The proposed theory of Survival of the Fitted explains biological evolution based on sustainability of multi-scale interaction networks. Winning in struggles, and reproductive advantages as associated with Darwinian Survival of the Fittest and Natural Selection is secondary. We define a species as the collective of organisms that share a set of core, sustaining interactions, termed the species interaction code. Interaction codes evolve in response to genetic, environmental, and other changes, in a process that relates to autoencoding in machine learning. This unique process, termed natural autoencoding, every biological interaction in the biosphere aids in multiple functions: encoding existing organisms and interactions; decoding such codes into ecosystems, and reacting to changes by adjusting interaction codes. DNA is only one element in this mechanism.

Natural autoencoding differs from typical artificial autoencoding:  1. Evolving codes stabilize around de-facto sustained interaction patterns, without employing a loss function or optimization; This is retrospective, what works works. 2. Natural autoencoding concurrently generates many species interaction codes, at multiple scales, including biochemical, cellular, and organism. 3. The input to the natural autoencoder is the entire biosphere, and thus includes the autoencoder itself.

Our group is now designing computer-based autoencoders to model natural autoencoding.

Principles of Survival of the Fittest have significantly affected human political, social, economical, and educational norms. Survival-of-the-fitted by natural autoencoding explains why a habitable environment requires diverse networked interactions. This perspective can help drive much needed changes in human behavior, both environmental and social, including the kind of social responsibility advocated by Moshe Vardi.

Acknowledgement: We thank D. Harel, S. Szekely and G. Frankel for valuable contributions.