From a cluster of New York startups, artificial intelligence systems have begun to do what only a rare class of human minds could once do reliably — see around corners. In September, AI models claimed the top prizes at the Metaculus Cup, outperforming the superforecasters who represent the apex of human predictive skill. The moment signals something larger than a competition result: the tools by which societies, markets, and governments understand the future are quietly changing hands, with consequences that reach from trading floors to policy chambers to the unspoken rivalries between nations
AI Forecasting Startups Challenge Human Experts, Reshaping Finance and Risk
They will soon just be dominated by AI forecasting systems
Mark
So these AI systems are actually winning prediction competitions against humans who've trained for years. How is that even possible?
Mimi
They're working with vastly more data than any individual could process. They can pull from public sources, private datasets, historical patterns. A human superforecaster is brilliant, but they're still one brain. The AI is synthesizing information at a scale that's just not human-compatible.
Luke
But we should be careful here—the article says they're "on par with" superforecasters, and then mentions two AI models won one competition. That's not the same as proving they're systematically better across all domains. One tournament result is meaningful but not conclusive.
Mark
Fair point. So if they're matching human experts, why are hedge funds moving so fast to adopt them?
Mimi
Because even a small edge in prediction translates directly to money. If you can forecast market movements or geopolitical events slightly better than your competitors, you trade faster and more profitably. Theo Summer turned $35 into $1.9 million in seven months using these tools. That's not a small edge.
Luke
Though we should note that's one person's result on one platform over one seven-month period. It's a striking data point, but it's not a controlled study. Markets can be volatile. We don't know what his loss rate was on other bets, or whether that result is repeatable.
Mark
What worries me more is what happens to regular people trying to trade on these platforms.
Mimi
They're going to be outgunned. Kalshi and Polymarket are already filling up with AI traders. If you're a retail investor without access to expensive forecasting systems, you're essentially competing against machines that process information faster and more comprehensively than you ever could.
Luke
The article quotes Veselovsky saying retail investors will "soon just be dominated by AI forecasting systems." That's a prediction about the future, not a current fact. It might happen, but we don't have data yet on how quickly that displacement will occur or how complete it will be.
Mark
And then there's the government and policy angle. Mantic is working with the British government. What does that even mean?
Mimi
It means governments are starting to use these tools to model regulatory scenarios, anticipate risks, understand second and third-order effects of policy decisions. It's risk management at a national scale.
Luke
But the article doesn't say what the British government actually used it for, or whether it changed any decisions. We know Mantic has worked with them, but the specifics of that engagement and its impact are not in the reporting. That's a gap worth naming.
Mark
And the final thing—Veselovsky's comment about Russia or China potentially having better systems. That feels like the real story underneath.
Mimi
It is. If forecasting AI becomes genuinely predictive—if it can see around corners better than human analysis—then whoever builds the best version has a strategic advantage that's hard to quantify and hard to defend against. It's not just about money anymore.
Luke
But that's also speculation. We don't know if Russia or China are building these systems, or whether they're better or worse. Veselovsky expressed a concern, which is worth reporting, but it's not the same as evidence of a geopolitical competition in forecasting AI. That's a story that hasn't been written yet.
Le Pouls
- AI models have crossed a symbolic threshold by defeating elite human superforecasters in open competition, signaling that machine prediction is no longer experimental — it is operational.
- One founding team turned $35 into $1.9 million in seven months using AI-driven probability trades, demonstrating that the financial edge these tools provide is already concrete and compounding.
- Retail investors on platforms like Kalshi and Polymarket are being quietly crowded out as AI systems flood prediction markets, tilting a once-open playing field toward those who can afford the technology.
- The tools are spreading beyond finance into government risk planning, medical research funding, insurance assessment, and corporate regulatory strategy — reshaping how institutions make consequential decisions.
- A geopolitical undercurrent is emerging: if advanced forecasting AI confers strategic advantage, the question of whether rival nations already possess superior systems has become impossible to ignore.
From a cluster of New York startups, artificial intelligence systems have begun to do what only a rare class of human minds could once do reliably — see around corners. In September, AI models claimed the top prizes at the Metaculus Cup, outperforming the superforecasters who represent the apex of human predictive skill. The moment signals something larger than a competition result: the tools by which societies, markets, and governments understand the future are quietly changing hands, with consequences that reach from trading floors to policy chambers to the unspoken rivalries between nations.
A handful of New York startups are building AI systems capable of forecasting stock movements, geopolitical shifts, and cultural trends — wagering that machines trained on vast data can outperform the humans who have spent careers mastering prediction.
The wager gained credibility in September, when two AI models claimed the top prizes at the Metaculus Cup, a competition that pits algorithmic and human forecasters against one another. The result was a threshold moment. Dan Schwarz, co-founder of FutureSearch, believes the methods societies rely on today will look primitive by 2030 — and even his own star superforecaster, Tom Liptay, understands that his position is not secure over the medium term.
The commercial proof is already striking. Preseen founders Theo Summer and Veniamin Veselovsky used their AI tools to grow $35 into $1.9 million in seven months on the prediction market Kalshi, by assigning probabilities to interconnected future events — oil prices, regional conflicts, cascading outcomes. Hedge funds have moved fastest to adopt these capabilities, and Mantic co-founder Toby Shevlane confirmed that clients are already using AI forecasts to drive trading decisions.
For ordinary investors, the trajectory is sobering. Prediction markets once open to millions are increasingly dominated by AI-driven trades, and Veselovsky warned that retail participants without access to expensive tools will soon find themselves effectively priced out.
The ambitions of these systems reach further still. The most advanced models can now generate their own questions about the future and model branching probabilities across scenarios. Mantic has delivered regulatory forecasts to a major corporation and worked alongside the British government. Medical researchers, insurers, and public agencies are all exploring applications.
Yet the power of these tools carries a quiet anxiety. Veselovsky raised the possibility that Russia or China may already be developing comparable or superior systems. If AI forecasting becomes a decisive advantage in markets, strategy, and governance, the nations and institutions that build the most capable versions will hold an edge that is difficult to see — and harder still to answer.
A small group of startups in New York are building artificial intelligence systems that can forecast everything from stock market swings to geopolitical events to how a pop song will chart. They are betting that machines trained on vast datasets can outthink the humans who have spent careers becoming expert at prediction.
In September, two AI models won the top prizes at the Metaculus Cup, a competition that pits prediction champions—both human and algorithmic—against each other. The victory marked a threshold moment: the machines had matched or exceeded the performance of superforecasters, those rare individuals whose estimates consistently beat ordinary expert opinion. Dan Schwarz, co-founder of FutureSearch, one of the startups building these systems, said the shift will be profound. By 2030, he predicted, the forecasting methods we rely on today will seem primitive. Even FutureSearch's own superforecaster, Tom Liptay, understands the trajectory. Schwarz noted that Liptay knows "he's not secure in his job" over the medium term—that these systems will eventually surpass him.
The business case is already compelling. Theo Summer and Veniamin Veselovsky, who founded Preseen, used their AI forecasting tools to turn $35 into $1.9 million in seven months on Kalshi, an American prediction market. The approach works by assigning probabilities to future events and cross-checking assumptions across interconnected scenarios—how oil prices might move, for instance, in relation to Middle East conflict. Hedge funds have become the quickest adopters. Toby Shevlane, co-founder of Mantic, another forecasting startup, said his company already has customers using AI predictions to make trading decisions. The impact on financial markets is likely to accelerate. Algorithms have shaped Wall Street for over a decade; AI forecasting will push that automation further.
The consequence for ordinary investors is becoming clear. Kalshi and Polymarket, prediction markets used by millions of Americans, are increasingly flooded with AI-driven trades. Veselovsky warned that retail investors without access to expensive AI tools will soon find themselves priced out. "They will soon just be dominated by AI forecasting systems," he said. The playing field, in other words, is tilting sharply.
But the reach of these tools extends well beyond trading floors. The most sophisticated models now do more than predict; they can formulate their own questions about the future and assign probabilities to different outcomes. Mantic has provided a large American corporation with scenarios of regulatory developments in its sector and has worked with the British government. Companies are hungry for this capability—not to make money, but to understand and monitor risk. Schwarz noted that medical researchers are interested in using forecasting tools to decide which treatments deserve funding. Insurance companies are exploring applications in risk assessment. Government agencies are considering how these systems might inform policy decisions.
Yet the very power of these tools raises an unspoken anxiety. Veselovsky acknowledged the potential for good but also expressed concern about the geopolitical dimension. "Maybe Russia or China has better systems," he said. The implication hung in the air: if forecasting AI becomes a decisive advantage in markets, policy, and strategic planning, then the nations or firms that build the most capable versions will hold an edge that is difficult to measure and harder still to counter. The question of who controls these systems, and what they can see about the future, has begun to matter.
Citations marquantes
By 2030, what we were doing for our decision-making will seem like complete Stone Age thinking— Dan Schwarz, co-founder of FutureSearch
Retail investors without access to expensive AI tools will soon find prediction markets dominated by AI forecasting systems— Veniamin Veselovsky, co-founder of Preseen
Envie de l'histoire complète ?
Lire l'original sur The Straits Times ↗
Politics & Governance