A robot that can beat the average person at table tennis with AI is more impressive than you think

Most people will think of lasers, glowing red eyes and end-of-the-world scenarios when you mention AI and robots. Google DeepMind, on the other hand, has made a huge step forward in robotic AI by training a computer to beat an average person at table-tennis instead of eradicating humanity.

Google DeepMind posted a video of the robot in action on X. As someone who has had some experience programming robot arms in a manufacturing engineering capacity, albeit limited, I was very impressed by what I witnessed. The study's research document is well worth reading.

It's not just a camera tracking the ball, and then motors swinging wildly a paddle around. Google DeepMind created a database that included all the possible initial states of a table tennis balls, including its position, speed, and spin. The robot arm then practised different movements, such as switching between forehand and backhand grips and applying topspin.

The AI system was designed to monitor how players behaved and played, and then use that information for the algorithm. The success rate of the algorithm was tracked, and the selected strategy was self-adjusted in real-time.

Google DeepMind claims that the robot competed against 29 opponents ranked in four different skill levels. After completing all of the matches, the robot came roughly in the middle--basically the same as an "intermediate amatuer."

It was beaten by the best players. One clip shows someone starting slowly, but then destroying the AI in one shot. Google DeepMind said that this was to be expected as aspects like how the rubber on the paddle affects spin were difficult to model in simulations.

Even so, I can imagine many applications for future versions of this technology. Robots are used in mass-manufacturing lines for welding and painting. They are great, but they can't always cope with minor lighting changes or mispositions. An AI system that is trained to respond to these issues will help to prevent production lines from being slowed down.

I can imagine a future where seriously injured people will rely on AI-trained robotic arms to do their work for them while they recover.

For now, I'm totally on board with the goal of improving at table tennis. It's better than improving at robotic nuclear war.