If you think technologies like Instant Replay are changing sports, wait ‘til Machine Learning catches steam.
There has been a lot of discussion in the Major League Baseball playoffs about the place of Instant Replay in the game of baseball. It has been a season long discussion, ever since the league introduced a “challenge” system similar to the one in the NFL. The discussion has intensified now that the games matter so much. The common refrain is that the introduction of replay inhibits the “spirit of the game” by changing the way it has to be played. But if instant replay is making purists uncomfortable, machine learning is going to be agonizing.
Machine learning is a field in Computer Science where algorithms are designed to learn from data and make data driven considerations. Given some input and some context, an algorithm can be trained to improve upon itself and get better and better at its “job”. As I will get into, “data” and “input” can be literally anything. If it can happen, it can be interpreted as data and machine learning is the field capable of making use of this data. The science is already now being used across many domains, even if it is only at the naissance of its popularity, so it can be best understood with real world examples. One example would be an algorithm trained by law enforcement to find and recognize child pornography on the web. It is difficult to think that a computer could “recognize” an image or video as child pornography and not just a blob of pixels, but that is actually what Machine Learning does best. An algorithm can begin to recognize patterns in the pixel data, weeding through petabytes of images and videos until the patterns show themselves. Machine learning is the computational strategy we turn to when there are too many variables to account for. “Typical” computer science relies on these cases and variables and is limited by human-level quantities. No human could ever “program” a self-driving car to be aware of every traffic situation and indication there is, but an algorithm can “learn” them through extensive pattern recognition.
The implications of machine learning in sports pale in importance to the implications on many other walks of life, like the few described above, but the ramifications are fascinating none-the-less. There is a “purism” in sports that does not exist in a lot of other facets of life. That purism is the reason steroid scandals are given the importance they are. Nobody is especially bothered if an actor is revealed to have used the drug to become a character on the screen, even if it does make you uneasy. Sports, however has a tradition that does not want asterisks attached to it. This does not need much argument, it’s inherently clear that people have a hard time embracing change in sports a lot of the time.
The thing I find most fascinating about the future of machine learning in sports is that I think this happening right now. Decades from now, these technologies will have obviously had a major affect on our lives, but the adoption of this technology will be measured season-over-season, not decade-over-decade. From the point of view of the creators of the game-specific technology, it will be as easy as adopting the software libraries already made available. While truly understanding the undelying mechanics of machine learning, implementing it will soon be a synch for spirited technologists and entrepreneurs willing to stand on the shoulders of the giants who have made the big leaps. Machine learning services are already being made available through Amazon Web Services. There are already many startups in the space and it only takes one to make a big impact. Many professional sports leagues have already moved to a data driven approach and coaches are increasingly expected to be literate in the way these approaches are handled.
The least intuitive part about understanding the impact machine learning can have is the fact that it can make a computer really good at what we typically think computers are very bad at. Specialized programs will soon be able to recognize the difference between a power run and a counter in football, just by watching the video, the way a seasoned coach would. The program also knows exactly how many times that team has run that play this year, field position, circumstance, injuries that might affect play calling and every other detail that might exist. It will know how often the middle linebacker overcommits to the play when he sees the pulling guard, and whether it happens more often in zone or man and whether or not it is available to exploit. This will happen with limited need for data entry. The machine will examine all the film and make these determinations. Opposing coaches will need to begin calling plays against the machine, not the human. In order to know how to accurately call “against the machine”, the coach will need a machine themselves. There exists a reasonable likelihood that it will ultimately make a lot more sense strategically to let the machine call the plays, rather than risk human error.
Moneyball demonstrated how baseball shifted from scouting for “5 tool players” to primarily examining data in order to make personnel decisions. There is still value in making scouting based inferences, even if they are flawed by human error. But what about when machines are able to identify athletic traits and injury risks through pattern recognition. Just capture as much video as you can and you have your answers. Machine-aided outcomes will never be perfect, but they only need to be better than human outcomes in order to be the obvious choice.
There will be resistance to the machines, perhaps even through lots of regulation of their use, but drawing all the lines and doing so predictively will be impossible and the competitive advantage the machines to those with access will ensure their adoption. I’m making a lot of assumptions in order to say that things will play out the way I have described but there are already people and companies making use of this technology in the games and this just the beginning. By the end of the decade some of the situations I have described will be taking form and someday not too far, a coach or manager somewhere may lose his or her job to a machine.