Writing Complex Information into the Brain

Recent advances in neuroengineering enable large-scale neural recording in animals and humans, which, when combined with AI, have produced brain - computer interfaces (BCIs) that decode motor intent or internal speech (“reading”). In contrast, precisely “writing” complex, behaviorally relevant information into the brain remains in its early days due to challenges in hardware, software, and basic neuroscience. Over the past decade, I have built (1) optical and genetic technologies for read/write BCIs (enabling cellular-resolution, bidirectional control over thousands of neurons), (2) interpretable, AI-powered dynamical systems frameworks to drive such BCIs (via data-driven system identification and optimal control), and (3) cognitive animal models to test and benchmark these tools (focusing on genetically tractable mice engaged in complex decision-making tasks). I will present how these advances have enabled the discovery of cell-type-specific and brain-wide dynamical structures underlying reward and memory computations, as well as early prototypes of foundation models for neural dynamics. Then I will outline concrete next steps toward closed-loop, AI-guided control of cognitive computations, thereby predictively steering animal behavior. I will conclude with a roadmap for generalizing this research program across contexts, scales, modalities, and species.

Speaker: YoungJu Jo, Stanford University