A stunning performance by a British AI in a global forecasting competition has added fuel to predictions that machines could be better than humans at predicting the future by the end of this decade. ManticAI, a startup with ties to Google DeepMind, placed eighth in the Metaculus Cup, leading the competition’s CEO to estimate that AI will be on par with or better than the best human forecasters by 2029.
The summer-long contest involved predicting the outcomes of 60 wide-ranging events, from Samoan elections to spats between billionaires. ManticAI’s ability to navigate this complex and uncertain landscape, outperforming many human specialists, underscores the accelerated development of AI’s reasoning faculties.
The AI’s success is attributed to its unique multi-agent system, which deploys a roster of machine-learning models from tech giants like OpenAI and Google. Each model is assigned a role suited to its strengths, such as historical analysis or scenario simulation. This allows the system to build a comprehensive and constantly updated forecast, a process that is incredibly labor-intensive for human teams.
The rapid progress is stark. A year ago, the best-performing AI was ranked around 300th. Today, multiple AIs are in the top tier. This dramatic improvement has caught the attention of the entire forecasting community, prompting a re-evaluation of AI’s potential role in fields ranging from finance to national security.
While experts caution that AI still struggles with certain types of complex forecasts that require deep, multi-step logical verification, the trend is clear. The future of forecasting is poised to become a collaborative effort. As forecasting company CEO Warren Hatch noted, the goal is to combine human judgment with AI’s analytical power to get the “best forecast possible as quickly as possible,” preparing for a world where AI-driven insight is indispensable.