What if your next head coach was an Artificial Intelligence?

During the late nineties, Billy Beane, general manager of the Oakland A’s, revolutionized baseball management by leveraging statistics in order to outsmart the richer ball clubs. His work was later subject to a book, Moneyball, which was made into film starring Brad Pitt as Beane. It is quite widely acknowledge now that Beane practices have revolutionized professional sport management and most professional team front office across every league in every country implemented at least part of his practices.

This revolution was achievable thanks to the emergence of computer statistics and the access to large data sets. As a matter of fact sport is a hobby where people are obsessed with tracking, making it an ideal experimental field for computer scientists and in particular those working around machine learning. Combining this big data with the recent improvements in CPU capacities, sport-related analysis a promising field for artificial intelligence. Twenty years after Beane success, are we on the breach of a new revolution lead by the emergence of AI ?

A Los Angeles based start-up called Second Spectrum is working on it quite intensively. Founded by former researchers at the University of South California 2 years ago, they focus on spatial dynamics and movement tracking in basketball in order to advise which play to go for. Figuring out all these relationships both relative and absolute of location, distance, timing, velocity is the real key to spatio-temporal pattern recognition or « the science of moving dots » as they like to call it.

Encountered difficulties

To start with, as one can imagine, like most data, sport data is hard to deal with and not that interesting. Body movements are very messy as people wiggle a lot.

On top of that, the raw data might seem meaningless or overwhelming. Second Spectrum expertise is to turn it into something that is consumable. For instance, data has to be put into perspective: during which event was this data taking place? Was the play successful or not? etc. Linking data to events requires a certain amount of sophistication in machine learning and the big data architecture.

Finally, one last difficulty is to code these machine learning algorithms accordingly. Explanations of a certain play by players can be messy, unclear or incomplete as they are numerous outcomes. Luckily machine learning enable to go beyond our ability to describe the things we know but that we yet are not capable to describe properly with our limited speech abilities.

Can a computer be better than professional NBA coaches?

I guess there are at least 2 facts that we can agree on easily:

• As machines have a greater computing capacity than a human being, they can proceed a much higher quantity of information
• A computer will always be more objective than a human brain, especially during game time where passion and excitements can interfere with decision making

Obviously, such company goal is to interpret a large quantity of acquired information to find relationships and patterns that may even be currently unknown to experts in sports activities. That being said, during a Q&A for the HP Basketball website, Rajiv Maheswaran, Second Spectrum CEO, describes their every humble approach with their prospects and customers: “we never go in and tell coaches and front offices that we have something they don’t know, and they should listen to us. What we always do is say we have the ability to get stuff out of this data that nobody else does. What would you like to know? What would you dream of having that you don’t right now?” In other words, they bring flexibility into big data that enable to extend the vision of a coach. Maheswaran claims that this year, all NBA championship contenders were using this product and that it has participated in changing strategies in very important game.

One example of potential discoveries they made is their analysis on relationship between the location of a shot and the odds of an offensive rebound. Based on the analysis of over 11,000 different shots, a pattern started to unveil where for every foot away from the basket, the chance of an offensive rebound decreases 1% until the 3-point line, where it suddenly improves. Hence, in order to get offensive rebounds, players need to move far closer to the basket: “90% of all missed shots can turn into an offensive rebound within 11 feet of the basket.”

I guess one aspect that would really improve their program would be real time analysis. If they need to turn their observations into “If this team is in this situation, here’s what they need to do to win”-type of recommendations, it needs to happen within a matter of second. I can clearly imagine 5 to 10 years from now where a Second spectrum consultant will be sitting on the bench right next to the head coach whispering him which players need to get on the court and which play to go for. Meanwhile, I think it is very interesting to observe that already today coaches who have been in the league for decades are willing to take advices from a machine.

Application beyond sport

Being able to influence the outcome of a game and eventually of a championship sounds really exciting to any sport fan but at the end of the day, the sciences of moving dots can be applied to other industries and to many areas of our lives. For instance, military commanders will have a great interest in being supported by an AI that have been through countless battle scenarios and offers suggestions on the ideal way of tackling the enemy. Retailers could easily track the movements of customers in a store and determinate the best spending rate depending on how items are laid out across their store. Even further out, office planners could get feedbacks on how people move around the space they have build out. I am convinced that the sciences of pattern recognition in a near future will help us build better buildings, better plan our cities and move in a more efficient way.