An artificial intelligence from a UK startup has proven its mettle in the art of prediction, securing a top-ten finish in a global forecasting tournament and beating a host of human experts. ManticAI’s eighth-place rank in the Metaculus Cup is being hailed as a milestone, signaling that AI’s ability to reason about future events is advancing at a breathtaking pace.
The contest was a rigorous test of foresight, involving 60 questions on topics ranging from international politics to environmental disasters. The goal was to accurately predict the state of affairs on September 1st. ManticAI’s performance shows that AI is moving beyond simple data processing and into the realm of sophisticated, probability-based reasoning about the unknown.
At the heart of ManticAI’s success is a system of coordinated AI agents. The startup, which has ties to Google’s DeepMind, uses a variety of large language models, each assigned to a different part of the forecasting process. One agent might assess the current situation, another might research historical precedents, while a third simulates potential outcomes. This division of labor allows for a relentless and thorough analysis that is difficult for humans to match.
The rapid improvement has not gone unnoticed. Professional forecaster Ben Shindel, who was outranked by the AI for part of the contest, remarked on the incredible progress made in just a single year. This swift evolution from a low-ranked participant to a top contender has led many to reconsider their timelines for when AI might surpass human predictive abilities entirely.
While the world’s best human forecasters still maintain a slight edge, especially on complex, interrelated predictions where AI can struggle with logical consistency, the gap is closing. The CEO of Metaculus, Deger Turan, has put a date on it, estimating AI could be on par with or better than humans by 2029. The ultimate future, however, points towards a powerful partnership between human judgment and AI’s analytical prowess.