British AI Startup Outshines Human Forecasters in Global Prediction Contest

Published: 20 September — The English Chronicle Desk | The English Chronicle Online

In a development that underscores the growing potential of artificial intelligence, a British startup has made waves in the global forecasting community after outperforming hundreds of human participants in an international competition. The system, developed by ManticAI, ranked among the top ten in the prestigious Metaculus Cup, a contest that challenges competitors to predict the outcomes of complex real-world events.

ManticAI, co-founded by Toby Shevlane, a former Google DeepMind researcher, secured the eighth position in the competition organised by the San Francisco-based forecasting platform Metaculus. Over the summer, participants were tasked with predicting the outcomes of sixty diverse events, including political scenarios such as whether Donald Trump would publicly clash with Elon Musk and whether Kemi Badenoch might be ousted from Conservative Party leadership.

Though still behind the world’s top human forecasters, ManticAI’s showing is seen as a milestone for artificial intelligence. Its strong performance suggests that AI systems are closing the gap with human experts far more quickly than previously thought. Deger Turan, chief executive of Metaculus, described the results as “impressive,” while noting that humans still generally hold the edge. He predicted that by 2029, AI may rival or even surpass the most skilled human forecasters.

The questions posed to competitors required predictions across diverse areas, from the outcome of Samoa’s general election to the extent of U.S. wildfires between January and August. Accuracy was measured on forecasts as of 1 September. Many professional human forecasters admitted that competing with AI was no longer straightforward. “It’s certainly a weird feeling to be outdone by several bots,” admitted Ben Shindel, a forecaster who initially trailed the AI but later finished above it. “Just a year ago, the best bot ranked around 300.”

ManticAI’s system works by dividing each forecasting problem into smaller tasks, handled by a suite of machine learning models including those developed by OpenAI, Google, and DeepSeek. This distributed approach enables the AI to evaluate scenarios, conduct historical research, simulate possible outcomes, and update predictions daily as new information emerges. Unlike humans, the AI can persistently reanalyse dozens of forecasts simultaneously without fatigue, a capability that many experts say is critical for accuracy.

Shevlane defended the originality of ManticAI’s predictions, arguing that its output demonstrates genuine reasoning rather than simple replication of past data. “Some say large language models just regurgitate their training data, but you can’t predict the future that way,” he said. “People often cluster around community averages, whereas our system frequently diverged. AI forecasters could be an antidote to groupthink.”

Despite these advances, experts agree that AI still faces limitations. Turan and others noted that AI struggles with logical consistency when predictions depend on complex interrelated events. Human forecasters also point to the continuing importance of intuition, though some, like Shindel, believe this capacity could eventually emerge in AI systems as well.

Leading human superforecasters, such as Philip Tetlock, co-author of Superforecasting, maintain that expert humans are still ahead overall. Tetlock’s recent research found that human specialists outperformed even the best bots in many scenarios. Warren Hatch, chief executive of Good Judgment, the company Tetlock co-founded, argued that the future lies in collaboration rather than competition. “AI will excel in areas with abundant structured data, like inflation forecasts, while humans will retain the advantage where sparse data and judgment are required. The real power comes from combining human and AI forecasting.”

Lubos Saloky, another competitor who finished third in the Metaculus Cup, offered a similar perspective. “I do not plan to retire. If you can’t beat them, merge with them,” he said, echoing the growing belief that hybrid forecasting could deliver the most reliable results.

The success of ManticAI highlights not only how rapidly artificial intelligence is evolving but also how it may transform decision-making in business, politics, and beyond. For now, human forecasters retain their lead, but the margin is shrinking — and the contest between human intuition and machine precision appears set to define the future of prediction.