AI ping pong robot beats top human players, but don’t freak out yet

If you're primed to fear AI-driven robots replacing human workers at complex physical tasks, consider this your trigger warning.

A robot arm built by Sony, and named Ace, has just been dubbed "the first autonomous system to be competitive with elite human table tennis players." That's a quote from the study splashed across the front page of Nature, the world's most venerable peer-reviewed science journal.

The Ace researchers brought receipts. As you can see in the video above, the eight-jointed robot arm is able to make split-second decisions via an AI that's being fed real-time data from nine cameras. It scored a lot of points and won a few games against some of the world's top ping-pong players at Sony HQ in Tokyo.

But here's the good news buried in all the data. Yes, within the confines of this study, Ace was competitive. That doesn't mean Ace could figure out how to win every time; it's nothing like the half marathon-running robot that simply has to master one speed. And, crucially, the human players started to spot flaws in Ace's ping-pong strategy.

Ace isn't the first ping-pong playing robot. Researchers have long been interested in the sport because of its speed and real-time decision-making, which is a major frontier in robotics. In this respect, Ace marks a milestone for the AI system and for the highly reliable arm.

That arm was able to track a ping pong ball with 10 milliseconds of latency — more than 10 times faster than the human brain can manage.

"Ace’s striking skills are trained entirely in simulation using reinforcement learning, then transferred directly to the real robot," Sony explained in a blog post. "This is analogous to a player who practices endlessly in a virtual training hall and then walks onto a real court without needing to relearn anything."

The human players fight back

But that's just the thing — ping-pong players learn on the go, and they're looking at more than just the ball.

Mayuka Taira, who lost a match to Ace last December, told Sony the robot effectively intimidated her at first. "Because you can't read its reactions, it's impossible to sense what kind of shots it dislikes or struggles with, and that makes it even more difficult to play against," she said.

But then Rui Takenaka, who has both lost and won against Ace, went that crucial human step further. Here's what he told the company, emphasis ours:

If I used a serve with complex spin, Ace also returned the ball with complex spin, which made ​it difficult for me. But ⁠when I used a simple serve, what we call a knuckle serve, Ace returned a simpler ball. That made it easier for me to attack on the third shot, and I think that was the key reason why I was able to win.

Got that? Ace, a profoundly smart system, was suckered by a knuckle serve.

"Professional human athletes are very good at adapting to their opponent and finding weaknesses, which is an area that we are working on," Ace project leader Peter Dürr told Reuters.

So we shouldn't exactly hang up our ping pong rackets just yet. But we should certainly be very concerned about the mentions of security applications through the various reports and blogs about Ace.

Because the most lucrative real-world application of speedy systems like this isn't at the Olympics. It's on the battlefield — where being faster than the human eye may mean game over for human soldiers.