From Prediction to Understanding: Will AI Foundation Models Transform Brain Science?

2025-09-21

AI Summary

Current AI foundation models in neuroscience are limited to predictive modeling of neural activity, lacking interpretability and mechanistic insight into brain computation and cognition. Method: We propose a novel generative pretraining paradigm integrating large-scale unsupervised learning, cross-domain transfer, and deep neural architectures to develop a multitask foundation model for neuroscience. Contribution/Results: The model achieves high-fidelity neural activity prediction while establishing, for the first time, systematic, interpretable mappings between its internal representations and both neural coding principles and cognitive functions. It provides a candidate unified computational framework for neuroscience and clarifies key challenges and viable pathways for interpretable modeling. This work marks a paradigm shift from data-driven prediction toward mechanism-driven explanation in computational neuroscience.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Generative pretraining (the "GPT" in ChatGPT) enables language models to learn from vast amounts of internet text without human supervision. This approach has driven breakthroughs across AI by allowing deep neural networks to learn from massive, unstructured datasets. We use the term foundation models to refer to large pretrained systems that can be adapted to a wide range of tasks within and across domains, and these models are increasingly applied beyond language to the brain sciences. These models achieve strong predictive accuracy, raising hopes that they might illuminate computational principles. But predictive success alone does not guarantee scientific understanding. Here, we outline how foundation models can be productively integrated into the brain sciences, highlighting both their promise and their limitations. The central challenge is to move from prediction to explanation: linking model computations to mechanisms underlying neural activity and cognition.

Problem

• Moving from predictive accuracy to scientific understanding in brain science
• Linking AI model computations to neural activity mechanisms and cognition
• Productively integrating foundation models while addressing their limitations

Innovation

• Using generative pretraining for unsupervised learning
• Adapting foundation models across multiple scientific domains
• Linking model computations to neural mechanisms