Is Natural Selection Like a Computer Algorithm?
Analogies can be useful, but they can also distract from what's really important. Consider a simple example. Say you are at a playground, and you hear a steady din of children's voices at play. You notice that the sound has spectral characteristics: a particular frequency range, dynamic range, and time variation, with occasional spikes. As a scientist, you produce a mathematical model to characterize the sound spectrum. You even make a prediction: the addition of n more children will cause the intensity to rise, but not the average pitch. Finding confirmation, you call your model the Weighted Playground Spectral Algorithm (WPSA) and publish it in a journal, where you apply it to galactic spectra. What have you done?
While your model might be useful in some contexts, it obscures what is most important about the playground: the minds, emotions and needs of the children and their parents. A star's light is not in the same ballpark (or playground) as a child calling out cheerfully, "Daddy! Watch me slide down the slide!" Yet we see a similar fallacy in a Commentary for PNAS by Nicholas H. Barton, Sebastian Novak, and Tiago Paixão of the Institute of Science and Technology in Austria. They compare evolution to computer science.
In "Diverse forms of selection in evolution and computer science," the authors notice that a particular mathematical model, the Multiplicative Weights Update Algorithm (MWUA) used in computer science, seems to fit some evolutionary scenarios, too. Launching from a paper by Chastain et al. called "Algorithms, Games and Evolution," they find this similarity interesting, and perhaps fruitful.
The astonishing diversity of complex adaptations that we see in the living world has been produced by natural selection, over ∼ 3.5 billion years of evolution. Information from the largely random survival and reproduction of past organisms has accumulated to produce genomes that are precisely fitted to diverse environments. In PNAS, Chastain et al. show that natural selection on freely recombining populations is equivalent to the multiplicative weights update algorithm (MWUA), an efficient optimization algorithm that has been discovered many times in computer science, statistics, and economics. Whether this equivalence explains the extraordinary effectiveness of natural selection, or conversely, of artificial algorithms, depends on one's perspective. Perhaps surprisingly, theoretical results in population genetics and in computer science look quite different, even when they deal with essentially the same questions. Thus, the equivalence identified by Chastain et al. allows these different results to be transferred between fields in both directions. (Emphasis added.)
The obvious problem with this line of thinking is that evolution has no algorithm! By definition, it is aimless, unguided, and mindless -- unlike an economist, statistician, or computer scientist. Like the playground voices compared to stellar spectra, any similarities will surely be swamped by more important differences. We also see that the commenters assume natural selection created all the "astonishing diversity of complex adaptations" in the living world, forcing them to try to fit an intelligent-design algorithm on what they believe originated by chance. Here's how Chastain et al. commit the same question-begging error:
Even the most seasoned students of evolution, starting with Darwin himself, have occasionally expressed amazement that the mechanism of natural selection has produced the whole of Life as we see it around us. There is a computational way to articulate the same amazement: "What algorithm could possibly achieve all this in a mere three and a half billion years?" In this paper we propose an answer: We demonstrate that in the regime of weak selection, the standard equations of population genetics describing natural selection in the presence of sex become identical to those of a repeated game between genes played according to multiplicative weight updates (MWUA), an algorithm known in computer science to be surprisingly powerful and versatile. MWUA maximizes a tradeoff between cumulative performance and entropy, which suggests a new view on the maintenance of diversity in evolution.
Once again, they first assume evolution, then try to force-fit an intelligently designed algorithm (MWUA) onto the whole of life. This is fallacious. It takes a mind to assign weights, play games to win, and run algorithms. It takes a mind to recognize a tradeoff, measure performance, and counteract entropy. A mindless world is incapable of such things. The authors have merely transferred their own mental activity onto a world that they must assume, being Darwinists, is blind and uncaring. Barton and colleagues never quite catch the absurdity of the comparison:
Although computer science and evolutionary biology appear to be very different fields, there are some surprisingly close parallels, which are apparent if one thinks of natural selection as providing a general algorithm by which populations can efficiently learn about their environment. Although the questions in each field differ, there are common themes -- not least, how rapidly can evolutionary algorithms act? In computer science terms, what is the complexity of an evolutionary algorithm, or in other words, how does the runtime scale with the dimensionality of the environment, genome size, population size, and so on? It is striking that, although these different fields deal with similar problems, their theoretical structures are quite different, giving an opportunity for fruitful transfer.
The only algorithm possible for evolutionary theory is what we might dub (after Berlinski) the SDLA: the "Sheer Dumb Luck" Algorithm. Unfortunately, that algorithm is weighted heavily in favor of entropy and extinction. We would hope that Darwinists would not try to transfer their algorithm back onto the computer scientists. It may be too late for the economists.