Hosted by the Turing Institute

Abstract

The conventional account of intelligent life is that cognition requires specialised machinery, like brains and neurons, and that therefore cognition is rare, and a late arrival in the evolutionary story. However, this is not correct. It turns out cognition does not require brains or neurons, and in fact, wherever we look in the tree of life, even in single-celled bacteria, we find some ability for memory and recall of past states, learning, and adaptive problem solving. How did these abilities evolve?

Since the mechanistic substrates of cognition in animals, plants, fungi and bacteria are all different, it seems unlikely to be common descent, and to invoke convergent evolution on this scale, spanning the whole of the tree of life, is also unreasonable. No, cognition and life are coextensive. Cognition is precisely what makes the difference between things that are living - with agential capabilities - and things that are not. But still, how did it evolve?

We consider a radical hypothesis: that cognition came first, before natural selection, and continues to be the main driver in evolutionary change. In order for phenotypic adaptations to be a genuine leader of genetic evolution, two things are needed. 1) adaptation in phenotypes independent of natural selection 2) a way to transfer adaptations into the genome. To the first, the widespread belief that evolution by natural selection is the only possible natural mechanism that can produce adaptive organisation without design (a.k.a. 'universal Darwinism') is incorrect. Any dynamical system described by a network of viscoelastic connections will exhibit memory, learning and adaptive behaviour spontaneously. This does not require selection or design, and can be demonstrated in a system as simple as a random network of masses connected by springs. Viscoelasticity simply means that the connections give-way slightly under stress (as real springs do when extended too far or for too long).

This kind of spontaneous relaxation in the dynamical coefficients of the system is sufficient for the same kind of behaviours that are commonplace in neural networks. Its adaptive capabilities are superior to those of natural selection. Since life is full of viscoelastic networks, this means that living things can learn and adapt without being evolved for these functions. The second step can be explained with the Baldwin effect, or genetic assimilation, whereby the solutions already found by learning modify the selective field that applies to random mutations. Together this makes a cognition-first account of evolution possible, and explains why a conventional selection-first account remains consistent with data. That is, genetic evolution occurs, and obviously fit mutations are retained by selection. But genetic evolution is  not the source of adaptations; it is one of many memory substrates in which life stores solutions it has already learned.

I will present this new hypothesis for biological evolution and discuss the deep implications that an intelligent evolutionary process would raise.  

This is joint work with Michael Levin, Professor, Tufts University, USA