Songs of Life and Mind

I use 'song' to mean a dynamical system composed of coupled periodic behaviours in harmonic relationships in any substrate. These dynamics recreate the conditions for their own behaviour and organisation (otherwise they wouldnt be periodic) and they have a nested or recursive organisation of cycles within cycles (which is where the harmonic relationships become important). With this framework I aim to unify development, evolution and cognition using the principles of resonance and the language of harmony.

 

A song is not just a metaphor or a conceptual construct: A song has many intrinsic properties that are relevant:

​

• A song is substrate independent - same tune, different instruments. For example, gene-regulatory dynamics, neural dynamics, bioelectric networks, cell-cell morphogen signalling, physiological tissues in a particular shape and form - or any combination - can all hold the same tune.

• A song is scale invariant (same song in different frequencies and wavelengths or physical sizes). High frequencies control small physical scales and large frequencies control large physical scales (for a given wave propagation speed).

• A song is self-sustaining. Its a cyclic attractor recreating its own necessary conditions.

• A song is information carrying. Harmonic relations have linguistic properties (compositionality, systematicity and productivity). And they are error-correcting dynamics (harmonic relations describe specific possibilities with big gaps inbetween these 'legal forms'. (More exactly, discordant harmonic relationships are only 'legal', curves of equal curvature all over, in higher-dimensional spaces that require more energy).

• Songs are physically causal. Resonance is a physical force that organises matter (e.g. see Chladni plates, Faraday worms, Bartons pendulums).

• Songs self-reproduce. Via resonance and embedding principles a song can be transferred or copied from one system to another (suitably irritable) system.

​

Songs are therefore legitmate as information-laden dynamical processes, or 'programs' that organise matter. They are abstract enough to be multiply realisable, and concrete enough that they enact physical forces on matter. They constitute a two way relationship between form and function. 

​

So thats all very interesting... we can imagine a complex song with enough information in it to organise matter into an organism (well, I can), but...

​

... where do songs come from?

​

My conjecture is that songs have cognitive properties (memory, learning and intelligence) - or rather, the intrinsic nature of cognition is a song.

​

• Memory: Songs are dynamical systems that can take a memory, imprint or engram from their environment. In general, its not difficult for a physical system to have memory in a simple sense - a lump of clay has a memory. Whats more interesting is to be able to remember more than one thing (see Natural Induction). Songs are sensitive to transformation through contact with periodic stimuli/other songs. They can take engrams that capture deep structural regularites in temporal and structural organisation. The basic mechanism is resonance and phase locking, and each frequency component of a song can be affected (pseudo-)independently.

​​

• Learning: The difference between being able to remember multiple things and learning is subtle. If you store and recall multiple things in a shallow way (like many leaves of a clay book) this captures no underlying structural regularities. But if the multiple imprints are compressed into a lower-dimensional space (like an auto-encoder with a compressed latent space) then deep structural regularities must be learned, not superficial details. This gives the memory generalisation capabilities - thats the difference between 'remembering a lot' and 'learning'. Songs get this 'for free' from the under-determination of sub-harmnic induction locking. All that means is that when f phase locks with 2f (the octave relation), the sign information is lost - the resultant phase of f is the same if it locks with 2f or -2f. This is a compression - a loss of information - a folding of the feature space. As harmonic locking is squeezed up successive octaves, this results in additional compression or 'folds' of the original space. As the same information resonantes back down again, through the octaves in the other direction, the result is generative - producing many patterns from the same underlying structural model. 

​​

• Problem-solving: Problem solving is just good optimisation. Regular optimisation is just local energy minimisation (see Natural Induction, and Agency, part-wholes). When physical systems do good optimsation that means they do better than finding the nearest local minimum of their energy function, and instead find something better than that (Agency, part-wholes). This is only possible because there were hidden variables (hidden dimensions to phase space) that you were not observing - these are private variables or unobservables. When you roll downhill in the deep low-dimensional space, it looks like you are doing something smart in the shallow high-dimensional space (see Deep Optimisation). Thats what all intelligent behaviour is, and the 'programming' of the hidden variables comes from learning that generalises well. In songs, phase locking between two deep models (low frequencies) is just a downhill energy-minimisation process in a compressed space. But the effect of this boring behaviour on the decompressed, high-dimensional space is not just the obvious rolling downhill behaviour you would expect (if you didnt know about the hidden state) (see Inner alignment model). But how did this hidden state (in the slow frequencies) know how to modify the behaviour of the observable state (in the fast frequencies) so that it looks intelligent!? Ans: Intelligent action is just 'oh, this again!' in a deep representation. The only way that anything ever looks intelligent is because its seen that situation before - it is part of a repeating pattern. Its not really anticipating or acting with foresight, its just in a cycle that repeats - and the right amount of late is the same as being early (like the wagon wheels in the strobe of the TV appear to spin the other direction). When you combine lots of 'oh this again' together at different levels of temporal compression, the result is intelligent action that acticipates future consequences (=agency) with the ability to improvise/generalise to avoid local minima and find better solutions (= intelligent problem solving).

​​

All of this is not selected or designed for the purpose of these outcomes - it all happens for free. When you put oscillators together - with strong coupling you immediately get one oscillator (boring). When they are weakly coupled - the only coherent result possible is a cycle, or a cycle with sub-cycles in harmonic relationships. It cant be otherwise. And all of the rest follows from that. Songs are cognitive spontaneously. They are self-reproducing, substrate-independent programs that learn deep knowledge about their environment from past experience and deploy it to organise their parts into intelligent action that anticipates future interaction with their environment. Natural selection is not required (see Natural Induction as a place to get started).

​

Its not so much that songs can do this clever thing like you, a 'real agent', do; A better way to look at is that you are a song. You are a pseudo-cyclic pattern of activity, written into and enacted by every aspect of your material fabric, at all organisational and temporal scales by past experience. Thats what makes you intelligent.... So sing your heart out!

​

Living things and cognition (in the 'basal cognition' sense) are coextensive (see Design for an Individual). Cognition is a special thing that only some living things have - all living things are intrinsically cognitive (exhibiting information integration and collective action, c.f. Levin). Information integration is the upward pass (high frequencies feed phase information up to lower frequencies) and collective action is the downward pass (in the other direction).

​

This enables the possibility that cognition comes first (not natural selection) - Cognition is the natural spontaneous result of self-organised active matter. It could therefore be the driver of adaptive change, orchestrating the activity and evolution of the genes, not the other way around. For sure, genetic changes still occur at random (random wrt some suitably ignorant level of description), and some changes survive and reproduce and others dont - natural selection is not wrong. But the selective field in which these mutations occur is not random. The conventional view is the selective conditions are external to the organism, acting on it (and its mutations) from outside. But actually, most of the selective pressure that mutations experience is entirely contingent on the behaviour of the organism - it is internally regulated (not externally regulated) and its is repeated through sufficiently many generations for genetic evolution to follow it.

​

By analogy, the grains of sand on the Chladni plate bounce irratically, but the pattern they produce is organised by the song the plate is singing - the frequency of its resonating activity and its fit with the geometry of the plate as a whole. In the same way, the song an organism sings - the harmonic ratios of its genetic, bioelectric and physiological activity, and their fit with the agential activity of the organism as a whole - organises the selective field that genetic mutations experience. The organism can therefore be a legitimate leader, and genetic change can be the follower (Uller & Helanterä).

​

This creates space for living things to be agents in the process of creation, not just products of it. And that, in turn, gives some respite to the crisis of meaning created by reductionism.

​

  

• Uller, T., & Helanterä, H. (2011). When are genes ‘leaders’ or ‘followers’ in evolution?. Trends in Ecology & Evolution, 26(9), 435-436.

• Buckley, C. L., Lewens, T., Levin, M., Millidge, B., Tschantz, A., & Watson, R. A. (2024). Natural Induction: Spontaneous adaptive organisation without natural selection. bioRxiv.

• Watson, R.A., Levin, M. & Buckley C.L. (2022)"Design for an Individual: Connectionist Approaches to the Evolutionary Transitions in
  Individuality", Frontiers in Ecology and Evolution (doi: 10.3389/fevo.2022.823588)

• Watson, R.A. & Levin, M. (2023) “The Collective Intelligence of Evolution and Development”, Collective Intelligence.

• Watson, R. (2023). Agency, Goal-Directed Behavior, and Part-Whole Relationships in Biological Systems. Biological Theory, 1-15.

• Tissot, T., Levin, M., Buckley, C. L., & Watson, R. A. (2024). An ability to respond begins with inner alignment: How phase synchronisation effects transitions to higher levels of agency. bioRxiv.

• Caldwell, J. R., Watson, R. A., Thies, C., & Knowles, J. D. (2018). Deep optimisation: Solving combinatorial optimisation problems using deep neural networks. arXiv preprint arXiv:1811.00784.

• Caldwell, J., Knowles, J., Thies, C., Kubacki, F., & Watson, R. (2022). Deep Optimisation: Transitioning the Scale of Evolutionary Search by Inducing and Searching in Deep Representations. SN Computer Science, 3(3), 253.

​