When Will Artificial Intelligence Replace This Man?

A 25-year-old Erik Spoelstra used to sit in a storage room in the old Miami Arena, evaluating hours of game film to review player performance as an entry-level NBA video coordinator. Eventually, he climbed out of the audio visual muck to become head coach of the NBA's Miami Heat, where he would go on to win two championships.

It's a classic story of rags to riches—rising from junior video coordinator to head coach in about 13 years—and it may now be unlikely to ever happen again, as computers take over the position that gave Spoelstra his start.

"If an AI were fed videos of a huge number of past NBA games, and were smart enough to understand the events occurring in the games, then it could do a better job at making tactical basketball decisions like choosing starting lineups," said Ben Goertzel, a prominent futurist and lead researcher in the OpenCog AI lab at Hong Kong's Polytechnic University.

"As AIs with robust video understanding become widespread, I'd expect that we could see AI sports assistants start to play a serious role," he said.

Within five years, he expects the video coordinator position Spoelstra had "could be mostly done by AIs." Goertzel thinks that once an AI can 'learn from the patterns in the game, and intelligently extrapolate them into the future', humans won't be needed to make the types of decisions they are needed for today. In other words, AI is on its way to replacing much of what a coach does.

That timeline might shock the old guard of the sports world. The purists are only just acclimating to a post-Moneyball world where data rules, but artificial intelligence does seem to have a habit of evolving faster than we expect.

In recent years data analytics, sports science, and the measuring of things we couldn't measure before has made sports like one giant science project where the math nerds are the cool kids and the jocks are only just catching on. With statistics now flooding in, better technologies are needed to help cut through the noise.

"The NBA has built out a number of algorithms used to consolidate data coming from our SportVU system, in order to make it consumable for each of our teams," said Ken DeGennaro, senior vice president of IT applications at the NBA.

SportVU, now equipped in every NBA arena, tracks player and ball movement from a set of fixed cameras in the rafters. DeGennaro described how the system turns the once-unquantifiable chaos of the game into obedient statistics ready for inspection.. SportVU generates huge amounts of data, however, and that's where machine learning—a type of AI—comes in.

"Machine learning is becoming increasingly popular for all data modeling," said Keith Goldner, a statistician who consults for NBA and NFL franchises. "For example with play recognition, you can show an algorithm a series of plays, and let it teach itself to identify every time it recognizes that play type."

The technique works by creating algorithms that can improve themselves, rather than requiring software engineers to create a set of deliberately programmed instructions. The algorithm self-designs its model by responding to positive or negative feedback on the result it comes up with, so the AI can teach itself to understand if it's seeing the correct type of play. In one case, scientists demonstrated a machine learning tool that could classify various NBA teams' version of the pick and roll.

By mimicking ways of processing data that was only possible from a human coach, machine learning AIs call into question the very limits of what humans are uniquely qualified to do. Teams that figure out how to use this data to make better decisions have a serious advantage, Goldner said. Coaches can use the system to track the frequency of different plays run by their opponents, for example, and prepare accordingly.

Eventually it seems likely that a system like SportVU would be able to process raw video footage, and once it does, AIs could further displace what human coaches do. So how long until a system like this can be developed?

Goertzel admits that we're not quite there. "Video understanding is still a hard problem, and sports events are fast and complicated. I do think it's a few years before it's compellingly solved, but video understanding is an area that big tech companies are putting huge effort into. The same sort of deep learning methods that have worked for image understanding can work for video recognition too. I expect massive breakthroughs in the coming years."

All this data-stuff is nice and fine, but fans watch sports to see very real athletes doing exceptional human things. Coaches deal in the currency of team chemistry, full of the intangible things we can't measure—at least, not yet. There's a social and emotional aspect to sports, and not everyone may welcome machines who meddle in areas that were so decisively human.

In describing the scientific changes taking place inside teams' front offices, Goldner can give off outsider vibes. "The culture is changing little-by-little and for guys like me it's about building trust. The Moneyball revolution has often been portrayed as quants versus scouts, with those who played the game bickering with the scientific types who did not. The truth is, in most successful organizations, everyone realizes the value of both."

Teams willing to adapt to the technological times ahead will continue to develop a competitive advantage, and for those convinced that coaching is too human an affair to be left to software, we're already witnessing the beginning of AI's creeping further in. IBM Watson recently announced a project with Spare5 to develop an app that can coach a golfer's swing, and an NFL source shared that some teams are using Kitman Labs, a machine learning company, to personalize each of their player's health.

Neither Goldner nor Goertzel expect human coaches to fully disappear from the sidelines, but going forward we're likely to see an increased partnership between coach and machine. We'll see the standard profile of a coach change, and the role of their assistants will evolve. Goertzel goes further. "In the medium term, AI coach assistants will deal with the tactical stuff. Eventually, mostly just the social and strategic aspects will be left for humans."

If the data analysis and decision making is left for these all-knowing AIs, coaches may be stuck with the very human task of leading their teams. In a future overrun with algorithms, coaching might then become more human than ever. That's great—but we may miss out on finding a few Erik Spoelstras along the way.