Evolvability and Artificial Evolutionary Systems

[Tim Taylor's home page] [Tim Taylor's research page]

[Niche Construction] [EvoCA] [NKPM Model] [Other Work]

I am also interested in theoretical questions of evolvability, and, in particular, of how to increase the evolutionary potential of artificial evolution systems. My doctoral research was an investigation of the evolutionary properties of a system of self-replicating software agents, and an analysis of how the evolutionary potential of such systems might be improved.

My PhD thesis, "From Artificial Evolution to Artificial Life", can be viewed in HTML or downloaded in a variety of formats.

I am currently working on a number of other projects by myself (when I can find the time to fit them in). These include:

Niche Construction

I have developed an individual-based model of niche construction, whereby organisms disturb the environment experienced by their neighbours. This disturbance in local conditions creates a niche that potentially could be filled by another species (which would then create still more niches and so on). The model is unique in allowing the complexity of the organisms---measured by the number of genes they possess in order to be well adapted to their local environment---to evolve over time, and is therefore the first model with which it is possible to study the contribution of niche construction to the evolution of organism complexity. Results of experiments demonstrate that the process of niche construction does indeed introduce an active drive for organisms with more genes. This is the first explicit example of a model which possesses an intrinsic drive for the evolution of complexity.

A description of the model and results of initial experiments appear in the paper "Niche Construction and the Evolution of Complexity". Further, more substantial, publications will be appearing soon.


A model of evolution in cellular automata (CA), based upon Howard Pattee's ideas on the emergence and role of symbol systems (e.g. genetic representations) in dynamical systems (e.g. the physical world). The idea is that the symbolic representations affect the dynamics of the system by generating boundary conditions for that system. Stated in this way, this may not sound like such a big idea, but looking at genomes in the context of dynamical systems actually admits a somewhat more active role for the genome than it is often granted in artificial evolutionary systems. My CA model (called EvoCA) evolves genomes which set the state of specific cells as the CA update rules are applied, and is therefore a tool for investigating the extent to which the evolution of boundary conditions can effectively control a dynamical system.

NKPM Model

An extension of Kauffman's NK model of fitness landscapes, which I am tenatively calling the NKPM model, to look at the evolution of structures which possess multiple selectively significant properties. The parameter P represents the number of properties, and M represents the degree of "overlap" (or "multimodality") in the genetic representations of these properties. The idea behind the model is that evolving organismic structures in the real world possess a variety of properties in various modalities (e.g. structural, electrical, chemical, mechanical, etc.), and that it might actually be easier to evolve such structures than uni-functional ones. The model is a tool to systematically explore this idea. Work on this model has (temporarily?) stalled, but if anyone is interested in it, I would be happy to discuss developing it further.

Other work

Recent collaborations include:

See my publications page for more details about these projects.

Document last updated: Tim Taylor, Monday, 15 January 2007