It has been said that the Achilles heel of Darwinism is the inability of the joint mechanism of natural selection and random mutation to account for the incredible complexity found in living organisms. In recent times this challenge to evolution has been met with several attempts to demonstrate otherwise. The May 8th issue of the journal Nature contains a scientific paper that describes such an effort. The paper titled "The evolutionary origin of complex features" [1] documents how digital organisms have been able to evolve complex logic functions by natural selection and random mutations. At first glance, this work seems to have parried the main threat to Darwinism by generating complexity without intelligence - or has it? In this response we will analyze the experimentation and evaluate the applicability of the exercise to the study of origins.
The work described in Nature falls within the realm of computer science named evolutionary computing. This field makes use of genetic algorithms to simulate biological evolution. Typically genetic algorithms are composed of a fitness function and the search space (or fitness landscape). The fitness function is used to evaluate the health of a given digital organism and is the basis for selection, and the fitness landscape is the continuous space of every possible fitness state. In the experiment, the fitness function [2] evaluated the number and complexity of the logic functions that a given digital organism could perform and assigned it a discrete value representing its fitness. The fitness landscape consisted of points (or valleys) where no logic functions could be performed, peaks where all of the possible logic operations set could be performed, and everything in between. The outcome was that as generations of digital organisms passed, the genetic algorithm often guided lineages of these organisms through the fitness landscape towards the ability to perform logic functions, even those considered to be fairly complex.
It should be clear by now that genetic algorithms are quite powerful. However, they do carry with them a certain pitfall. The results of genetic algorithms are often overstated because these evolutionary algorithms are subject to what I call the "design in, design out" principle. The "design in, design out" principle is patterned after the computer science axiom "garbage in, garbage out" which is used to describe the expectation that when invalid data is entered into a system that it will result in an invalid output. Likewise, in the "design in, design out" case, if design is put into a system, especially one that is supposedly modeling a process that is directionless, then one should not be surprised if the output is the result of teleology.
In the case of the article described in Nature, the "design" or the intelligent human influence can be found in the fitness function. As stated earlier, the primarily fitness function assesses health (or fitness points) on a monotonically increasing scale based upon the number and complexity of logic functions that a given digital organism could perform. For example, both NOT and NAND logic operations were associated with 2 rewards points, AND and OR_N were associated with 4 reward points, and so on up to the EQU operation which was associated with the greatest reward - 16 points. The award system was not an arbitrary one, but one that was chosen very carefully. Just ponder the number of variations of different reward systems that could have been used in the genetic algorithm. The entire set is extraordinarily large, and the vast majority of them would have not resulted in convergence on logical complexity.[3] Therefore the fitness function is highly specified. In fact, the fitness function was designed so well that it provided a 'map' that guided the genetic algorithm in fairly regular fashion [approximately 50% of the time] to the target in mind - complex logic functions and in particular, the EQU [4] function.
This point is actually underscored by the fact that in an environment when no merit points were rewarded for simple logic operations that the most complex operation, EQU, never evolved. Consider the following text from the paper:
"..50 populations evolved in an environment where only EQU was rewarded, and no simpler function yielded energy. We expected that EQU would evolve much less often because selection would not preserve the simpler functions that provide foundations to build more complex features. Indeed, none of these populations evolved EQU, a highly significant difference from the fraction that did so in the reward-all environment."
But the observation that genetic algorithms require a well designed fitness function to be successful is not a new one. In his book, "No Free Lunch" [5], William Dembski describes this particular pattern of "smuggling in" information to allow genetic algorithms to achieve their targets (EQU in this case). Dembski stated that "if an evolutionary algorithm actually proves successful at locating a complex specified target, the algorithm has to exploit a carefully chosen fitness function." He goes on further to say that "[t]his means that any complex specified information in the target had to first reside in the fitness function." His statements are entirely true for the work described in Nature.
So where does this leave us from the perspective of intelligent design? Should the work described in the paper be considered invalid and the result of “garbage in” experimentation? No it should not. The paper actually provides support for biological teleology since it suggests that the evolution of complexity is only possible with a carefully fined tuned fitness function. Of course, it remains to be seen whether or not nature contains within it such a series of fitness functions, but that’s another discussion for another day.
Specified complexity like that resulting from the genetic algorithm described in Nature has one known cause and that is intelligent agency. Therefore we should not be surprised to find out that very smart scientists have played a role in the incredible evolution of the logic savvy digital organisms described in the Nature journal article. So the next time that you are confronted with a newsflash that makes extraordinary claims about the results of a genetic algorithm, get a pencil and a notepad and prepare yourself for a lesson in design.
References
[1] “The evolutionary origin of complex features”, Nature Vol 423 Web Access to article here
[2] Actually, the paper describes a few variations of fitness functions, but the one that I am referring to is the primary algorithm used.
[3] The reward system must provide a smooth gradient for the evolutionary algorithm to climb with very few local maxima.
[4] EQU stands for Equals ("=") and logically is expressed as follows EQU = (A and B) or (~A and ~B) and is the most complex of the 9 one and two input logic functions.
[5] “No Free Lunch” by William Dembski
