Artificial Intelligence Drives Real Insights in Sports

By Adam Grossman and Alberto Rios

How do you teach a computer to see? For a large part of the twentieth century, leading scientists and mathematicians thought this was impossible. In the late 1980s, however, the first computer neural network was created that programmers could train to recognize images, logos, and people in videos or images. Today, consumer grade graphics processing units (GPU) have become powerful enough to enable computers to deliver near real-time results. Teaching a computer to “see” enables the machine to collect more data more quickly and more accurately than ever before to answer when, where, and how long a sports sponsor’s logo was on screen during a game broadcast or press conference.

Within the last 30 years, the concept of machine learning has transformed from being a near-impossible task to now being an integral part of the sports industry’s future. A good example of the use of machine learning on the field is PitchF/X in baseball. PitchF/X “tracks the position, speed, and break of a ball, outcomes like hits, strikes, foul balls, and so on, and uses a machine learning algorithm to categorize the pitch type (e.g.: four-seam fastball, changeup, curveball, slider)”. PitchF/X removes the need for humans to watch videos and document every at-bat. Instead, teams can collect information about both hitters and pitchers more quickly and more accurately than ever before. This information then can be provided to hitters as insights on what pitch is more likely to be thrown by a player during a certain pitch count. In addition, this information can be provided to pitchers on what pitches a hitter will be more likely to swing at during an at-bat.

How exactly does a computer learn to see? At our company, Block Six Analytics (B6A), we have developed a neural network for object identification using computer vision software we call our Media Analysis Platform (MAP). One can think of a neural network as the brain of a computer. The human brain learns through repetitive practice to acquire knowledge and by storing said knowledge in neurons, or brain cells. Each neuron collects a tiny bit of information and combines it via neural pathways in the brain to create an output. For example, students can learn algebra or how to speak in Spanish by practicing tasks repeatedly. Each neuron in the brain learns a bit of information about the subject and the combines so that one becomes fluent in algebra or Spanish. Our neural network takes a similar approach to learning how to identify a specific object. We teach our system by showing it hundreds pictures of that object and it “practices” on how to classify that object based on features such as shape and color. Each practice session enables our neural network to learn how to become better at object recognition until it masters the concept.

Machine learning enables everyone in the sports industry to collect and analyze more data than ever before. The challenge is how to communicate the insights using this data to improve decision-making on important strategic challenges. In the Pitch F/X example, baseball players have fractions of seconds to determine whether to swing or not swing at a pitch. Giving players all of the data that PitchF/X provides may slow down their decision-making processes and make them worse hitters. As Travis Sawchik described in his book Big Data Baseball: Math, Miracles, and the End of a 20-Year Losing Streak, the Pittsburgh Pirates’ data analysts spent hours talking with coaches and players talking about machine learning. In addition, their data analytics team created spray charts, heat maps, and other data visualizations tools that made as easy as possible for players and coaches to understand the data.

Artificial intelligence is easier to leverage than ever before and creates real insights that impacts strategic decision-making. Big data analysis, however, requires a communication strategy to translate machine learning into actionable intelligence that can be used by everyone in a sports organization.