revolutionary tools.  groundbreaking articles.  proven results.

A Robot Could Do Your Job: The Future of NFL Playcalling – The Wrong Read, No. 27

Welcome to the 27th installment of the “The Wrong Read,” an article series that reflects on recent podcast episodes, pushing the ideas discussed on the podcasts to their logical conclusions and offering some further thoughts on the topics broached by the guests and hosts. Darryl Slater recently came on RotoViz Radio to say the Jets could take Baker Mayfield sixth overall. Here are my thoughts on why that could be a great pick. You might want to click on that link if you’re interested in a traditional football article. What follows below gets a little weird. What can I say?—it’s the offseason. You’ve been warned.

Most industries today are embracing some sort of automation. Airplanes more or less fly themselves. Amazon uses robots to pick items off shelves and drones to deliver them. A growing number of people have their investments managed by robo-advisors. Almost all manufacturing and assembly involves only minimal human intervention. The upshot of all this automation is not only decreased labor costs, but also increased safety, accuracy, and quality.

A recent episode of the Coaches Box gave me the occasion to think about automation in a new industry. On the episode titled “Fade the Fade,” host Anthony Amico opined about the suboptimality of using the fade route near the goal line. Fade routes yield completed passes only about 33 percent of the time. They do tend to have a higher touchdown rate than other routes. But this is mainly because they are generally only used near the goal line, whereas most other routes are used in all areas of the field.

Anthony suggests that teams should use the slant route more often near the goal line. Slants are caught more than 67 percent of the time. Based on data from Football Outsiders, passing to a slant route is about 11 percent more productive than the average pass. Passing to a fade route, on the other hand, is about 12 percent less productive than average, according to DVOA. As chronicled here before, these are not the only suboptimal playcalling decisions NFL coaches tend to make.

Suboptimal outcomes are being phased out in most other areas of life. Self-driving cars promise to make roads safer. Artificial intelligence promises to deliver more accurate medical diagnoses. Humans are increasingly being replaced by machines to handle tasks that humans are simply not as efficient at performing. So why haven’t NFL playcallers been replaced by machines yet? Some NFL team should do this right away.

Why Not AI Playcallers?

Counterarguments could be posed along two possible lines:

Machines Aren’t Good Enough

First, one might argue, computers lack the sophistication required for this proposal. I beat the computer in Madden all the time. If AI was really better at calling plays, the computer should beat me. But this objection makes a number of assumptions that we can dispense with. First, it assumes that Madden’s AI is the best version of an NFL playcalling AI—it probably isn’t.

Even if it is, however, it also assumes that the AI in Madden is engineered to call the optimal play, rather than the most realistic play. That is to say, it fails to take into account that the creators of Madden wanted to make a realistic simulation. They wanted to mimic NFL coaching tendencies, not to make a computer that could beat anybody at Madden. The truth is that machines already can defeat the world’s best at chess and Go; Madden is trivial. Assuming Madden is a reasonable analogue for the NFL, NFL playcalling is trivial.1

Of course, more goes into being a coach than calling plays—you can’t expect a computer to manage the responsibilities of dealing with a team full of egos and human personalities. This I can grant. But I’m not suggesting computers take over all head coaching or coordinating duties—only the playcalling.

Machines Are Boring

The second sort of counterargument imagines a world in which my proposal has already been realized league wide. In that case, machines simply call plays against each other. Winning or losing the game then depends less on the playcalling and more on the skill of the players a team has been able to acquire, and the ability of those players to execute the plays. Coaching doesn’t matter in this world.

I admit, this objection has some merit. But the scenario being described is probably a long way off. The fact that it’s newsworthy that the Rams intentionally get to the line quickly so Sean McVay can call audibles for Jared Goff before his mic cuts out indicates just how slow the NFL is in embracing innovation. Radio communication between coaches and QBs has been legal since 1994. Why did it take so long for a team to start doing this? This quote from a Boston Globe article on the NFL headset sums up the attitude that will likely keep AI from gaining ground in the NFL:

“Some people were still wary of technology,” said Brian Billick, an NFL Network analyst and former Ravens coach who was the offensive coordinator for the Vikings in 1994. “They’d use rotary phones if you could find them, still to this day. Some people were just naturally, ‘Nah, nah, nah, that’s too much, that’s not football, we don’t want that,’ but there’s no question that it made things more efficient, that it saved time.”

The “that’s not football” attitude is prevalent enough to prevent any rapid widescale adoption of AI as a playcalling tool for the foreseeable future.

And in any case, by the time a few teams have installed AI playcallers and gained a massive advantage as a result, the NFL may change the rules to limit AI’s usefulness during the game.2

AI Playcallers Might Be Inevitable, Eventually

To be perfectly honest, I’m unsure of all the hurdles this proposal would have to surmount to become a reality. But I also think it’s probably only a matter of time before some forward-thinking team publicly employs AI in some capacity when calling plays, and has great success doing so. In theory, the role of NFL playcaller (as distinct from other coaching duties) should be one of the most automatable jobs around. Your time has almost come, fade route.

  1. Calling the best play is really a math problem anyway.  (back)
  2. At which point the most innovative coaches will probably attempt to enhance themselves to improve their “natural” decision-making. Seriously.  (back)

recent and related...

in case you missed it...

Week 16 Primetime Slate DFS Breakdown

Who are the best plays for the Week 16 DFS primetime slate on DraftKings and FanDuel? Tyler Loechner runs through notes on the primetime short slate featuring SNF/MNF games.1 DSF Week 16 Primetime Slate DFS Week 16 Primetime Slate theory thoughts: All four QB options have some angle of appeal

Read More

NFL Week 16: Coulda, Woulda, Shoulda

Monty Phan wraps up the weekend of NFL action. For nearly four months every year, my wife puts up with a husband who is totally useless on Sundays, a slave to the Red Zone Channel and NBC Sunday Night Football. Just when the season crescendos to a climax – the

Read More

Michael Thomas Sets Records As Surprise WRs Make Their Case to Dynasty Owners

Shawn Siegele looks at all the top wide receiver performances from Week 16, including big games from Sterling Shepard, Diontae Johnson, and Steven Sims Jr. Championship weekend featured a title-winning fourth-down score from Tyler Boyd and monster performances from two legends. Michael Thomas and Julio Jones combined for 22 catches

Read More

Sign-up today for our free Premium Email subscription!

© 2019 RotoViz. All rights Reserved.