or 
Biologists have long considered Darwin's Theory of Evolution to be a concise and powerful explanation of all the biodiversity and complexity that we see. For many laymen, however, it's a big leap from the Theory of Evolution, which is essentially a process, and the results of that process running over billions of years. Evolutionary time scales are just too long for most people to reason about. A common stumbling block is that we don't see the process of evolution happening in the time scales in which we live our lives. Evolution is just too slow. Yes, the rise in drug resistant bacteria is an excellent example of evolution happening in our lifetimes, but many people just don't find it compelling as it is not an example of the creation of new species, but instead a modification of an existing one. Furthermore, "proving" the Theory of Evolution just doesn't lend itself to the traditional scientific method of hypothesis, experiment, and repeated verification. While the Theory of Evolution has immense explanatory power, evolution of living organisms is just too slow for direct scientific experimentation.
Recently, we've come to realize that the Theory of Evolution is more broadly applicable than just explaining the complexity and biodiversity of life. It can be applied in at least two other areas--genetic programming in the field of Computer Science and memetics, the study of memes. (We will save the comparison to memes for another essay and here just focus on the first two areas.) The broad class of things to which evolution applies is called replicators. While the specifics of evolution may be different for each kind of replicator, the general principles of Darwinian Evolution are the same. They are:
- Outward traits as an expression of instructions. The form or behavior of the replicator should be largely due to its instructions. In living organisms, this simply means that an organism's phenotype (physical traits and instincts) is largely determined by its genotype (DNA).
- Inheritance of instructions from one generation to the next. The instructions received by the child replicator are to be mostly (or even almost completely) based on those of the parent(s). Some non-systematic mistakes (called mutations) may be introduced in the transfer of instructions from parent to the child. Here, non-systematic, means that they are random in nature, or at least not consciously decided. In living organisms, this transfer of traits is the transfer of genetic material (DNA) from parent(s) to child. (In most complex biological systems, the DNA of the child is a blending of the the DNA of the parents.)
- Natural selection of individuals who pass instructions to the next generation. Natural selection implies that some sets of instructions will be more prevalent in the successive generations--precisely those instructions from individuals who were able to pass them to their descendants. By contrast, instructions from individuals who were culled become less prevalent in the next generation. In living organisms, natural selection largely has to do with the qualities an individual possesses that help it to live long enough to have offspring.
- Natural selection operating through a set of fitness criteria. Unlike mutations, which are random, the fitness criteria are largely consistent over time. The fitness criteria is effectively the measure of the quality of the instructions as they express themselves in the form of the replicator and play out in the environment in which the form lives. In living organisms, fitness criteria include things like the ability of the organism to feed itself, evade predators, and have many offspring.
We would like to highlight the work of John R. Koza, Forrest H. Bennett III, David Andre, and Martin A. Keane who have authored the book Genetic Programming III: Darwinian Invention and Problem Solving based on their ongoing research in the area. The authors have focused primarily on the use of genetic programming as a means to generate computer programs for specific purposes. Genetic programming is especially interesting in this arena because it requires only a specification of a problem to be solved and not a specification of how it should be solved. The authors have also applied these techniques to the area of electronic circuits, which we will highlight here.
Electronic circuits are comprised of a number of interconnected components, which may include resistors, capacitors, inductors, transistors, etc. Each of the components may have different values--resistors, for example, are rated in ohms. There are an infinite number of possible circuits and relatively few circuits do anything of use. Electrical engineers go through formal training to learn various techniques in circuit design, such as building radios and amplifiers. Such techniques are not at all obvious and designing electronic circuits is an art.
In order to apply the evolutionary principles to the arena of circuit design, Koza's group created an environment where a base circuit could evolve over successive generations based on a specific fitness measure. Over many generations, the process produces circuits that are better and better at satisfying the fitness measure. After a time, the circuits produced would achieve high quality solutions to the "problem" posed by the fitness measure. We now explain some of the details of their work by describing these evolving circuits as replicators.
The genotype, or instructions of the "circuit organism" is the schematic of the circuit, or the description of the components and their interconnections. One phenotype would be an actual physical circuit built from the specified components, but in the computer world, such a physical circuit is not really necessary. Instead, the schematic can be fed to a circuit simulator, that allows a virtual run of the circuit under various simulated conditions, thus the phenotype is the schematic of the circuit as needed by the simulator. It is easy to see how the "circuit organism" has the first property of replicators--that the outward traits result from its instructions.
The second replicator principle is that successive generations inherit instructions from earlier ones. Koza's group set up their computer-based environment explicitly for this purpose. Each experiment started with a single basic circuit--a "circuit Eve", if you will--that contains no components but just connects its inputs to its outputs. We call the circuit Eve generation number 0. To create generation one, a random component with a random rating is connected into the circuit. In generation one, not just one circuit is created, but a number of different ones. In the next generation, another random component may be added, the circuit may be rearranged, or a component removed. Again, there may be many different individual circuits each modified in a (usually) different random way. In general, from one generation to the next a circuit undergoes a small set of specific transformations. The possible transformations are rich enough so that the lineage of any circuit imaginable can be traced back to circuit Eve through a series of finite steps. Because each circuit is a small modification of its parent circuit, we satisfy the principle that successive generations inherit nearly all of its instructions from its parent.
We now discuss natural selection and the associated fitness measure for our circuit organisms. The fitness measure is usually based on the input/output (voltage, amperage, frequency response, etc.) function desired of the circuit. So before the process begins, someone specifies what is the desired behavior of the circuit. When a new circuit comes into existence, its simulated behavior can be compared to the desired behavior and the difference between the two reduced to a single numeric rating. The fitness measure need not only consider the behavior of the circuit, but it can also take into account the number of components or monetary cost of building the circuit. Ultimately, it doesn't matter what the fitness measure is, so long as it can be readily computed from each circuit individual.
The final essential aspect of the group's genetic programming environment is the notion of natural selection. It doesn't take very many generations in the genetic programming environment to generate a large number of individuals. In Koza's group's environment each individual gets some amount of "life" in that each individual has a potential to generate offspring, but individuals which have a better fitness measure are allowed to generate more offspring. Keep in mind that at no time does the programmer (Koza's group) specify how the circuit is to be laid out to satisfy the fitness measure. Instead, the circuit layout is an output of the evolutionary process carried out by the genetic programming environment. Such an environment acts as a designer that uses only the principles of Darwinian evolution. But do the designs exhibit intelligence?
The group experimented with various fitness measures given by classic circuit design problems. These include filters, amplifiers, computational circuits, source identification circuits, a robot controller, a temperature measuring circuit and a voltage reference circuit. Out of 14 machine produced results, 10 either infringe upon or duplicate patented circuits. Note that patents are an excellent measure of intelligent designs by presumably intelligent humans. The evolved solutions therefore exhibit intelligent design according to a rigorous measure--these patents were originally granted for nontrivial circuits that did a good job of solving real problems.
Replicators of genetic programming create intelligent designs without the need for any intelligent designer. It could be further claimed that such intelligent designs are an emergent property--a natural consequence--of replicators. The essential principles of Darwinian Evolution were distilled into the genetic programming experiments, yet those same principles apply to the replicators of living organisms. It is not hard to see how the apparent intelligent design in living organisms is simply an emergent property of the process of Darwinian Evolution. No intelligent designer is required.
For further reading on the topic of biological evolution and its unconscious design, we recommend Richard Dawkin's book The Blind Watchmaker.
As a final note, we would like to point out an important difference between the fitness function in Koza's group's genetic programming environment and the fitness function for biological life. In each of the runs of the circuit virtual environment, the fitness measure was based on a specific design problem. Koza's group used their environment to elicit the circuit design that solved that problem--effectively driving the evolutionary process to a desired outcome. In the biological world, by contrast, there is no single fitness function and no god-like decision maker driving the direction of the evolutionary process. In particular, humans are by no means the end goal of the evolutionary process. In all biological cases the fitness measure for a living organism has to do with its survival and its fecundity (ability to proliferate). These qualities depend heavily on the organism's environment which include other organisms that may compete with it for food or other resources or even that consider it food. Biological organisms do not evolve in isolation, but rather co-evolve with all other biological organisms in its ecosystem. The classic example is the lion and the gazelle which are in a kind of evolutionary arms race to build bodies that can outrun each other. Each advance by one side changes the environment, and therefore the fitness measure of the other.
For further reading about Koza's work, see the February 2003 issue of Scientific American.
Postscript:
Since this essay was written, the Intelligent Design movement has gathered momentum and is actively challenging the state laws of Ohio (and other places) to have Intelligent Design taught as a plausible alternative to Darwin's Theory of Evolution. Isn't it ironic that the Christianity Meme has adapted this new argument Evolution itself through evolutionary means? Isn't it amazing that the Christianity Meme finds such self-serving things to do with your tax money and the minds of your children?Here, we add, hopefully, another nail in the coffin of Intelligent Design.
Suppose that the argument for Intelligent Design is correct--that the complexity of life on this planet is proof that there must be a designer intelligent enough to have created the (apparently purposefully designed) complexity. (For proponents of Intelligent Design, the designer is none other than God, though they deny that in public forums.) Clearly, this designer must, Himself, be quite complex. How was He created? If He exists (or existed) in the natural world, the only tool that the ID folks have for explaining such complexity is Intelligent Design. Clearly, the Designer could not have arisen by chance and therefore He must have been designed. But the Designer's Designer must be even more complex. Invoking Intelligent Design again, we must conclude that there must have been an even more complex Designer's Designer's Designer. Continue invoking Intelligent Design over and over again and you are left with an infinite series of ever-more-complex Designers going back infinitely in time. Intelligent Design admits no beginning to this infinite series. This infinite series is:
- completely at odds with a universe that appears finite in both space and time;
- at odds with the fact that we have zero direct evidence of these infinite number of ever-more-complex creators; and
- is contrary to the increase in complexity in life we have seen on this planet.