As soccer fans everywhere wait anxiously for the next round of World Cup matches, experts, gamblers, artificial intelligence (AI) agents, and even animals are weighing in on who will win each match.
These predictions have become big business. Just seven years since the US federal ban against sports betting was struck down in 2018, the American Gaming Association announced that Americans bet $16.89 billion on sports in 2025. And that’s just the legal figure. But not everyone is interested in the commercial aspect of sports predictions.
In a new project called Forecast 2026, researchers at Trinity College Dublin are pitting experts and AI agents against the crowdsourced knowledge of soccer fans. The study isn’t aiming to craft the ultimate big-winning prediction bot, said study lead Konstantinos Tsetsos, a cognitive and decision sciences researcher at Trinity College Dublin. Instead, the team hopes it can help answer a simple question about how people make decisions.
“Do the decision-making principles we uncover in the lab generalize to a real-world setting where people care about the outcome, have strong prior beliefs, and receive external information and feedback as events unfold?” Tsetsos wrote in an email. He added that the World Cup was a strong “natural laboratory” to test this question. Each of the 104 matches in this year’s edition has a clear outcome, and the huge interest in the competition means there’s no shortage of armchair experts keen to get involved.
Anticipating the Bounce of the Ball
Sports predictions are normally made through the prism of betting, which is a highly emotionally charged activity. “We wanted to remove the money, remove the gambling frame, and give people a fairer way to show what they know,” said Tsetsos.
Rather than asking participants to predict a winner or loser, Forecast 2026 asks them to assign probabilities to a win, a loss, or a draw. “This can help us separate forecasting skill from lucky guesses,” Tsetsos said.
Another motivation for the study, he added, was the rise of generative AI. Humans are now regularly using AI to augment their decision-making, but researchers are still learning about how AI agents make choices. The new project pits humans against language-based AI agents, like ChatGPT or Claude, and conventional statistical forecasting systems, which use team- and match-related statistics to make numerical predictions.
The Wisdom of the Crowd?
The project has now amassed a substantial chunk of data on how these three groups—humans, AI agents, and conventional forecasting systems—weigh up against the mighty power of the gambling industry. “Betting markets use sophisticated predictive models, aggregating team strength, historical data, team news, injuries, lineups, professional analysis, and betting patterns. Consistently beating that benchmark is difficult, if not impossible,” said Tsetsos.
Forecast 2026 assigns each human or AI predictor a point score based on how closely their correct or incorrect guesses match the official betting odds. Currently, AIs have a small advantage over the human crowd. Against the betting baseline, AIs are averaging a -158-point score, while the humans’ average is down at -725 points. The difference is closer between AIs and humans who have predicted more than half of the matches played. Traditional betting models are far behind that specific group at -2,322 points.
Classical psychology studies would suggest that as more people make predictions, those predictions would become more accurate. This is known as the “wisdom of the crowd” effect. But Forecast 2026 has found that both human and AI predictions have become less accurate as the tournament goes on. “The market baseline has become sharper and harder to beat,” said Tsetsos.
When Big, Bold Calls Pay Off
The research team hopes that the remaining matches in the tournament will provide more insight into how humans and AI agents learn from previous results. This improvement might be borne out by insightful, but contrarian, predictions becoming more common. Qatar’s draw with Switzerland in the first round of matches was an outcome that both humans and AI agents said was far more likely than the betting market did. On the other hand, the human crowd wildly overpredicted the likelihood that plucky Caribbean minnows Curacao would not be beaten by three-time Africa Cup of Nations champions Ivory Coast.
One of the biggest surprises from the study so far, said Tsetsos, is that humans’ and AI’s boldest calls, like the Qatar-Switzerland draw, often mirror each other. “One possibility is that the market was simply off, making both humans and AI look good. Another is that humans and AI were picking up on something that was not fully captured by the statistical models feeding into the market,” said Tsetsos. As of today (July 2), a human calling themselves CopacabanaKickaboutFan was top of the charts, with a +579 score.
CopacabanaKickaboutFan remains anonymous, but the authors have collected information about each of their participants' numerical ability, cognitive style, and soccer knowledge. “At the end of the study we hope to understand what kind of person forecasts like this,” said Tsetsos. “For now, they are a useful reminder that humans should not be written off.”
Users can make predictions right up until the tournament ends on July 19.
