# Reinforcement Learning and its Connections with Neuroscience and Psychology

**Authors:** Ajay Subramanian, Sharad Chitlangia, Veeky Baths  
**arXiv:** [2007.01099](https://arxiv.org/abs/2007.01099)  
**Date:** 2020-06-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2007.01099v5/

## Key points

- Reinforcement learning (RL) methods have achieved success in complex sequential tasks such as playing Atari games, Go, and Poker.
- RL algorithms have surpassed human performance by learning from scratch and relying solely on scalar rewards obtained through interaction with the environment.
- Fundamental concepts in RL draw inspiration from animal learning, psychology, and neuroscience.
- The paper conducts an extensive review of findings in neuroscience and psychology to establish a comprehensive understanding of RL's potential for modeling learning and decision-making in the brain.
- Mapping modern RL algorithms to specific discoveries in neurophysiological and behavioral literature highlights the relationship between RL, neuroscience, and psychology.
- This mapping provides insights into how RL aligns with biological processes underlying learning and decision-making.
- The observed connection between RL, neuroscience, and psychology has implications for advancing research in both artificial intelligence (AI) and brain science.
- Bridging these disciplines allows researchers to enhance our understanding of how RL algorithms function within biological systems.
- The integration of insights from neuroscience and psychology enhances our understanding of RL and opens up new avenues for interdisciplinary research at the intersection of AI and brain science.

## Summary

In the paper titled "Reinforcement Learning and its Connections with Neuroscience and Psychology," authors Ajay Subramanian, Sharad Chitlangia, and Veeky Baths explore the success of reinforcement learning (RL) methods in complex sequential tasks such as playing Atari games, Go, and Poker. These algorithms have surpassed human performance by learning from scratch and relying solely on scalar rewards obtained through interaction with their environment. While there has been significant independent innovation in RL, many fundamental concepts in this field draw inspiration from animal learning, psychology, and neuroscience. To establish a comprehensive understanding of RL's potential for modeling learning and decision-making in the brain, the authors conduct an extensive review of findings in both neuroscience and psychology. By mapping various classes of modern RL algorithms to specific discoveries in neurophysiological and behavioral literature, the authors highlight the relationship between RL, neuroscience, and psychology. This mapping provides valuable insights into how RL aligns with biological processes underlying learning and decision-making. The paper also discusses the implications of this observed connection between RL, neuroscience, and psychology for advancing research in both artificial intelligence (AI) and brain science. By bridging these disciplines, researchers can leverage knowledge from each field to enhance our understanding of RL algorithms' functioning within biological systems. Overall, this paper contributes to the growing body of evidence supporting reinforcement learning as a promising approach for modeling cognitive processes in the brain. The integration of insights from neuroscience and psychology not only enhances our understanding of RL but also opens up new avenues for interdisciplinary research at the intersection of AI and brain science.