AI Forecasting Startups Challenge Human Experts, Reshaping Markets and Policy

In early September, two artificial intelligence models did something that would have seemed impossible just years ago: they won a forecasting competition against the world's best human predictors. The Metaculus Cup, where champions of prediction—both flesh and algorithm—compete to estimate the likelihood of future events, crowned AI as the victor. It was a symbolic moment, but it pointed to something already underway in finance, policy, and risk assessment: machines trained on vast troves of public and private data are now making forecasts that match or exceed the estimates of superforecasters, those rare individuals whose predictions consistently outpace ordinary experts.

A handful of startups have built their entire business model on this premise. FutureSearch, Preseen, and Mantic are among the companies betting that AI can predict stock market swings, geopolitical ruptures like the reopening of the Strait of Hormuz, or even how a pop song will chart. Dan Schwarz, who cofounded FutureSearch, put it bluntly: by 2030, the decision-making methods we rely on today will look primitive. His own company employs Tom Liptay, a superforecaster of genuine renown. Schwarz acknowledged that Liptay understands his position is temporary. The machines will surpass him, and soon.

The money flowing into these systems tells its own story. Theo Summer and Veniamin Veselovsky, cofounders of Preseen, used AI forecasting tools to turn thirty-five dollars into one point nine million dollars in seven months on Kalshi, an American prediction market. The mechanism is elegant: rather than simply predicting yes or no, the models assign probabilities to outcomes and can cross-reference interconnected events—say, oil prices and Middle Eastern conflict—to refine their estimates. Hedge funds, which move capital at the speed of opportunity, have become the quickest adopters. If a fund believes an AI system can identify profitable trades, it will deploy capital accordingly.

What began as a niche application in finance is spreading. Toby Shevlane, cofounder of Mantic, noted that his company already has customers using AI forecasts to inform trading decisions. The impact is measurable, if still modest. But the trajectory is clear. Prediction markets like Kalshi and Polymarket, which millions of Americans use, are increasingly populated by AI systems making bets at speeds and with precision that retail investors cannot match. Veselovsky warned that ordinary traders may soon find these platforms unnavigable—not because the rules have changed, but because they will be dominated by algorithms that see patterns and probabilities invisible to human eyes.

The most sophisticated models have evolved beyond mere prediction. They now formulate their own questions about the future and assign probabilities to different outcomes without human prompting. This capability is opening doors beyond finance. Mantic has provided a major American corporation with scenarios of regulatory change in its sector and has consulted for the British government. Companies want to understand risk, to map the terrain of what might happen and prepare accordingly. Medical researchers are exploring whether these systems can help identify which treatments deserve funding. Insurance companies see applications in risk assessment. The technology is becoming infrastructure.

Yet infrastructure carries weight. Veselovsky acknowledged the upside: better forecasts could improve decisions across sectors, from policy to science. But he also named a concern that sits uneasily beneath the optimism. If Russia or China has developed forecasting systems superior to those available in the West, the asymmetry could reshape geopolitical advantage in ways we have not yet reckoned with. The question is no longer whether AI can predict the future. It is who gets to see those predictions first, and what they do with that knowledge.