In the 1980s Andrew Barto AND Rich Sutton They were considered eccentric admirers of an elegant, but finally convicted idea – saving machines learn, like people and animals, from experience.
Ten years later, with a technique, which a pioneer, now more and more critical for contemporary artificial intelligence and programs, such as Chatgpt, Barto and Sutton, received the Turing award, the highest honor in the field of computer science.
Barto, a retired professor at the University of Massachusetts Amherst and Sutton, a professor from the University of Alberta, followed the technique known as learning to gain, which consists in persuading the computer to perform tasks through experiments in combination with a positive or negative opinion.
“When this work began for me, it was extremely unfashionable,” Barto recalls with a smile, speaking because of the magnification from the house in Massachusetts. “It was amazing [it has] He achieved some influence and attention – adds Barto.
Learning to strengthen was perhaps best known by Google Deepmind in 2016 for the construction of Alphago, a program that learned how to play in an incredibly complex and subtle board game to the level of experts. This demonstration caused new interest in the technique, which was used in advertising, optimizing energy consumption, finance and chip design. This approach also has a long history of robotics, in which it can help machines learning to perform physical tasks through test and mistakes.
Recently, reinforcement learning is crucial for conducting the results of large language models (LLM) and creating extremely talented chatbot programs. The same method is also used to train AI models to imitate human reasoning and building more talented AI agents.
Sutton, however, notes that the methods used to conduct LLM include people who provide goals, not the learning algorithm only through their own exploration. He says that having machines can be more fruitful on their own. “The great division is whether [AI is] Learning from people or whether learning from your own experience, “he says.
Over the past few decades of Jeff Dean, senior vice president of Google, said “Sutton’s work” Sutton’s works will be progressed in AI, “said Jeff Dean, senior vice president of Google, in a statement issued by Association of Computer Machines (ACM), which passes the Turing award. “The tools they developed remain the central pillar of the AI boom and contributed to significant progress.”
The reinforcement has a long and chessboard in AI. It was at the dawn of the field, when Alan Turing suggested that machines could learn through experience and opinions in their famed 1950 article “Computing machines and intelligence“, Which examines the view that the machine may someday think like a human. Arthur Samuel, AI pioneer, used the reinforcement learning to build one of the first machine learning programs, A system capable of playing checkersin 1955