Artificial intelligence failed to curb excitement: the results of the simulation from General Reasoning
Experts from General Reasoning conducted a large-scale study to find out whether modern neural networks are capable of becoming successful cappers. The 2023/24 English Premier League (EPL) season was chosen as a testing ground. The experiment "took part" 8 leading AI agents created by technology giants: OpenAI, Google, Anthropic, and xAI.
Experimental conditions
The methodology was as close to reality as possible. Each neural network was provided with:
- Historical background: Full statistics on teams, players and results of previous meetings.
- Starting capital: Virtual bank in the amount of $130,000.
- Freedom of action: The simulation was run three times for each model to eliminate the factor of random luck.
The algorithms worked like full-fledged players: they independently analysed the upcoming matches, compared their predictions with the odds of bookmakers, and determined the size of each bet.
The fiasco of digital minds
The results were disappointing for lovers of easy money with the help of technology. None of the eight models could close the season in the black.
The leader among the losers was the Claude neural network from Anthropic — its losses amounted to "only" 11% of the deposit. The rest of the participants showed much sadder dynamics, losing a total of about $1 million during the simulations.
Why does AI lose?
The head of General Reasoning commented on the failure as follows: AI does an excellent job of local, well-structured tasks. However, sports betting is a long-term process in the face of ever-changing variables. The real world turned out to be too chaotic and unstable for current models, who are not yet able to adapt to sudden injuries to players, refereeing errors and other unpredictable factors of a "live" game.
Olivia Grant is an iGaming news reporter at CasinosInCanada who covers casino industry updates, regulations, and market developments.
