether the counterparty had Trump memecoins in their wallet. “There’s free money on the floor,” he told her. “But you have to be there. And you have to be very fast at clicking.” On what the work did to him: “I don’t have thoughts any more. I can’t write anything.” He's also not forecasting anything beyond who is likely to have left their own bets unmanaged.

This license works the way the founder’s license does — self-bestowed, and answerable to nobody. And like many founders, these prediction market “superforecasters” are reasonably likely to be serial losers. In June, four finance academics at ESSEC, HEC Montréal and Toronto went through $67 billion of Polymarket trading—588 million individual trades—and published the arithmetic. Among the accounts that finished in profit, the top 1 percent took 76.5 percent of all the money made. The winners got it by posting limit orders, which are standing offers left in the order book for someone else to take; the losers got it taken off them by accepting those offers at market price. On whether any of this is skill, the authors won’t go further than calling performance “modestly persistent, though this may reflect selection rather than skill.” Read into that what you will.

The cost is not mainly to the losers, though some do rack up heavy losses in the quest to scratch the itch of anticipating what particularly shallow futures hold. The real problem is a generation that now believes pricing the likelihood of something is the same as understanding the underlying dynamics enough to really forecast it.