Friday 27 June 2014

Tim Tyler: Darwinian physics

Transcript:

Hi. I'm Tim Tyler and this is a video about Darwinian physics.

In the 1980s there was a movement to incorporate elements from physics into evolutionary theory. It was claimed that not all of the complex-looking patterns of organisms were there in order to benefit the genes of their owners. Instead, some of them were the product of self-organizing systems. Reaction-diffusion patterns were invoked to explain zebra stripes and cheetah spots - challenging the idea that these were adaptations whose purpose was to improve camouflage. Cellular automata were found that reproduced the decorative patterns found on some seashells. The branching tree-like pattern that is found everywhere in biology was likened to dendritic frost patterns, electrical discharges and propagating cracks. The search was on for the new rules that governed these self-organizing systems.

Now, we have found many of these rules. A big and important one has turned out to be: Darwinian evolution via natural selection. So, now the time is ripe for Darwinism to claw back some of the explanatory territory that was lost to complexity theory - and it is time for it to perform some invasions of its own - into realms that have been traditionally occupied by physics and chemistry.

Darwinism is based on processes involving copying with variation and selection. It was once widely thought that copying of the required kind was unique to biological systems - and that before the origin of life copying fidelity was too poor to support any kind of Darwinian evolution - with any copying processes producing instead an error catastrophe and a mutational meltdown. This argument was made by A. G. Cairns-Smith in 1982 - for example. However, this idea has turned out to be dramatically wrong. It turns out that the high-fidelity copying suitable for supporting cumulative adaptive evolution is ubiquitous in simple physical systems.

Lightning strikes are one of the easiest examples of Darwinian physics to visualize. Lightning makes a branching tree. This is a family tree - with regions nearer to the trunk being the ancestors of regions nearer to the tips. If you ask what information is transmitted from parent to child in this case, the answer is that the offspring inherit their parent's position. They inherit this extremely precisely - down to the nearest millimeter.

Position isn't the only thing that is inherited in this way, other attributes - such as velocity and chemical composition - are inherited too, although this inheritance is often less precise.

Examples of Darwinism in simple physical systems are ubiquitous. Most tree-shaped structures are the product of copying with variation and selection. This includes many crystals, fractal drainage basins, propagating cracks and electrical discharges. Re-radiation of photons after hitting dust particles is one of the most common events in the universe and it follows the classical Darwinian algorithm - producing a family tree of photons with the older photons with higher energies being ancestral to their more numerous, but less energetic offspring.

Many other cases do not so obviously exhibit a simple tree. For example, raindrops divide, but they also join creating a network - rather than a classical phylogenetic tree. Of course there are also broadly similar examples of joining in classical evolutionary biology: gametes fuse together, parasites inject DNA into hosts, bacteria assimilate other bacteria and lineages join forces in symbosis. Evolution involves joins and mergers as well as splitting and subdivision.

In other cases, the phylogenetic tree is only visible if you take a historical perspective. For example, every asteroid has an associated ancestral tree of other asteroids, moons, planets and stars that it was formed from. However, you don't see the tree by looking at the asteroid. Only if you look back in time does the family tree become visible.

It's well known that organisms can be modeled as optimizing a fitness function. The same thing applies to lightning - it too is optimizing a function. It acts as a crack in space the seeks out the path of least resistance in its search for the ground. The fact that organic evolution acts as an optimization process is widely exploited by genetic algorithms. There, humans allocate a fitness function and let simulated evolution do their optimization work for them. Optimization features maximands - and these give evolution an apparent teleological character. Biologists can ask what wings are "for" - and then give a reasonable-sounding functional answer. Darwinian physics brings the concept of adaptation to the evolution of inorganic systems - and it becomes reasonable to ask what their features are "for".

Ultimately, both living organisms and lightning strikes can be modeled in the same optimization framework using the same maximand - by treating them both as maximizing entropy.

Entropy maximization is a formulation of optimization processes from within physics which some physicists are already familiar with. Some might ask what the advantages are of taking a Darwinian perspective over one based on entropy maximization. This is a complex question which risks going beyond the scope of this video, but briefly, entropy maximization and Darwinian evolution are largely equivalent ideas which make most of the same predictions - since nature's maximand is closely correlated with entropy production. These are mostly two different ways of looking at the same phenomenon. However, it is possible to select locally for other things apart from entropy maximization - including entropy minimization. Path dependency within a Darwinian framework provides a context which helps to explain these apparent deviations from maximum entropy production.

Lastly, Darwinism neatly explains observation selection effects. While these have long been considered to be part of physics, they are obviously a case of selection acting on observations. We can reformulate the concept of survival of the fittest as observation of the observable. This more general concept acts as a grand unified theory of selection - which neatly covers both Darwinian evolution in biology and observer selection effects in physics and many other areas of science.

Over the last two centuries, Darwinism has had a revolutionary impact on biology, leaving most of the theories that preceded it in the dust. Physics and chemistry are relatively new and unexplored frontiers for Darwinian explanations - but it seems clear that incorporating Darwinism into physics will involve some rewriting of textbooks, and a fair number of psychological paradigm shifts, as scientists gradually awaken to the idea.

Historically, Darwinism has had a hard time penetrating areas that go beyond inheritance of nucleic acids. For example, cultural evolution is still a topic mired in confusion and controversy. However, physicists are supposed to be intelligent people. Hopefully they will prove to be more open to Darwin's ideas than many social scientists have been - and less inclined to treat the Darwin enthusiasts as hostile invaders.

Enjoy,

References

1 comment:

  1. Hi Tim,

    Good to see you are thinking about the important topic of Darwinism in physics.

    I recently came across a paper by Gerard 't Hooft, the Nobel Prize winner who amongst many other accomplishments introduced the Holographic Principle to physics. This monumental 207 page paper is title 'The Cellular Automaton Interpretation of Quantum Mechanics' and summarizes 't Hooft's research program on this subject over the past couple of decades. (http://arxiv.org/abs/1405.1548)

    While his theory is not explicitly Darwinian he considers that fundamentally Quantum physics takes place near the Planck scale where bits of information are confined to Planckian areas and may interact only with their nearest neighbors in a fully causal and deterministic manner. He considers standard quantum theory to be a statistical summary of these interactions. While information coded on an area that codes for physics within a volume may be a close analogy to biology's genotype/phenotype duality 't Hoofts theory requires a lot of development before it may be considered Darwinian.

    Lee Smolin's recent paper Temporal Naturalism (http://arxiv.org/abs/1310.8539) reads like a Darwinian manifesto and one gets the impression that Smolin's views have undergone a fundamental shift. He suggests that our conception of nature should be based on a model of evolution rather than on a mathematical model (quote):

    I propose we develop a conception of nature contrary to the mathematical universe, based on taking time and its passage through a succession of present moments to be real and fundamental

    Wojciech Zurek has extended Quantum Darwinism into the realm of quantum gravity. One of the most promising approaches to quantum gravity in the 90s was the theory of Decoherent Histories which worked with the wave function of the universe. Unfortunately it was found that the theory predicted a huge amount of things that didn't happen along with those that did. Zurek is attempting to use Quantum Darwinism as a selection mechanism (http://arxiv.org/abs/1312.0331). He has promised a further paper on this subject.

    I am currently working on a book titled 'Darwin does Physics' and given your interest in the subject was wondering if you would be like to read some sections for the purpose of discussion.

    In the Social Sciences I found Herbert Gintis' paper 'A framework for the unification of the behavioral sciences' to be of interest. Have you seen this paper and do you have any views on it?

    Cheers
    John Campbell


    ReplyDelete