Brains and Machines

Martin Schrimpf

The last decade has seen the rise of AI models to explain the brain and mind. Mapped onto brain regions, these models predict neural activity within sensory cortices and also higher-level systems like language. To make progress on system models, we are developing the Brain-Score platform which, to date, hosts over 100 benchmarks of neural and behavioral experiments that models can be tested on. By systematically evaluating a wide variety of model candidates, we not only identify models beginning to match a range of brain data, but also discover key relationships: models' internal representations are more brain-aligned the higher their performance on object categorization in vision and their next-word prediction performance in language. Using the integrative benchmarks, we develop improved state-of-the-art system models that more closely match neuroanatomy and internal function, and thereby yield benefits for machine learning such as improved robustness. Our latest generation of models can further be used to predict the behavioral effects of neural interventions such as micro-stimulation, and can drive new experiments. By fostering integrated models of brain and behavior, NeuroAI will not only deepen our understanding of the neural mechanisms underlying cognition but also potentially transform clinical approaches to brain-related disorders.