How Robot Umpires Apply To The Sports Industry
BY ADAM GROSSMAN
The independent baseball Atlantic League announced that will it will use robot umpires to call balls and strikes during games for the remainder of this season. The robot umpires “uses the TrackMan, a network of calibrated lasers that work to call strikes and balls that (if all goes well) are accurate to each hitter's unique strike zone.” The Atlantic first used robot umpires during its All-Star Game earlier in July and will start using them full-time on Thursday.
The idea of using machines instead of humans for specific tasks is something we are familiar with at B6A. For example, our Media Analysis Platform (MAP) leverages machine learning to identify and classify logo activations in linear, OTT, digital, and social video / images. The reason we adapted this approach is that machines have the ability to be more accurate and analyze content more quickly than humans particularly when logos are moving quickly across the screen (i.e., NASCAR races or even NBA jersey patches).
This is a similar thesis to using robot umpires. More specifically, professional baseball players routinely pitch baseballs in excess of 90 mph which provides human umpires four-tenths of a second to analyze a pitch for a ball or strike. While this is an extremely small amount of time for a human, the typical computer central processing unit (CPU) “can execute 1,800,000,000 instructions” per second. This means four-tenths of a second is a relatively long time for a machine to analyze this information.
It is admirable how accurate umpires are, particularly at the MLB level, given how fast pitches get to home plate. However, between 2008-2018 MLB umpires made incorrect calls 12% of the time. A study published in The Conversation also found that, “when batters had two strikes, the error rate for all umpires increased – incorrect calls happen 29 percent of the time.”
There most common explanation for the increased error in two-strike situations is a false negative issue in that umpires do not want to call a third strike because that ends the at-bat with an umpires decision rather a player’s action. If an umpire calls a ball then (except for counts with three balls and two strikes) the at-bat can continue with the hitter continuing to have the ability to swing and miss or make contact with the pitch.
These are the types of biases that are common and completely reasonable for humans to have but can be eliminated by machines. For example, a machine is not going to care if a player or manager yells at it for calling a third strike. And yet this same The Conversation umpire study that documented umpire error also expresses the belief that umpires should not be replaced by machines completely, particularly given the complexity of what an umpire has to do outside of calling balls and strikes.
The Conversation article recommends that “Umpires could easily be fitted with ear pieces connecting them to a control center that conveys real-time ball and strike information. These tech-assisted umpires could then make calls correctly, quickly and effortlessly…Umpires could remain the final arbiter, having override ability under certain circumstances, such as if a ball hits the ground before crossing the plate or if a system outage occurs.”
The ability of a human to potentially override a machine is a core part of MAP as well. In particular, we use our Visual Verify tool within our General Manager backend platform to manually analyze each positive frame to identify if / when a machine has made an error. We particularly focus on the beginning of a season, new client relationships, or new activations to ensure we can be as accurate as possible with our analysis.
We then can update our algorithms in the appropriate way when needed so we can maximize the processing speed of a machine to deliver results typically within 72 hours after a game occurs. This also enables our team to be more focused on delivering insights to our clients on what can be done to improve performance and how that can be communicated to internal and external stakeholders.
Leveraging machine learning to help with specific and often repetitive tasks where it is better-suited than humans is a good example of how this type of technology can work both on and off the field. While this may change in the future, for now machines provide the ability to free up humans to focus on more complex issues that can provide more value to core sports industry challenges.