In a previous post I described the hunch engine, an exploration-support tool based on interactive genetic algorithms. There is an intriguing parallel I want to expose in more detail between biological evolution and decision making: search and evaluation in decision making are similar to variation and selection in evolutionary theory. Search is all about creating a variety of options and possible answers to a query; evaluation is the process through which some or none of the options are selected. Nature thus provides us with a powerful metaphor for decision making, and in that context genetic algorithms are decision-support tools. With interactive genetic algorithms, variation is performed by a non-human device while options generated by the device are evaluated by a human being.
In fact, we humans have been using this technique for hundreds of years, it is known under various names such as breeding, animal husbandry, or directed evolution. To name one famous example, corn was bred about 9000 years ago by Mexican farmers. Teosinte, the plant they started with, is so different from modern corn, that it was originally classified in a different genus. Teosinte is barely edible, while corn is today one of the leading sources of calories for humans. The story of how such a transformation was made possible, by the combination of careful selection by farmers with a genetic structure that enabled dramatic morphological changes, is still being uncovered by ongoing research. Which means that humans have been using a powerful biological engine called variation which they did not understand at all; all they knew was that it worked for producing the requisite amount of variation and they could provide selective pressure. The philosopher Daniel Dennett uses the phrase “competence without comprehension” to describe the strange value proposition of Darwinian evolution, that “to make a perfect and beautiful machine, it is not requisite to know how to make it’’ [MacKenzie RB (1868) (Nisbet & Co., London), cited by Dennett]. Indeed, what McKenzie, a 19th century critic of Darwin, calls “a strange inversion of reasoning”, has been one of the weapons creationists have used, as in the pamphlet from the 1970s. But directed evolution is a highly successful embodiment of that inversion of reasoning –of competence without comprehension. Corn is the descendant through directed evolution of teosinte. The domestic dog, in its apparently infinite variety, is the product of many generations of breeding from just one common ancestor, the gray wolf. Examples abound. In silico evolution, in the form of genetic algorithms, creates an opportunity for competence without [necessary] comprehension. You may be able to comprehend –either during or after the design process, but you don’t have to. The machine takes care of the variation process. This is a powerful concept: think about all the situations in which you have to rely on an expert to produce variations for you –an architect, a designer, a brand naming consultant, etc. The expert is the gatekeeper between you and your dreams, and defines the possible on the basis of her own biases. Your dreams are bounded by the expert. Yet, you are an expert on knowing whether you like something or not. You may not understand how the expert comes up with the variations, but you’re competent (in fact, you’re likely the most competent) when it comes to your own tastes. Competence without comprehension empowers you to innovate far beyond your comprehension. One caveat is obviously that whatever new stuff you produce be safe. Consider the following situation: you took a picture with your camera-phone, but it’s too dark. You would like to make your photograph prettier, but you have no comprehension of digital photography. There is often a “magic button” you can use with digital photography software –yet you know that the outcome is arbitrary. One thing you are very competent about is deciding whether a picture is good or not. How do you exert your competence without comprehension of the underlying technology? The hunch engine offers you multiple choices, each of which is an image obtained from applying a random filter to your original picture. Following your competent hunches, you select one or more of the pictures that seem to offer some degree of improvement over the original. And you press “evolve”.
The next iteration presents you with another set of choices, this time variants (mutated and recombined filters) of the filter(s) you selected in the previous iteration. Again, select the images that offer the most improvement. After two or three steps of this directed evolution, you end up with just the right picture –and you know it. At no time did you have to understand what was going on under the hood.