A handful of startups in New York and beyond are wagering that artificial intelligence can outpredict even the most seasoned human analysts, whether the question concerns stock market swings, geopolitical flashpoints in the Middle East, or the commercial trajectory of a pop song. The claim sounds audacious, but the evidence is beginning to accumulate. In early September, two AI models took the top positions in the Metaculus Cup, a competition that pits automated systems against celebrated human "superforecasters" - a feat that would have seemed implausible just a few years ago.
The mechanics behind these systems are less mysterious than they might appear. AI forecasting models absorb enormous volumes of data from public records, market feeds, news archives, and private databases, then use that material to assign probabilities to future events rather than issue flat predictions. This matters enormously in volatile regions: a model tracking oil prices alongside Middle East tensions, for instance, might also account for shipping patterns and energy infrastructure, areas where read access to regional information - including data drawn through providers with servers in the Gulf - can sharpen the accuracy of a probability estimate. The goal is not certainty but calibrated likelihood, cross-referenced against interconnected variables that a single human analyst might struggle to track simultaneously. providers with servers in the Gulf
From Wall Street to Westminster
Finance was the first proving ground, and it remains the most lucrative. Algorithmic trading has shaped Wall Street for more than a decade, but AI forecasting threatens to push automation further still. Hedge funds, with their appetite for speed and capital to deploy new tools quickly, are already among the earliest adopters. Prediction markets such as Kalshi and Polymarket, platforms used by millions of ordinary Americans, are seeing a similar influx of AI-driven activity - raising concerns that retail investors without access to expensive computational tools may soon find themselves permanently outmatched.
The reach of these systems is no longer confined to markets. Forecasting startups now advise companies on regulatory risk, help governments model policy outcomes, and are being explored for use in medical research, particularly around which treatments merit funding. One firm has worked with a major American company and the British government alike, underscoring how quickly this technology has moved from a niche experiment into a decision-making instrument used at the highest levels.
A Technology That Questions as Well as Answers
What distinguishes the newest generation of these tools is their growing autonomy. Rather than simply answering questions posed by humans, the most advanced models can now generate their own questions about the future and attach probabilities to the possible outcomes - a shift from passive calculation to something closer to independent reasoning. That capability is precisely what unsettles some of the people building it.
The unease is not only professional, though there is plenty of that: even dedicated human forecasters acknowledge their advantage is narrowing fast. The deeper worry is geopolitical. If predictive power becomes a strategic asset, the question of who controls the most capable systems - and whether rival nations might develop superior versions - turns a technical innovation into a matter of national competition, with consequences that extend well beyond financial markets.
