13 December 2024
Richard Watson
Institute for Life Sciences & Department of Computer Science
University of Southampton, UK
There are two well-known sources of adaptation in biological systems. One is evolution by natural selection and the other is learning. Learning is often assumed to require specialised cognitive machinery, such as neurons and brains. However, some capacity for learning appears to be present wherever it is looked for in the tree of life, including plants and fungi, and single-celled organisms, even bacteria. Here we consider the possibility that cognition is coextensive with life because some ability to integrate information and coordinate collective action is possible without any specialised machinery that is evolved for this purpose. We study physical dynamical systems described by networks of viscoelastic connections (connections that give way slightly under stress). We show that such systems can form memories (i.e. dynamical attractors that cause the system to revisit past states), show associative learning with generalisation (i.e. form behaviours that accumulate information from multiple past experiences and generate novel patterns from the same distribution) and adapt or improve in their ability to problem-solve with experience (i.e. to find configurations that are exceptionally effective at resolving multiple simultaneous conflicting constraints). These behaviours can be explained by recognising that associative learning, familiar in the context of neural networks, results from ordinary physical energy minimisation when it affects connections that control the relationships between variables rather than controlling variables directly. This kind of behaviour is therefore shown to be possible in a system as simple as a set of masses connected by springs, without any special arrangement, and without any natural selection involved. Since dynamical systems described by networks of interactions are ubiquitous in biology, and physical interactions usually give-way under stress (i.e. are rarely perfectly elastic), we suggest that this might help explain how basal cognition is coextensive with life. If true, this suggests that living systems, even systems that are not evolutionary units and therefore not adapted by natural selection (e.g. at the origin of life prior to the onset of evolution, and in current ecosystems of numerous kinds), can nonetheless exhibit adaptive organisation and behaviour. If natural selection is not required for the presence of adaptative organisation, many assumptions and arguments about what is and is not possible in biology would require revision.