Sports matches, horse races, beauty contests… New technologies make it possible to analyze unprecedented amounts of data and different parameters.
Manchester City, big winner of the Champions League in June? It is not a sports forecaster who says it, but the artificial intelligence of Opta Analyst. To anticipate the outcome of a competition, Paul the octopus and other medium animals are relegated to the closet. More streamlined software based on artificial intelligence (AI) replaced them with impressive results. As early as November, the artificial intelligence of EA Sports, the publisher of the famous football game Fifa, had predicted the success of Argentina in Qatar in 2022. One day before the victory of Albiceleste, Indira Ampiot, Miss Guadeloupe , was crowned Miss France 2023. Nothing surprising for the consulting firm Avisia, which had announced its coronation using software.
A multitude of data to take into account
“AI aggregates statistical data to establish possible scenarios”, details Anis Ayari, artificial intelligence engineer and popularizer on Twitch and YouTube. Player performance, match histories, coaches, ball possession rate, etc. Sports statistics are used to feed predictions. Many extra-sporting data can also be taken into account, such as the weather, the quality of the pitch, the teams’ budget, or even the sporting culture of the country. “In these models, Argentina and Brazil are well rated because football has become a religion. On the contrary, the United States is poorly ranked. This can be explained by the choice to build a baseball stadium instead of a football field”notes Virginie Mathivet, director of the modern data department at TeamWork.
Some models use subjective data. In 2016, Microsoft analyzed the opinion of fans around the 2018 World Cup. France came out on top with a 76% chance. Same observation for Miss France. Avisia’s machine learning tool uses the theory of the wisdom of crowds. “AI analyzes the feelings and opinions expressed by Internet users on social networks to identify voting trends”underlines Marie-Alice Blete, data architect.
AI and survivor bias
Fifa announced the last four winners of the World Cup: Spain in 2010, Germany in 2014, France in 2018 and Argentina a few days ago. For 16 years, EA sport’s artificial intelligence has been flawless. Or almost. One detail, and not the least: most of the intermediate predictions are wrong. The IA announced a semi-final between France and Argentina, lost 1-0 by the Blues (France beat Morocco 2 goals to 0 in the semi-final). In its prediction, Brazil successively eliminated Belgium, Germany, then Portugal on penalties to reach the World Cup final. Final where she bowed to the sworn enemy Argentina, on the smallest of margins (0-1). This raises questions about the reliability of the predictions of these new tools. “It’s called survivorship bias. Exceptions are set up as models of success. And we forget the 99% of AI who predicted a bad result”, nuance Anis Ayari, engineer in artificial intelligence.
Depending on the elements analyzed and the technology chosen, the forecasts can vary completely. Opta Analystthe most elaborate artificial intelligence on the market saw Brazil as the winner. Scientists from the University of Innsbruck (Austria) had also predicted a Brazilian victory while the Oxford AI announced, at the start of the competition, a victory for Belgium.
Chance, a determining factor
Fallible prognoses in more than one way since AIs do not integrate a determining element of competitions: chance. An injured player, a sick Miss or a pandemic are all factors that an algorithm, no matter how well trained, cannot predict.“An artificial intelligence is as effective as a football or horse racing expert”slice Virginie Mathivet, director of the modern data department at TeamWork.
Especially since sport is far from being an exact science. No machine or expert had anticipated the course of Morocco, which reached the semi-finals of the 2022 World Cup. Or the victory of Saudi Arabia against Messi’s team. “That’s the whole point of sport or a competition, continues the doctor in artificial intelligence. If a machine can determine human behavior in advance, then why participate? » Just take a look at the Blues’ final in Qatar. Until the 79th minute, all the AIs and experts gave France no chance. In the 82nd, the France team turned the tide… and the forecasts changed.