A British artificial intelligence startup has beaten dozens of human rivals in an international forecasting competition, raising questions about whether machines are now closing in on one of the most human of skills — predicting the future.
ManticAI, founded by former Google DeepMind researcher Toby Shevlane, secured eighth place in the Metaculus Cup, a summer-long contest that asked entrants to forecast the likelihood of 60 political, economic and social events.
AI edges closer to human forecasters
The competition, run by US-based Metaculus, attracts professional “superforecasters” as well as enthusiasts. They were asked to estimate outcomes ranging from who would win Samoa’s general election to how many acres of US land would be destroyed by wildfires.
Until recently, AI systems struggled to get close to the top. Last year, the best-performing bot was ranked in the 300s. Mantic’s finish inside the top 10 has been described by observers as a breakthrough.
“It’s a strange feeling to be beaten by bots,” said Ben Shindel, a professional forecaster who competed in the event. “It shows just how far the technology has come in only a year.”
How the system works
Mantic doesn’t rely on a single model. Instead, it spreads forecasting tasks across different AI systems, including those built by OpenAI, Google and DeepSeek. One model might dig into historical patterns, another might simulate scenarios, while others monitor live data. Their results are then pooled into a single prediction.
“Some say large language models only repeat what they’ve seen before,” Shevlane said. “But you can’t forecast the future by parroting. Our system had to reason. Often, it deliberately disagreed with the crowd, which shows AI can cut through the herd instinct humans sometimes fall into.”
AI also has stamina on its side. While human forecasters may revisit a question every few days, the machine can refresh dozens of predictions daily, adjusting probabilities as events shift.
Humans still hold the edge — for now
Despite Mantic’s strong showing, experts say humans remain ahead in complex, judgment-heavy areas.
Philip Tetlock, co-author of Superforecasting and a pioneer of the field, recently published research showing human experts still outperform AI on nuanced questions.
Metaculus chief executive Deger Turan agreed that, for now, the best forecasters are people. But he believes the gap will close quickly: “AI could match or beat top humans by 2029.”
Some argue the future lies in partnership. “AI is likely to excel at areas with lots of data, such as inflation figures,” said Warren Hatch, head of forecasting firm Good Judgment. “But in questions with sparse data, human judgment remains crucial. The best results will come from combining both.”
Forecasting’s future
For Mantic, the competition marks an important milestone. “This shows AI can do more than crunch numbers — it can reason, evaluate and predict,” Shevlane said.
Not everyone is ready to hand over the reins. Lubos Saloky, a human forecaster who finished third, put it simply: “I don’t plan to retire. If you can’t beat them, merge with them.”
With AI climbing the leaderboard and humans still holding the edge on judgment, the future of forecasting may well be hybrid.
