Artificial intelligence is a broad sector, from chatbots to performing more or less complex tasks. DeepMind opened up to the sports sector and started playing table tennis after mastering the game of Go.
DeepMind, the AI company best known for beating the world champion at Go, is now playing another sport: table tennis.
Sport as a training tool
If the trend is towards generative AI based on large language models, this is not the only sector that AI is covering. Sports is a sector that researchers prefer, because it allows AI to be trained on the basis of strategic and physical skills, all in real time.
Ping-pong is particularly chosen for its fast-paced nature and complex movements. Google DeepMind researchers explain: X (formerly Twitter) “The robot must master low-level skills, such as returning the ball, as well as high-level skills, such as strategizing and long-term planning to achieve the goal.”. Mastering skills that follow a learning mechanism that is almost similar to that of generative AI, which is more common among the general public.
Advanced training for amateur level
To develop this robot, engineers collect a huge data set about the ball or even existing movements. From there, the AI trains on this data by creating virtual simulations to perfectly reproduce the movements. This is followed by a practical application where the robotic arm is responsible for reproducing these movements and putting them into practice against a human opponent. A new learning loop is created from this data to master everything. Unlike generative AI, here the robot does not respond to a prompt, but must adapt to the situation it faces.
All that remained was to put it into practice. DeepMind recruited 29 human players, who were ranked into four skill levels: beginner, intermediate, advanced, and “advanced+.” At the end of the tournament against the bot, the machine won a total of 13 matches, or 45 percent of its duels. Performance comparable to that of “human-level amateurs.” Researchers.
From there, it’s a step up to seeing this robot compete in the Olympics. The researchers say the robot won 55 percent of its duels against intermediate players, but none against advanced players. The AI field continues to advance, aiming to surpass human intelligence, and now it’s looking to appropriate gestures.
Want to join a community of enthusiasts? Our disagreement Welcome, it's a place of mutual help and passion for technology.
