Discovering cognitive strategies with tiny recurrent neural networks

2025/07/02 by Jian Li, Marcus K. Benna, Marcelo G. Mattar · 2 voices · 37 citations
Biochemistry, Genetics and Molecular Biology · Neuroscience · Psychology · #Artificial intelligence #Artificial neural network #Cognition #Cognitive science #Computer science #Curse of dimensionality #Inference #Machine learning #Neural and Behavioral Psychology Studies #Neural dynamics and brain function #Normative #Psychology #Receptor Mechanisms and Signaling #Reinforcement learning

paper · pdf · doi:10.1038/s41586-025-09142-4

published in Nature 644(8078), 993-1001 (Nature Portfolio)

openalex publication_date 2025/07/02 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/06

Abstract

Abstract Understanding how animals and humans learn from experience to make adaptive decisions is a fundamental goal of neuroscience and psychology. Normative modelling frameworks such as Bayesian inference 1 and reinforcement learning 2 provide valuable insights into the principles governing adaptive behaviour. However, the simplicity of these frameworks often limits their ability to capture realistic biological behaviour, leading to cycles of handcrafted adjustments that are prone to researcher subjectivity. Here we present a novel modelling approach that leverages recurrent neural networks to discover the cognitive algorithms governing biological decision-making. We show that neural networks with just one to four units often outperform classical cognitive models and match larger neural networks in predicting the choices of individual animals and humans, across six well-studied reward-learning tasks. Critically, we can interpret the trained networks using dynamical systems concepts, enabling a unified comparison of cognitive models and revealing detailed mechanisms underlying choice behaviour. Our approach also estimates the dimensionality of behaviour 3 and offers insights into algorithms learned by meta-reinforcement learning artificial intelligence agents. Overall, we present a systematic approach for discovering interpretable cognitive strategies in decision-making, offering insights into neural mechanisms and a foundation for studying healthy and dysfunctional cognition.

Citations

• Interpreting neural computations by examining intrinsic and embedding dimensionality of neural activity
• Estimating the Dimension of a Model

Cited by

• A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems
• Fast dynamical similarity analysis
• Setting up for failure: automatic discovery of the neural mechanisms of cognitive errors
• From Pixels to Feelings: Aligning MLLMs with Human Cognitive Perception of Images
• When Can AI Models Explain Learning? Validity Criteria for AI as Cognitive Models in Education
• A Group Theoretic Analysis of the Symmetries Underlying Base Addition and Their Learnability by Neural Networks
• Large Language Models as Psychological Simulators: A Methodological Guide
• ATLAS: Active Theory Learning for Automated Science
• Hybrid neural-cognitive models reveal how memory shapes human reward learning. [europepmc]
• Digital Twin Brain: Generating Multitask Behavior from Connectomes for Personalized Therapy. [europepmc]
• Inferring latent behavioral strategy from the representational geometry of prefrontal cortex activity. [europepmc]
• Contextual inference through flexible integration of environmental features and behavioural outcomes. [europepmc]
• AI-driven complex systems redefine cognitive science. [europepmc]
• Low dimensional latent structure underlying the choices of mice [europepmc]
• Evolving learning state reactivation and value encoding neural dynamics in multi-step planning [europepmc]
• Environmental dynamics shape human learning: change points versus random walks [europepmc]
• Dopamine drives a positive reward bias on human reinforcement learning [europepmc]
• Integrated multiobjective optimization of RFSSW parameters for AA2024-T3 using ANOVA machine learning and NSGA II. [europepmc]
• Designing Neural Dynamics: From Digital Twin Modeling to Regeneration. [europepmc]

Discussions

• Thrilled to see our TinyRNN paper in @nature! We show how tiny RNNs predict choices of individual subjects accurately while staying fully interpretable. This approach can transform how we model cognit [bsky, 329 points, 8 comments]
• In the next #BrAInWorldsJournalClub, Georgia Koppe will present the work of @ji-an-li.bsky.social, Marcus Benna & Marcelo Mattar @marcelomattar.bsky.social 🧠in Nature, 2025: doi.org/10.1038/s41586- [bsky, 3 points, 0 comments]

Related

• Neural Basis of Reinforcement Learning and Decision Making
• Reinforcement Learning and its Connections with Neuroscience and Psychology
• Neural Mechanisms of Human Decision-Making
• Neural Dynamics Model of Visual Decision-Making: Learning from Human Experts
• What Neuroscience Can Teach AI About Learning in Continuously Changing Environments