Our Perspective on the cybernetic future of large-scale model-centric integrative neuroscience is now on arXiv. Predictive accuracy is not understanding: a Transformer-style foundation model that captures neural time series at scale does not, by itself, explain how behavior arises from the closed-loop coupling between brain, body, and environment. We argue for redirecting the model class itself toward integrative neurocybernetic models – closed-loop, embodied, dynamically structured, and organized around a controller pursuing latent objectives, with internal dynamics that correspond to biologically identifiable components.

The piece grew out of the Neurocybernetics at Scale symposium (Champalimaud Foundation, October 2025), and synthesizes a community effort around four desiderata – understandability, neurocybernetic structure, structured variation across scales, and scalability – and a program of complementary building blocks (nonlinear state-space models with meta-dynamical extensions, scalable inference, knowledge distillation, mixed open/closed-loop training, and connectomics-informed architectures).

Park, I. M., Vermani, A., de Polavieja, G. G., Gallego, J. Á., Esfahany, K., Saxena, S., Orger, M., Ijspeert, A., Dowling, M., McNamee, D., Turaga, S. C., Mainen, Z., Paton, J. J., and Renart, A. (2026). Integrative Neurocybernetic Modeling in the Era of Large-Scale Neuroscience. http://arxiv.org/abs/2604.23903

Check it out!