Once upon a time, a cephalopod in Oberhausen, Germany named Paul the Octopus captured human hearts by picking football match winners at the Euro 2008 and 2010 FIFA World Cup with a success rate of 85.7 percent.
Now, AI is following in Paul’s tentacle steps, taking the challenge to the FIFA World Cup 2018 in Russia.
A number of global AI research centers have run machine learning models to predict outcomes for the multi-stage competition, and they have come up with entirely different results. Synced looks at a selection and how the picks were made:
German University Researchers: Spain
A research team from the German Technische University of Dortmund and the Technical University in Munich did not let bias affect their model’s prediction for 2018 World Cup Winner. In association with the Ghent University in Belgium, the team trained an AI model that predicted Spain will win the championship.
Researchers used a popular statistical method for various machine learning tasks, Random Decision Trees. They trained a model with data such as the FIFA rankings, whether national team players are club teammates, players’ average age, how many Champions Leagues they’ve won, etc. Even each country’s population and gross domestic product (GDP) were included in the exhaustive dataset.
The model ran 100,000 tournament simulations, calculating each team’s probabilities to advance in each round. Spain emerged with a slight advantage (17.8 percent) over Germany (17.1 percent) and Brazil (12.3 percent).
However, even AI hadn’t imagined the sudden firing of national team coach Julen Lopetegui just days before the start of the World Cup, a move many observers believe could hurt Spain.
Goldman Sachs: Brazil
Every four years Goldman Sachs economists publish a “World Cup and Economic Report,” predicting World Cup winners. This time the investment giant generated a model that ran through over a million tournament simulations and concluded that Brazil will win the World Cup with a win over Germany.
Goldman Sachs’ Random Decision Trees had fewer data factors, focusing on team rating, player rating, historical team and opponent performance, team’s current scoring and winning momentum, etc.
Electronic Arts: France
Electronic Arts (EA), developer of the popular FIFA series football simulation video game, has a remarkable record in predicting professional sports contests. The company bet on Germany for 2014 World Cup, and has predicted American football’s Super Bowl winner nine times out of thirteen.
This time EA says France and Germany will go to the final, and the Gallic Rooster will prevail 5-4 in a penalty shootout.
EA ran its simulations using new ratings for its video game FIFA 18. While the EA game engine is not technically AI, it is backed by numerous machine learning techniques designed to make player performance as realistic as possible.
Russian Researcher: Germany
Russian news media reports that a fourth-year student of the Perm State National Research University created a neural network to identify a team’s relative character and chemistry based on the final player selections, trained on a complex combination of data including each player’s cost, age of the coach, and injuries on the team.
The model identified Germany, Brazil and Argentina as the best built squads, and suggested Germany will take the trophy. The model ranked host Russia’s chances of winning the championship 27th out of 32 teams.
Meanwhile, an AI-augmented human expert based Swarm organized by Unanimous.ai, a San Francisco-based company that has pioneered so-called “Artificial Swarm Intelligence,” also picked Germany just ahead of Brazil.
Bayesian virtual octopus Paul is a fun prediction engine too http://paul.octane.mx/matches/Fra-Bel