An AI-informed model of human reward-based learning: Hybrid approach could aid studies of mood disorders

image via Medical Xpress - medical research advances and health news

March 6, 2026, 5:00 PM

• •The study introduces a new approach for studying human reward-based learning using a combination of AI algorithms and psychological theories.
• •Researchers found that traditional RL approaches are less effective in predicting human decisions than models that also represent human memory and mental processes.
• •The researchers successfully bridged AI systems, including RL and neural networks, with ideas rooted in psychology.
• •The model could be improved further and adapted by other research groups to carry out behavioral science research.

Researchers at Google DeepMind and the University of Oxford have developed a new approach to studying human reward-based learning by combining AI algorithms and psychological theories. Their research, published in Nature Human Behavior, suggests that traditional reinforcement learning (RL) algorithms do not effectively replicate how human decisions are influenced by past experiences. The team created AI models that integrate conventional RL algorithms with flexible neural networks, which are trained to represent human mental processes. Their findings highlight the need for alternative models that better account for human internal representations, potentially offering new insights into disorders characterized by disruptions in reward-based learning.

Read original article