Monday, November 12, 2007

Evolution in civil engineering

One of the issues at the heart of the debate between proponents of darwinian evolution and proponents of intelligent design is the ability of a darwinian process to come up with solutions to problems. One of the things about a "system", beyond its inputs, outputs, energy supply and processing, is that the system as a whole carries out some function, and the function is a solution to a problem.

It is difficult to identify systems in nature where an evolutionary process has solved problems. The famous example of the apparent development of antifreeze glycoproteins in notothenioid fish is a candidate. A partial solution to the "problem" of malaria can be found in sickle cell anaemia - a genetic problem in humans that otherwise would not have provided a selective advantage (see "Edge of Evolution"). So it is quite often the case that when challenged to offer evidence of the power of evolution to solve problems, proponents of darwinism may point to its success in non-biological areas.

One example I've been pointed to more than once is the "evolutionary" process which led to novel designs for IC layout. Another one featured in the February 2007 issue of "Civil Engineering", the Proceedings of the Institution of Civil Engineers. An article entitled, "Going organic: using evolution in civils design," by Pasquale Ponterosso (what a good name for a structural engineer!) and Dominic Fox from Portsmouth University, introduced the genetic algorithm to civil engineering designers, and presented "summaries of recent research areas of application, including reinforced earth embankment design, truss optimisation, masonry arch collapse loads and mechanisms, and yield-line analysis of reinforced-concrete slabs."

Good stuff! How powerful is darwinism, if it can do all that for us!

But it's not quite that simple. The achievements claimed are somewhat more modest. In engineering terms, the problem has to be fairly well defined, and the parameters of the solution are also set by the programmer/engineer.
Due to its random nature, the genetic algorithm is not expected to provide the optimum solution. The normal procedure is to run the genetic algorithm several times and use the best answer obtained over a number of runs. This tends to limit applications of the genetic algorithm to problems where a good solution is acceptable (rather than the optimum one), and where the search space is so large that conventional numerical optimisation techniques are not practical in a reasonable timeframe.
So in engineering terms, this is good for finding "local optimums" in a large search space, but doesn't provide a means of knowing whether the best solution has been obtained. If this sort of genetic algorithm is to be used as an analogue for real biological systems, I think it is then necessary to apply this back to those systems, and demonstrate that the problems solved are of this sort. Certainly the search space is large enough. But is it too large? - is it the case that improvements in fitness are too scarce to start with for darwinian processes to make any headway in establishing fitness? This harks back to my queries from some time ago about the actual size of the search space for biological systems, which was never satisfactorily answered. It is a crucial question if darwinism is to be a creditable explanation.

Finally, the authors warn:
As with most engineering software, it cannot be blindly assumed that the output is correct or that the result is aesthetically pleasing.... An incomplete evolutionary system or fallacious or incomplete input, or inaccurate boundary conditions, will lead to erroneous output. Engineers must always be the final authority.

The genetic algorithm cannot replace an engineer's experience or judgment, but may be useful as an aid to design thinking or to creativity.
Obviously, nobody will willingly surrender their own responsibilities to a computer. But there is more to it than that. The writers are arguing that in addition to the definition of the problem and the parameters of the solution, there is a need for intelligent input to evaluate the solution. A darwinist would perhaps argue that the real world provides an environment in which evolutionary solutions face the ultimate test of their fitness. However, at least in the context of evolutionary design in civil engineering, a great deal of intelligent input is required to make this blind, random process yield something worth looking at. This input isn't available to the materialist.