Teaching machines in a manner akin to animal trainers shaping the behavior of dogs or horses has been a crucial technique for advancing artificial intelligence. This approach was distinguished on Wednesday with the prestigious A.M. Turing Award, the computer science community’s equivalent of the Nobel Prize. Andrew Barto and Richard Sutton, two trailblazers in the realm of reinforcement learning, have been named this year’s recipients for their groundbreaking work.
Their research, initiated in the late 1970s, laid the foundation for numerous AI advancements over the past decade. At the core of their efforts was the development of “hedonistic” machines capable of dynamically adjusting their actions in response to positive feedback. Reinforcement learning, championed by Barto and Sutton, led to significant breakthroughs in AI, including a Google program triumphing over human Go players in 2016 and 2017, enhancing AI tools like ChatGPT, and optimizing financial trading.
Although the concept of reinforcement learning was not in vogue when Barto and Sutton first delved into it at the University of Massachusetts, Amherst, they persisted in crafting their theories and algorithms. Barto reflected on those early days, noting the lack of recognition for their work at the time. However, the field has now gained prominence and relevance, as evidenced by their well-deserved award.
This year’s $1 million prize, sponsored by Google, was announced by the Association for Computing Machinery. Barto, now retired, and Sutton, a distinguished professor at the University of Alberta in Canada, have contributed significantly to the field of AI, aligning with Alan Turing’s vision of machines that can learn from experience.
Their research, drawing from psychology and neuroscience, explored how pleasure-seeking neurons respond to stimuli, resulting in innovative approaches to tasks like balancing a pole on a moving cart. These contributions have not only shaped the landscape of AI but have also inspired countless researchers and attracted substantial investments.
In a joint interview, Barto and Sutton offered differing perspectives on evaluating the risks associated with AI agents that continually seek self-improvement. They also drew a distinction between their work in reinforcement learning and the current trend of generative AI technology employed by companies like OpenAI and Google.
Sutton downplayed concerns about AI’s potential threats to humanity, while Barto emphasized the importance of considering unforeseen consequences. Despite their differing viewpoints, their combined expertise and contributions have significantly advanced the field of artificial intelligence.
Sutton, with his eyes fixed on a future where beings of higher intelligence surpass current humans, contrasts sharply with Ears, who proudly identifies as a Luddite. This vision, often associated with posthumanism, envisions a realm where humans are not the pinnacle of evolution but instead a stepping stone to something greater.
“People are complex machines,” Sutton reflects, acknowledging the remarkable capabilities of the human mind. However, he also recognizes that humans are not the final product of evolution but rather an intermediate stage. “It’s an inherent aspect of the AI endeavor,” Sutton elaborates. “Our goal is to not only comprehend ourselves but also to create entities that can surpass our own capabilities. Perhaps even to evolve into such entities ourselves.”
