AI outperforms humans in predicting future at international forecasting contest

A British artificial intelligence startup co-founded by a former Google DeepMind researcher has been ranked among the top 10 in an international forecasting competition, showcasing the growing capabilities of AI in predicting future events.

The startup, Mantic AI, secured eighth place in the Metaculus Cup, where participants were asked to forecast the likelihood of 60 events over the summer. These included high-profile occurrences such as the public disagreement between former US President Donald Trump and tech mogul Elon Musk, and the potential removal of Kemi Badenoch from the Conservative Party leadership, according to The Guardian.

While Mantic AI did not take the top spot, lagging slightly behind the leading human forecasters, experts say AI could rival or surpass human prediction abilities by 2029. Deger Turan, CEO of Metaculus, described the achievement as “impressive.”

Professional forecaster Ben Shindel, who ranked behind several AI systems, said, “It’s certainly a weird feeling to be outdone by bots at this point. We’ve come a long way compared to last year, when the best bot ranked around 30th.”

Before the competition, analysts had estimated that the top AI bot would achieve only 40 percent of the average score of leading human participants. However, Mantic AI exceeded expectations by scoring over 80 percent.

Mantic AI’s co-founder Toby Shevlane explained, “Some say large language models just regurgitate training data, but predicting the future requires genuine reasoning. Our system’s predictions were often more original than most human entrants, who tend to cluster around community averages.”

The startup uses a combination of machine-learning models, including OpenAI, Google, and DeepSeek, assigning tasks based on each model’s strengths.

AGI on the horizon?

As AI models grow increasingly accurate in forecasting, experts suggest they may be approaching Artificial General Intelligence (AGI), or human-level AI. A recent Google DeepMind research paper predicts AGI could emerge as early as 2030.

Unlike conventional AI, which is designed for specific tasks, AGI would possess the ability to understand, learn, and apply knowledge across a broad range of activities, mirroring human intelligence. In other words, AGI would be capable of reasoning and problem-solving in diverse domains, much like a human being.