Andrew Gehret Barto (also recorded as Andrew Garrett Barto, Andrew Greig Barto, Andrew Geret Barto, or Andrew G. Barto; born in 1948, around 1948, or 1948 or 1949 in the United States) is an American computer scientist and engineer who is currently professor emeritus of computer science at the University of Massachusetts Amherst.^[1,7,19,25] He is best known for his foundational contributions to the field of modern computational reinforcement learning, of which he is recognized as one of the founders.^{[3,18,26,4,27,28,14]}

Barto grew up as the son of an artist and a machinist, and he was fascinated by mechanical apparatuses from childhood.^[15] He studied at the University of Michigan, earning a Bachelor of Science in mathematics in 1970, followed by a master's degree and a Ph.D. in computer science in 1975 under Bernard P. Zeigler with the dissertation Cellular Automata as Models of Natural Systems.^[16,6] In 1977, Barto joined the University of Massachusetts Amherst during his postdoctoral stage, advancing to the professorial career track in 1982, becoming a full professor in 1991, and serving as chair of the Department of Computer Science from 2007 to 2011 before retiring as professor emeritus after teaching as a professor there from 1991 to 2012.^[25,8]

Barto's research employs an interdisciplinary perspective bridging psychology, neuroscience, and biology to explore learning processes in biological systems, such as organisms, as well as in machines. Alongside his students and collaborators—including Richard S. Sutton (Ph.D. 1984), Amy McGovern (Ph.D. 2002), and Satinder Singh—Barto developed foundational conceptual frameworks, mathematical formulations, and algorithmic innovations such as temporal difference learning, policy gradient methods, actor-critic architectures, the integration of neural networks, and methods for constructing autonomous agents.^[4,18,17,27] Although reinforcement learning was not fashionable when its foundations were drafted in the late 1970s and 1980s, the paradigm became central to artificial intelligence by the 2010s, powering achievements such as Google's AlphaGo defeating world champions in Go in 2016 and 2017, OpenAI's ChatGPT, DeepSeek chatbots, financial trading optimizations, and robotic hand manipulation; Barto also co-authored the standard textbook Reinforcement Learning: An Introduction (MIT Press, 1998; 2nd ed. 2018).^[27,28]

Barto is a fellow of the Association for the Advancement of Artificial Intelligence (AAAI), the American Association for the Advancement of Science (AAAS), and the IEEE. His honors include the IEEE Neural Networks Society Pioneer Award in 2004, the IJCAI Award for Research Excellence in 2017, and the ACM Turing Award together with Richard S. Sutton in 2024 or 2025 for developing the conceptual and algorithmic foundations of reinforcement learning.^{[4,8,14,3,9,24,26,27,28]}

Andrew Gehret Barto was born in the United States in 1948 or 1949 (according to some sources, in 1948).^[6,19] In 1970, he graduated with a Bachelor of Science degree with honors and distinction in mathematics from the University of Michigan, after initially majoring or specializing in architecture, naval architecture, and marine engineering. After reading works by Michael Arbib, Warren Sturgis McCulloch, and Walter Pitts, he became interested in using computers and mathematics to model the brain and brain function, and five years later earned a doctorate (Ph.D.) in computer science for a thesis on cellular automata.^[7]

In 1977, Barto joined the College of Information and Computer Sciences at the University of Massachusetts Amherst as a postdoctoral research associate, was promoted to associate professor in 1982, and became a full professor in 1991. From 2007 to 2011, he served as department chair and was a core faculty member of the Neuroscience and Behavior program.^[8] During his time at the university, Barto co-directed the Autonomous Learning Laboratory, which generated several key ideas in the field of reinforcement learning. Richard Sutton, with whom he co-authored the influential book Reinforcement Learning: An Introduction, was his doctoral student.

Barto received a bachelor's degree in mathematics from the University of Michigan in 1970, and five years later received a PhD in computer science. In 1977, Barto joined the College of Information and Computer Sciences at the University of Massachusetts Amherst as a postdoctoral research associate, was promoted to associate professor in 1982, and became a full professor in 1991. From 2007 to 2011, he served as department chair, and he was a core faculty member of the Neuroscience and Behavior Program.^[8,11] During his time at UMass, Barto co-directed the Autonomous Learning Laboratory (originally the Adaptive Networks Laboratory), which produced several key ideas in reinforcement learning.^[8,4,29,11]

Barto graduated 27 doctoral students, 13 of whom went on to become professors.^[8,11] Richard S. Sutton, who was his PhD student, co-authored with him the influential seminal book Reinforcement Learning: An Introduction (MIT Press 1998, or 1988 according to Chinese Wikipedia; 2nd edition 2018). Starting in 2012, Barto retired, though he still holds the position of co-director of the Autonomous Learning Laboratory. Currently, he also works as an associate editor, serves on the Advisory Board of the Journal of Machine Learning Research, and is an editorial board member.

When Barto began working at the University of Massachusetts, he joined a group of researchers exploring the behavior of neurons in the human brain as the basis of human intelligence, a concept that had been advanced by computer scientist A. Harry Klopf. Together with his doctoral student Sutton, he used mathematics to advance this concept and apply it as a foundation for artificial intelligence.^[5] This concept came to be known as reinforcement learning and developed into a key fundamental technique in artificial intelligence.^[5] Barto and Sutton used Markov decision processes (MDPs) as a mathematical foundation to explain how agents (algorithmic entities) make decisions in a stochastic or random environment, receiving rewards at the end of each action. While traditional MDP theory assumed that agents knew all information about the MDPs in their attempt to maximize cumulative rewards, Barto and Sutton's reinforcement learning techniques allowed both the environment and rewards to be unknown, enabling this category of algorithms to be applied to a wide array of problems.^[9] Barto established a laboratory at UMass Amherst to develop ideas on reinforcement learning, while Sutton returned to Canada.

Barto and Sutton have been widely credited and accepted as pioneers of modern reinforcement learning, with the technique considered foundational to the modern AI boom.^[10] The reinforcement learning framework developed within academic circles has transcended academia, finding practical applications across various sectors.^[5,4,10] In gaming, one of its first major real-world achievements saw DeepMind's AlphaGo program defeat the prevailing human Go champion.^[5,4,10] In robotics, systems that learn motor skills such as object manipulation rely on Barto's algorithms, including Boston Dynamics' Atlas and OpenAI's Dactyl, while companies such as Tesla and Waymo apply reinforcement learning for autonomous driving. The framework also finds optimization applications in network control, chip design, internet advertising and recommendation systems on YouTube, Netflix, and Amazon, global supply chains, and improving chatbot reasoning capacity. In conversational AI, systems like ChatGPT, Grok, and Gemini use reinforcement learning from human feedback (RLHF) to improve their responses.

In a 2025 interview, Barto emphasized that achieving real intelligence requires AI to learn by doing through trial and error rather than from curated data, a philosophy promoting adaptive systems mimicking human learning across the technology, healthcare, and finance sectors. The integration between reinforcement learning and other AI techniques is evolving rapidly, particularly where other methods help reinforcement learning build world representations for more efficient exploration, as seen in the linguistic domain where it complements pretrained large language models.^[11] An evolutionary perspective underpins Barto's thinking on multi-criterion reinforcement learning, where systems respond to multiple reward signals rather than a single function, reflecting how different regions of the human brain process diverse forms of feedback. Regarding AI safety and alignment, Barto acknowledged the difficulty of ensuring systems act according to human values, warning that as an optimization technology, reinforcement learning may cause agents to discover unexpected, unsafe ways to achieve rewards that system designers never anticipated.

Barto has published more than 100 papers or chapters across journals, books, and conference and workshop proceedings. With Richard Sutton, he co-authored the book Reinforcement Learning: An Introduction (MIT Press 1998, 2nd edition 2018); he is also identified in most sources as co-editor of the Handbook of Learning and Approximate Dynamic Programming (Wiley-IEEE Press, 2004) alongside Jennie Si, Warren Powell, and Don Wunsch II, though Russian Wikipedia describes him as having co-authored the handbook with Sutton.^[4]

He received the IEEE Neural Network (or Neural Networks) Society Pioneer Award in 2004 and the University of Massachusetts Amherst Neuroscience Lifetime Achievement Award (also referred to as the Neuroscience Achievement Award) in 2019.^[13,14] In 2017, he received the IJCAI Award for Research Excellence for his contributions to reinforcement learning, with his citation recognizing him for his foundational, groundbreaking, pioneering, and impactful research in both the theory and application of reinforcement learning.^{[13,8,14,3,9,24,4]}

Sources disagree on whether it was in 2024 or 2025 that he received the Turing Award from the Association for Computing Machinery alongside his former doctoral student Richard S. Sutton for their work on reinforcement learning.^[8,4,5,14] The citation for the award read: "For developing the conceptual and algorithmic foundations of reinforcement learning."^{[8,14,3,9,24]} The award was endowed with one million USD.

The scientist has published over a hundred articles and chapters in journals, books, as well as in conference and workshop proceedings. Together with Richard Sutton, he co-authored the 322-page book Reinforcement Learning: An Introduction, published in 1998 by The MIT Press (ISBN 9780262193986).^[10] He is also a co-editor of the Handbook of Learning and Approximate Dynamic Programming (2004) alongside Jennie Si, Warren Powell, and Don Wunsch II.^[10]

CMIgrCgAAAAJ. http://www-all.cs.umass.edu/~barto/