In a new paper in the journal BIO-Complexity, “Complexity in Computer Simulations,” computer scientist Winston Ewert reviews much of the literature claiming to show, including via computer simulations, how irreducible complexity might have evolved by undirected means. He finds that “Behe’s concept of irreducible complexity has not been falsified by computer models.” The models include Avida, Ev, Steiner trees, geometric models, digital ears, and Tierra. Ewert reports that in many cases, the “parts” that compose the irreducibly complex system are “too simple.” The programs are designed such that systems that the programs deem “functional” are very likely to evolve.
Almost all of the cases of proposed irreducible complexity consist of parts simple enough that a system of several components could be produced by chance, acting without selection. As such, they fail to demonstrate that their models can evolve irreducibly complex systems, especially on the scale of biological complexity.
For evolutionary theorists, this leads to a conundrum. Since “Darwinian evolution is an ateleological process,” operating without a goal, this means that “If a model is designed to assist the evolution of an irreducibly complex system, it is not a model of Darwinian evolution” and “Any decision in the construction of a model made with an eye towards enabling the evolution of irreducible complexity invalidates the model.” Ewert finds that this is precisely what many of these models do. In the one case that a truly irreducibly complex system was produced by a program, Ewert found it was “designed as part of the ancestor used to seed the … simulation,” and thus did not evolve in the Darwinian sense. He concludes that computational attempts to explain the evolution of irreducible complexity have “failed on a number of fronts”:
Two of the models fail to satisfy the knockout test, in that they maintain functionality after parts have been removed. Almost all of the models use parts that are trivially complex, on the order of an amino acid rather than a protein in complexity. None of the models attempt to show why the mechanism used necessarily requires its parts. Finally, some of the models have been carefully designed to evolve. Thus, none of the models presented have demonstrated the ability to evolve an irreducibly complex system.
He concludes, “The prediction of irreducible complexity in computer simulations is that such systems will not generally evolve apart from intelligent aid” and this prediction “has thus far stood the test in computer models.”