Oakland Ballers’ manager let AI take his job. A bold call, some glitches and a win followed

A few minutes before first pitch on Saturday, Oakland Ballers manager Aaron Miles paced the dugout fiddling on his iPad. 

This was no personal piece of tech, or a device he used regularly in his manager duties. This iPad was new and came equipped with an application that would take his job — just for the evening. 

Comfortably sitting in first place with homefield advantage for the postseason, the Ballers decided to get experimental, controversial even, and have artificial intelligence manage their game against the Great Falls Voyagers at Raimondi Park in Oakland. 

Lighting up Miles’ iPad was the AI application designed specifically for this game, programmed to do anything Miles would do. That means creating lineups, recommending pitching changes, pinch hitters, pinch runners and, as AI engineers who were on hand at the ballpark would later learn, selecting a hitter for the Pioneer League’s “Knock Out” round. The app even selected the Beer Batter, the most-likely player to strike out and earn the crowd a $5 beer.

“AI is scary to everyone in the fact it can take everybody’s job,” Miles said. “But it’s a cool thing what we’re doing right now because baseball is a game of stats and analytics. If you have the ability to compile stats at warp speed like AI has, it could have a place in the game and it probably will. In what capacity, I don’t know.”

“To our knowledge, this is the first time in human history that AI is going to make key decisions for a major sporting event,” Paul Freedman, the Ballers CEO, said. “So we’re breaking ground here.”

Yes, Miles’ job is safe. The 2006 World Series champion-turned-manager led the Ballers to an astounding 73-22 record, including a 3-2 victory Saturday, entering Sunday’s regular-season finale. Jokingly, he brought the iPad out at the official lineup exchange and had the iPad shake “hands” with all the umpires and opposing manager. The crowd laughed, he smirked.

Still, Raimondi Park was abuzz with AI’s infiltration into baseball. In the heart of tech-savvy Bay Area, locals were intrigued, but also dubious. 

One Ballers fan, Diane Wang, brought a sign voicing her displeasure. “AI prefer Aaron Miles,” with the “A” crossed out, it read. 

“I mean, Aaron Miles is probably the only manager secure enough that he would even allow this experiment,” Wang said. “The relationship between the manager and the team can’t be reduced to bits of information. It’s a fun experiment, but it’s reductionist thinking to avoid complexity of human acuity.

“AI should do the hard work. Let us do the creative stuff. We should be doing the writing, the arts and baseball.”

Kevin and Zack Benson, Pleasanton natives, mused about AI “taking peoples’ jobs” and that baseball might as well “be played by robots,” but were curious how the game would look.

“I would say it’s intriguing in a game like this to see what happens,” Kevin Benson said. “Otherwise, taking the human element out of the game would be awful. But it’s a fun idea to see what happens.”

The AI application — which dubbed itself AaronLytics, after its muse — was far from perfect upon arrival. The Mexico City-based company Distillery created AaronLytics in a little more than two weeks, and by Saturday morning engineers realized that the live statistic source it needed, called PrestoSync, wasn’t working with Open AI. 

That meant live stats — every pitch and play — were manually applied to the application in order to provide feedback for Miles during the game. Gavin Anderson, a partnerships representative from Distillery, flew down from Seattle on Saturday morning to pick up whatever would be lost in translation. He moved between sideline to dugout, laptop in hand.

Even before that, Miles’ human touch was needed. AaronLytics crafted a lineup identical to the one used on Friday, except that Dillon Tatum, the catcher on Friday, came to the ballpark sore on Saturday. Something AI wouldn’t know unless informed. So, Tyler Lozano replaced him. 

AaronLytics, though, provided a perfectly good rationale for the rest of the lineup — the application is built not just to recommend managerial decisions, but provide a reason for the suggestion. 

Tremayne Cobb Jr., a contact hitter with speed, would lead off. Esai Santos, batting second, boasted a .432 on-base average with a lefty-bat to work counts. Christian Almanza, a traditional power hitter, batted cleanup and Cam Bufford, batting .343, batted fifth to protect him. 

AI’s infiltration into baseball isn’t much of a shock to baseball lifers who’ve seen the game’s evolution within the analytic universe, this is just the first time the actual decisions were presented on an iPad. 

Since the Moneyball revolution began 25 years ago at the Coliseum — just a few miles down the road from the Ballers’ home — professional baseball teams have come to rely heavily on data and their own statistical models to inform nearly every decision they make. Information is so robust that data has become part of public consumption, from hitters’ launch angles, bat path and exit velocities to pitchers’ induced vertical break and arm angle. 

“In my mind, it’s going to be significant in sports. Implementing AI won’t just take over the decision-making, but help to justify it,” Distillery CEO Andrey Kudievskiy said. “This will help managers become better team managers and will help professional teams to win more games. Even if it’s a 1% chance, it’s significant in the baseball world.”

The Ballers as an organization have become a testing ground for major league baseball’s technological and tactical experiments. They use Automatic Ball and Strike, a Knock Out home run derby to decide extra-inning games and a little-used rule that allows batters to try for first base on any pitch that isn’t caught by the catcher. 

Data is so entrenched that Miles found himself consistently on the same page as AaronLytics. Anyone watching the first seven innings might not even remember that AI was calling the shots. Ballers starter Reed Butz was efficient, as was Voyagers starter Sam Lavin. Few baserunners and clean defense meant few opportunities for Miles to trot his iPad out to make a change. But AaronLytics lit up with suggestions, anyway. 

“If Cobb. Jr or Santos reaches, put on a hit-and-run with Drawek to exploit Lavin’s fastball tendencies and open holes,” the app read. “Consider a straight steal with Cobb Jr. or Santos if you sense a good jump — Lozano batting eighth gives you flexibility if you need to generate a run with the lower third.”

AaronLytics made its first big recommendation in the seventh inning, when the Ballers were up a run and needed desperately to create cushion. After Lou Helmig reached on an error, Miles replaced him with a speedier pinch runner in Darryl Buggs, who promptly stole second base. The next inning, T.J. McKenzie came in as a pinch runner and later scored a key second run.

“Good job, AI!” the ball boy said.

The eighth inning also featured a decision that had the Ballers dugout murmuring. Reliever James Colyer allowed a two-out baserunner, prompting AaronLytics to call for closer Connor Sullivan for the final out of the inning. Sullivan hadn’t completed many four-out saves, so the decision appeared bold to those who’d watched Oakland all year. 

Sullivan got the final out in the eighth, but surrendered the game-tying two-run blast in the ninth inning. Miles later revealed that he and AaronLytics were on the same page; he’d have Sullivan in for a four-out save, too.

“We’ve done it before. He called it just the way I’d normally do it, too,” Miles said. “Obviously he got the out in the eighth. He’s our closer in the ninth. Just didn’t happen.”

Pure chaos ensued when Anderson realized what happens in the Pioneer League in the event of a tie after nine innings: A Knock Out round, meaning each team gets one player to hit as many home runs as he can before recording five outs. Most wins. 

AaronLytics didn’t know about the rule. With help from Miles, Freedman and rules off the Pioneer League website, Anderson input the new rules into the system. AaronLytics picked Bufford, whose solo home run cleared the entire stadium earlier. 

He needed just two in the Knock Out round to win Oakland’s 73rd game.