If you pay attention to the countless academic studies devoted to the subject or the dire prophecies from Elon Musk or Stephen Hawking, an artificial intelligence (AI) age is rapidly approaching that promises to make most of our jobs redundant.
A McKinsey and Company report published earlier this year estimated that more than half of the world’s activities could be automated within 40 years.
Even football will not be free of its effects. Jobs requiring repetitive tasks or data collection are most at risk but the robotic takeover does not begin and end with ultra-efficient document checking; in the two decades since IBM’s Deep Blue computer beat grandmaster Garry Kasparov at chess, artificially intelligent machines have learnt to make art and compose symphonies.
Who is to say that AI could not one day park a bus better than Jose Mourinho, spot an opposition’s tactical weakness quicker than Pep Guardiola or scout the next N’Golo Kante before anyone else?
Expected goals 2.0
This season’s introduction of the expected goals (xG) metric — which aims to measure how effective a team’s attacking play is — to the mainstream via its use on Match of the Day demonstrates the increasing sophistication of statistics in the modern game.
Every top-flight club has its own analysis departments — Arsenal even bought an entire company specialising in the field — while modern players double as endless data providers, spending their days hooked up to various GPS trackers and biometric devices.
Yet with the sheer wealth of granular data threatening to exceed our ability to interpret it, some companies, such as Dutch start-up SciSports, are using machine learning — systems that can adapt to tasks they’re not explicitly programmed for — to ingest and interpret numbers at scale.
“Expected goals is too limited,” founder and chief executive Giels Brouwer told City A.M.
“It doesn’t take into account the context of a situation. So we built a system that calculates the game state of every action — determining what the chance of scoring is before a pass takes place and after a pass takes place. Then we can add an x value to every thing that happens on the pitch.”
While scouting now routinely takes into account a player’s statistics, those numbers only provide a clearer picture of how they have already performed. Working out how a player may develop in the future is still somewhat reliant on sage-like intuition.
SciSports is using machine learning to change that with a product called Insight.
“It can look at how a player’s game is improving and then look for players who have demonstrated similar growth,” says 26-year-old Brouwer. “It’s worked out how good those older players were at their peak level, and then give a potential rating to the younger player.
“The most beautiful thing about machine learning is it can cope with such large amounts of data. Our initial models weren’t using machine learning but then we saw really good improvement when we started to implement it.”
It’s not just assessing players with which AI could help football clubs in the transfer market, but also with working out which clubs are likely buyers or sellers.
As well as discovering new genes related to motor neurone disease or making its own film Hollywood film trailer, IBM’s pre-eminent supercomputer Watson can analyse reams of historical transfer data and this summer predicted where players would move to with 63 per cent accuracy.
Alongside the performance stats taking up space on clubs’ hard drives is physical data on prize assets whose prolonged absence from a team could be worth millions in wasted wages.
Silicon Valley-based Kitman Labs provides Premier League clubs, as well as leading teams in a host of other sports, with machine learning technology capable of analysing the millions of data points produced on each player and the multivariate different contexts in which injuries occur to uncover their unique stress responses and combat future issues.
Smart sports marketing
AI is being used for off-field ventures too. Data company Sportradar, which works with a number of football leagues across Europe as well as the NFL and NBA, uses machine learning to clean its data, perform analysis and even build visualisations.
“You want to be hands-off because you don’t want the biases of the human mind to affect the process,” Sportradar senior vice president of product management and technology Ashok Balakrishnan told City A.M. “Once you train the machine, you want it to be making its own data-driven conclusions.”
Californian company GumGum, meanwhile, uses image recognition technology to scan through the web and assess how far the sports sponsorships of the likes of Toyota and McDonalds proliferate, helping blue-chip companies to better understand how valuable their inventories are.
Football’s war against the machines appears to be pretty much over before most even knew it had begun. Mourinho probably does not need to worry about being replaced by the Robotic One just yet, however.
A 2013 Oxford University study ranked “athletic trainers” as the 35th least-likely occupation to become automated. That is safer than physios, chief executives, statisticians – and even athletes themselves.