From GT Sophy to Table Tennis Ace: Why Reinforcement Learning Will Power the Next Generation of AI

Recent breakthroughs such as GT Sophy, the superhuman racing agent developed for Gran Turismo, and Table Tennis Ace, a robotic system capable of competing with human players in one of the world’s fastest and most dynamic sports, demonstrate a profound shift in artificial intelligence. These systems are not simply predicting the next token or recognizing patterns—they are learning to make decisions, adapt in real time, and master complex interactions with the physical and digital world.In this talk, we will explore how reinforcement learning (RL) enabled these achievements, why game-playing and robotic control remain some of the most demanding benchmarks in AI, and what they reveal about the path toward more capable, autonomous systems. We will examine how RL complements foundation models, allowing AI to move beyond knowledge and reasoning into action, planning, and continuous adaptation.As AI expands from chat interfaces into agents, robots, vehicles, and interactive environments, the ability to learn from experience may become as important as the ability to learn from data. Through the stories of GT Sophy and Table Tennis Ace, we will argue that reinforcement learning is not merely a specialized research area—it is a foundational technology for the future of AI, enabling systems that can operate, improve, and collaborate in the real world.