Evolution Icon Evolution
Intelligent Design Icon Intelligent Design

Evolutionary Computing: The Invisible Hand of Intelligence

invisible hand.jpg

Darwinian evolution is characterized by an utter lack of guidance; it is the “blind watchmaker” of Dawkins. It doesn’t know where it’s going. It doesn’t know it is building a watch. It doesn’t know what a watch is. There is no goal, no plan, and no target in “natural evolution.” Mutations bounce around against environmental constraints, and whatever happens, happens.

William Dembski and Robert Marks have shown that no evolutionary algorithm is superior to blind search — unless information is added from an intelligent cause, which means it is not, in the Darwinian sense, an evolutionary algorithm after all. This mathematically proven law, based on the accepted No Free Lunch Theorems, seems to be lost on the champions of evolutionary computing. Researchers keep confusing an evolutionary algorithm (a form of artificial selection) with “natural evolution.”

A recent example is seen in a Review paper in Nature by Agoston Eiben and Jim Smith, “From evolutionary computation to the evolution of things,” where they claim:

Analogous to natural evolution, an evolutionary algorithm can be thought of as working on two levels. At the higher level (the original problem context), phenotypes (candidate solutions) have their fitness measured. Selection mechanisms then use this measure to choose a pool of parents for each generation, and decide which parents and offspring go forward to the next generation. At the lower level, genotypes are objects that represent phenotypes in a form that can be manipulated to produce variations (Box 1). Genotype-phenotype mapping bridges the two levels. At the genotypic level, variation operators generate new individuals (offspring) from selected parents. Mutation operators are based on one parent (asexual reproduction) and randomly change some values. Recombination operators create offspring by combining values from the genotypes of two (or more) parents. Finally, an execution manager controls the overall functioning of the algorithm. It regulates the initialization of the first population, the execution of the selection-variation cycles, and the termination of the algorithm. It also manages the population size (typically kept constant) and other parameters affecting selection and variation. For instance, it determines the number of parents per generation, and whether mutation, recombination or both produce the offspring for a given set of parents. [Emphasis added.]

This is no analogue to natural evolution. Indicators of intelligently guided information saturate this paragraph. We see decisions, operations, choices, management, and regulation. Don’t the authors recognize their error?

They know that the two processes (evolutionary computing and natural evolution) are only analogous. They produced a table that compares and contrasts the two (see Table 1, “Main differences between natural evolution and evolutionary algorithms”).

It’s apparent that they recognize some serious differences between computer evolution and natural evolution. For one thing, evolutionary computation is vastly simpler than the “highly complex biochemical and developmental process” that they know characterizes life. For another, the definitions of fitness, selection, mapping, and other key words in evolutionary computing are so different from those in Darwinian theory that you wonder how the comparisons convey any meaning at all.

What is entirely lost on Eiben and Smith is the most important distinction: intelligent intervention. Everything from hardware to software has been intelligently designed — including the definition of a problem, the choice of processes and their parameters, and their recognition of what constitutes a solution. To really compare what they do with natural evolution, they would have to stand by a sterile volcano or hot spring with their hands over their eyes, ears, and mouth, and just let things happen. Then, when something happens, say, “Who cares?” That would be Darwinian evolution.

There’s no question that intelligently guided “evolutionary computing” has been very successful at solving problems. Marks and Dembski can testify to that from work in their own Evolutionary Informatics Lab. It can be as successful as artificial selection in breeding. Even when random variation is involved, a little intelligent guidance can go a long way. The authors’ blindness to this is evident:

Evolution has provided a source of inspiration for algorithm designers since the birth of computers. The resulting field, evolutionary computation, has been successful in solving engineering tasks ranging in outlook from the molecular to the astronomical. Today, the field is entering a new phase as evolutionary algorithms that take place in hardware are developed, opening up new avenues towards autonomous machines that can adapt to their environment. We discuss how evolutionary computation compares with natural evolution and what its benefits are relative to other computing approaches, and we introduce the emerging area of artificial evolution in physical systems.

By contrast, Marks and Dembski account for the invisible hand required in evolutionary computing. The Lab’s website states, “The principal theme of the lab’s research is teasing apart the respective roles of internally generated and externally applied information in the performance of evolutionary systems.” So yes, systems can evolve, but when they appear to solve a problem (such as generating complex specified information or reaching a sufficiently narrow predefined target), intelligence can be shown to be active. Any internally generated information is conserved or degraded by the law of Conservation of Information.

But is such an idea too “spooky” and thus outside the bounds of science? The website answers that objection:

Simply put, intelligent design, when applied to biology, seems to invoke ‘spooky’ forms of causation that have no place in science. Evolutionary informatics eliminates this difficulty associated with intelligent design. By looking to information theory, a well-established branch of the engineering and mathematical sciences, evolutionary informatics shows that patterns we ordinarily ascribe to intelligence, when arising from an evolutionary process, must be referred to sources of information external to that process.

What is spooky is failing to see the obvious, believing that information-rich patterns can emerge without intelligence. What is unscientific is explaining an effect without a necessary and sufficient cause, thinking you can get something from nothing.

What Marks and Dembski prove is as scientifically valid and relevant as G�del’s Incompleteness Theorem in mathematics. You can’t prove a system of mathematics from within the system, and you can’t derive an information-rich pattern from within the pattern. The information in a book, for instance, cannot be derived from the paper and ink used to print it. It’s impossible to bootstrap a book from the bare ingredients.

Eiben and Smith, in contrast, see humans as pulling themselves up out of their own evolutionary bootstraps and taking control of evolution:

From a historical perspective, humans have had two roles in evolution. Just like any other species, humans are the product of, and are subject to, evolution. But for millennia (in fact, for about twice as long as we have used wheels) people have also actively influenced the course of evolution in other species — by choosing which plants or animals should survive or mate. Thus humans have successfully exploited evolution to create improved food sources or more useful animals, even though the mechanisms involved in the transmission of traits from one generation to the next were not understood.

Question: Did humans influence, choose, exploit, create, and improve things by intelligent design? Think about that as they continue:

Historically, the scope of human influence in evolution was very limited, being restricted to interfering with selection for survival and reproduction. Influencing other components, such as the design of genotypes, or mutation and recombination mechanisms, was far beyond our reach. This changed with the invention of the computer, which provided the possibility of creating digital worlds that are very flexible and much more controllable than the physical reality we live in. Together with the increased understanding of the genetic mechanisms behind evolution, this brought about the opportunity to become active masters of evolutionary processes that are fully designed and executed by human experimenters ‘from above’.

So ask: Did humans interfere, design, influence, create, control, understand, and take mastery of these things as an intelligent cause?

It could be argued that evolutionary algorithms are not faithful models of natural evolution (Table 1). However, they certainly are a form of evolution. As Dennett said “If you have variation, heredity, and selection, then you must get evolution”.

There you go — a proof by assertion and argument from authority. But can they really dismiss critics so easily? Artificial selection is “a form of evolution,” too. What’s the difference between that and natural evolution? Intelligence. The breeder makes use of natural processes to obtain the outcome that his designing mind sought. Without intelligent interference, no evolutionary algorithm would be superior to blind search.

In the early days of intelligent design, Thaxton, Bradley, and Olsen recognized the fallacy at work here: the problem of “investigator interference.” In their milestone book The Mystery of Life’s Origin (1984), they showed the many ways that investigators in the origin-of-life field “played a highly significant but illegitimate role in experimental success.” At the end of Chapter 6, they concluded with a quote by Brooks and Shaw:

These experiments… claim abiotic synthesis for what has in fact been produced and designed by highly intelligent and very much biotic man. (p. 110)

The results would never have occurred without the investigators’ help. Like puppet masters, they guide outcomes even if some randomness is thrown in. The same applies to evolutionary computing. Eiben and Smith fail to see that the “invisible hands” running the show are their own. And unless they are willing to claim their hands are unguided by any intelligence, they don’t establish Darwinian evolution; they confirm intelligent design.

Image: � Rafinade / Dollar Photo Club.

Evolution News

Evolution News & Science Today (EN) provides original reporting and analysis about evolution, neuroscience, bioethics, intelligent design and other science-related issues, including breaking news about scientific research. It also covers the impact of science on culture and conflicts over free speech and academic freedom in science. Finally, it fact-checks and critiques media coverage of scientific issues.

Share

Tags

Computational SciencesMathematicsResearchScience