Integrating artificial and biological neural networks to improve animal task performance using deep reinforcement learning

Artificial neural networks have performed remarkable feats in various domains but lack the flexibility and generalization power of biological neural networks. Given their different capabilities, it would be advantageous to build systems where both network types can synergistically interact. As proof-of-principle, we show how to create such a hybrid system and harness it to improve animal performance on biologically relevant tasks. Using optogenetics, we interfaced the nervous system of the nematode Caenorhabditis elegans with a deep reinforcement learning agent, enabling the animal to navigate to targets and enhancing its food search ability. Agents adapted to strikingly different sites of neural integration and learned site-specific activations to improve performance on a target-finding task. The animal plus agent displayed cooperative computation and generalized to novel environments. This work constitutes a demonstration of how to improve task performance in animals using artificial intelligence interfaced with a nervous system.