Is your Euro 2012 fantasy football team sliding down the league at an alarming rate? If you thought Chris from the office was hard to beat, watch out – soon you might be competing against a computer.
Sarvapali Ramchurn, a computer scientist at the University of Southampton, UK, and colleagues have developed an artificial soccer manager that ranks in the top one per cent of the 2.5 million players on the official English Premier League fantasy football game.
The AI manager evaluates players’ previous performance statistics to predict how many points they will receive during the forthcoming week. It then picks a squad that will maximise the expected fantasy football score while adhering to the rules of the game, such as how much it costs to buy or exchange a player.
Humans do the same, of course, but the AI manager goes into more depth than even the most dedicated fantasy football player. “It cares about every single decimal point,” says Ramchurn. The AI also plans weeks ahead, so it can take into account the impact its decisions will have in the future. “Just like [Garry] Kasparov playing chess against Deep Blue, it was about the machine being able to look ahead many more moves than the human,” he adds.
Ramchurn tested the AI manager offline during the 2010/11 season and it achieved a rank of just over 26,000, putting it ahead of 98.9 per cent of all other players, and at one point managed to finish in the top 500 players. The system will compete online in this year’s season and Ramchurn will add human input to test whether this will further boost the score – for example, unlike its human counterparts, the system is currently unable to avoid picking players who are injured or who have had a row with their manager.
The work, which was led by Tim Matthews at the University of Southampton, will be presented at the Association for the Advancement of Artificial Intelligence conference in Toronto, Canada, next month. The team will also explain why the work was carried out in the first place.
Matthews and colleagues are researching methods for selecting the best emergency response teams for disasters like earthquakes, and realised that the data-rich world of fantasy football provided the perfect testbed for their theories. Winning the office league is just a happy by-product.