What does evolution make? Learning in living lineages and machines.

What is the origin of the complex, highly functional patterns of form and behavior observed in the living world? Specifically, how does the information in the genome unfold and give rise to self-constructing living organisms with problem-solving capacities implemented at all levels of organization? Here we review recent progress that unifies work in developmental genetics and machine learning, in a powerful way that goes beyond traditional mappings of genes to traits. We emphasize the discovery of deep symmetries between evolution and learning, which cast the genome as instantiating a generative model. The layer of physiological computations between genotype and phenotype provides a powerful degree of plasticity and robustness, not merely complexity and indirect mapping, which strongly impacts individual and evolutionary scale dynamics. Powerful ideas from machine learning and neuroscience now provide a versatile, quantitative formalism for understanding what evolution learns and how developmental and regenerative morphogenesis interpret the deep lessons of the past to solve new problems. This emerging understanding of the informational architecture of living material is poised to impact not only genetics and evolutionary developmental biology but also regenerative medicine and synthetic morphoengineering.