The Limitations of AI: A Reality Check in Predicting Soccer Outcomes

In an era where artificial intelligence (AI) has made tremendous strides in various domains, a recent study highlights the significant gap between AI’s capabilities and its limitations when applied to real-world problems. The “KellyBench” report by General Reasoning, a London-based AI start-up, reveals that even the most advanced AI systems struggle to accurately predict soccer match outcomes over an extended period.

The study tested eight top AI systems from prominent organizations such as Google, OpenAI, and Anthropic in a virtual simulation of the 2023-24 Premier League season. The AI agents were provided with detailed historical data and statistics about each team and previous games, allowing them to build models that would maximize returns while managing risk. The AIs then placed bets on match outcomes and goal scores to test their adaptability to new events and updated player data as the season progressed.

The results were telling: none of the AI systems managed to turn a profit consistently across all three attempts. In fact, most AIs struggled significantly, with some even going bankrupt. Anthropic’s Claude Opus 4.6 fared relatively better, boasting an average loss of only 11 percent and nearly breaking even on one attempt. However, xAI’s Grok 4.20 was a notable underperformer, going bankrupt once and failing to complete the other two attempts. Google’s Gemini 3.1 Pro managed to turn a 34 percent profit on one occasion but ultimately went bankrupt.

This study serves as a sobering reminder of AI’s limitations when applied to complex real-world problems. While AI has made significant strides in areas such as software development and language processing, its ability to accurately predict outcomes over long periods remains uncertain. The findings also highlight the importance of domain knowledge and understanding human behavior in AI applications.

Moreover, this study underscores the need for further research into the biases and limitations of AI systems, particularly when applied to domains with complex dynamics, such as sports betting. As AI continues to transform various industries, it is essential that we acknowledge its limitations and strive to develop more robust and accurate AI models that can effectively adapt to changing circumstances.

Source: https://arstechnica.com/ai/2026/04/ai-models-are-terrible-at-betting-on-soccer-especially-xai-grok/