Sunday, 1 December 2013

The limits of evolutionary theory

The quest for the edge of evolutionary theory has been pursued by many. Often they conclude that morality - or some other aspect of human nature is "beyond the edge of evolution".

My search for the limits of Darwinism is a bit different. It's obvious that Darwin's theory of evolution covers all living things. However, it is now also now clear that the same theory applies to a range of inorganic systems - drainage systems, propagating cracks, electrical discharges - and so on. I discuss this in more detail in various places - including:

So: what are the limits of evolutionary theory? One solution is to have a theory that covers all change. I've briefly discussed this possibility in:

Such a complete theory of evolution has also been proposed by Nathalie Gontier. However, notoriously a theory that predicts anything also predicts nothing. This is not necessarily a fatal flaw - Turing's theory of computers applied to the universe in the form of the Church, Turing, Deutsch principle also predicts nothing. However, not making concrete predictions is not a very positive feature. This motivates the search for more easily falsifiable evolutionary theories.

Evolutionary theory usually involves a reproducing population undergoing mutations and subject to selection. However, this leaves open the issue of what counts as a "population", what a "mutation" is and what qualifies as "selection".


One classical way of constraining the predictions of evolutionary theory is to insist that mutation occurs at random. Obviously unconstrained variation can't be permitted - since otherwise you can explain any form of change by saying: "that's just a macromutation". Random mutations represents the opposite extreme.

The problem with the "random mutations" solution is lack of realism. Mutations aren't really random. Variation isn't really undirected with respect to fitness. The idea may be a pillar of neo-Darwinism, but it's just a confused mistake.

I think that Donald Campbell nailed the nature of the constraints on variation in 1974 - by saying that they depended on existing "knowledge" - where "knowledge" is intended in a broad sense that includes the accumulated "wisdom" stored in organisms' genomes.

Donald Campbell's solution might have been technically correct, but it isn't very neat. To apply it one must know what counts as "existing knowledge" - which is a non-trivial problem. Modelers typically tend to stick with random mutations - since these are more tractable.


A population is a clearly defined concept in the case of sexually-reproducing organisms. In most other cases it isn't so clear. I generally advocate that a broad conception of what counts as a population be used: any collection of one or more entities.


Selection takes the form of production and elimination. Entities are produced and eliminated. Production and elimination are a bit like creation and destruction - except that production could take the place of revealing an existing thing - while elimination could just hide something - rather than destroying it.


A clue about the domain of universal Darwinism is that it appears to cover the same domain as maximum entropy thermodynamics - whose application domain is fairly well established.

Another clue is that it appears to underpin all goal-directed systems and all adaptations.

Most systems that are macroscopically reversible are not usefully modeled by universal Darwinism.


The options for falsifying theories based on universal Darwinism are mostly based on evolution taking place too fast - and insufficiently gradually. Insisting on random (or undirected) mutations is the classical way of doing this. However, even if you relax this notion - and only insist that the mutations are based on existing knowledge - that still provides reasonable possibilities for falsification.

No comments:

Post a comment