Davies, Adam Paul. “ELECTRONICS AND COMPUTER SCIENCE.” (2014).
URL1 URL2
Biological evolution contains a general trend of increasing complexity of the most complex organisms. But artificial evolution experiments based on the mechanisms described in the current theory generally fail to reproduce this trend; instead, they commonly show systematic trends of complexity minimisation. In this dissertation we seek evolutionary mechanisms that can explain these apparently conflicting observations. To achieve this we use a reverse engineering approach by building computational simulations of evolution. One highlighted problem is that even if complexity is beneficial, evolutionary simulations struggle with apparent roadblocks that prevent them from scaling to complexity. Another is that even without roadblocks, it is not clear what drives evolution to become more complex at all. With respect to the former, a key roadblock is how to evolve ‘irreducibly complex’ or ‘nondecomposable’ functions. Evidence from biological evolution suggests a common way to achieve this is by combining existing functions – termed ‘tinkering’ or ‘building block evolution’. But in simulation this approach generally fails to scale across multiple levels of organisation in a recursive manner. We provide a model that identifies the problem hindering recursive evolution as increasing ‘burden’ in the form of ‘internal selection’ as joined functions become more complex. We show how having an ontological development process that occurs by local growth, as present in most complex biological organisms, resolves this problem, enabling evolution to occur recursively. Meanwhile, to understand what drives complexity in evolution we provide a model showing that under certain conditions a well-studied concept from the computational study of algorithms – complexity lower bounds –applies in evolution.