Our brains are remarkable examples of learning systems. But learning is a potentially broader metaphor, applicable to physical and biological phenomena that may not involve neurons at all. This program puts forward the nontrivial hypothesis that learning can be a useful framework to organize questions about a broad range of systems that acquire functional behaviors by accumulating incremental changes due to environmental stimuli over their history. These examples range from reconfiguration of vasculature networks in slime mold and changed behaviors in ciliates and other single celled organisms in response to structured environmental stimuli, to the generation of broadly neutralizing antibodies in the adaptive immune system. Minimal systems that have provided insight into possible mechanisms for such `physical learning’ include mechanical and molecular systems that learn to deploy specific elastic, phase separation or self-assembly behaviors. In these and other examples, systems adapt locally but confer global function.

Despite the recent proliferation of examples, the unifying key principles of such adaptive processes have not yet been distilled. In parallel, the theory of learning is undergoing an accelerated development propelled by machine learning. This program will bring together physicists, biologists and neuroscientists to refine key questions–what kinds, or classes, of natural local processes in different physical and biological systems allow them to learn? What kinds of statistical structure in environmental stimuli can be learned by a given physical system? What physical properties and architectures allow for learning more complex internal models of complex environments (expressivity)?