The Edge of Evolution

by Michael J Behe

  To grasp the problem of incoherence, picture a huge castle—much bigger than Iacocca Hall, much taller than six floors. At the very top of the castle is a single small room with a balcony, from which, say, a hero can wave to adoring crowds below. Getting to the top requires navigating a maze. In the castle there are no coherent staircases, only single steps. The bottom of the castle is a large room with a hundred doorways. Behind each door is a single step that opens onto another room (a different room for each of the doorways), which again has a hundred doorways leading to a single step, leading to a new room with a hundred doorways, and so on. Frequently, however, the step up through a doorway leads to a room with no other doorways and no other steps—a dead end. Only one route of the many possible ones leads to the window. An elderly, blind knight who is retiring from service begins climbing at the bottom of the castle with one thought in mind—to limp up any step he comes across, hoping to reach the balcony at the summit and receive the adulation he deserves. Unfortunately, because he is elderly, once he goes up a particular step he can’t go back down; he would stumble and injure himself. So if he reaches a dead end, he’s stuck. Although in a storybook the knight would surely find his way somehow, Darwinism professedly has no use for fairy tales. Almost every knight in a Darwinian story should get stuck in a dead end, never to reach the summit.

  The blind knight’s quest to climb to the summit of a castle that has no coherent staircase—only disconnected steps—mirrors a staggering difficulty for Darwinism: Even if there is some gradual route to a distant pinnacle, it is not “biologically reasonable” to expect random mutation and natural selection to navigate a maze to get there. Because steps are not preorganized into a staircase, and because so many wrong turns and dead ends lie in wait, an unseeing search would almost certainly fail. Without a floor plan or a guide, the knight would languish in some lower windowless room.

  RUGGED LANDSCAPES

  A problem akin to the knight’s predicament has been discussed fitfully with no resolution in evolutionary biology journals under the name “rugged fitness landscapes.”8 In the 1930s the mathematical biologist Ronald Fisher pictured evolution as an exercise in hill climbing. The idea is that a species would gradually evolve to get better and better—to become more “fit”—until it was as good as it could be under the circumstances. In a sense, the species would rise to the acme of an evolutionary hill. Once there, it would be stuck—going back down the hill means getting less fit, which in a Darwinian competition should almost always be prohibited.9 Well, what if, more realistically, instead of a single hill, the evolutionary geography actually resembled a badlands: a whole rugged landscape filled with many hills—big ones, little ones, tiny ones? The tiny ones are by far the most common, bigger ones much less frequent. There is only one highest peak. If so, then in a rugged evolutionary landscape, it is much more likely that a species will climb a tiny hill and get stuck there, unable to become less fit, yet forever isolated from the surrounding peaks. Random mutation and natural selection can’t solve the rugged landscape dilemma—they actually cause the dilemma.

  Even in the shadow of an evolutionary Mount Everest—the promise of some terrific new biological feature—the challenge of a rugged landscape would remain. In fact, that is where it would become especially difficult. The more complex and interactive a system, the more its simple variations will short-circuit evolutionary hill climbing. As a physical example, think of the goal of building a structure like Iacocca Hall. An evolutionary story might start with a small shack, useful as a shelter, and hope to build on that. But the materials one would use to build a shack (wood, straw, nails) are not the ones one would need for a larger structure (cement, steel). The shack would serve, for a while, but could not be altered into a large building without essentially being replaced. Yet tearing down the building would remove the only shelter available at the time. Even construction of a small building that improbably used cement and steel would not include spaces for future staircases, electrical wiring, and so on that would be needed for a larger building. A smaller building that did have space for them would very likely be less efficient and more costly than one that didn’t.

  FIGURE 6.1

  Evolutionary fitness landscapes. The top figure represents a simplistic evolutionary landscape, where only one or a few traits can vary, and fitness can increase smoothly. Ordinary Darwinian processes would easily drive a species to the single pinnacle. The bottom figure represents a more realistic, rugged evolutionary landscape, where many traits can vary. Here random mutation and natural selection would drive a species to some local peak, where it would remain stuck. Natural selection would actively inhibit a species from traversing such a landscape. If a limited scientific study focuses on just one peak of a rugged landscape, the results can misleadingly seem to match the smooth peak. (Reprinted from Gavrilets, S. 2004. Fitness landscapes and the origin of species. Princeton, N.J., Princeton University Press. Courtesy of Sergey Gavrilets.)

  To mix metaphors, how many steps should we expect random mutation and natural selection to climb before getting stuck on a tiny hill of a rugged landscape? Very few. Using a sophisticated mathematical model, H. Allen Orr decided that the likeliest number for a single gene was between just one and two.10 That count fits pretty well both with John Maynard Smith’s reasoning about proteins and with what we know from the best relevant data on evolution we have available—the effects of malaria on the human genome. The evolutionary response of the human genome to Plasmodium falciparum has been exactly what you’d expect of a Darwinian process—disjointed and incoherent. In one group of humans the G6PD gene is broken, in another band 3 protein is defective. Both are single steps to small, local adaptive peaks. The sickle mutation pops up once or a few times, and then, separately, alterations in fetal hemoglobin ameliorate its side effects—several steps to an unrelated adaptive peak. Like some blind knight stumbling through a castle maze, in the case of sickle/fetal hemoglobin, Darwinism has managed to walk up two steps, but has become stuck in an evolutionary dead end. Random mutation and natural selection are operating at full steam, but they lead nowhere.

  This is not the kind of process that could have coordinated the many proteins that work in concert in intraflagellar transport. It is not the kind of process that leads to any significant degree of coherence.

  AN ANALOGY OF INCOHERENCE

  To get a better feel for the helplessness of Darwinism in the face of the problem of incoherence, let’s return to another analogy. Suppose you enter a large room to find those proverbial million monkeys chained to keyboards. (Unlike the unruly, real-life macaques, these hypothetical fellows politely keep pecking away at all the keys.) They are hard at work revising Moby Dick. Because they are thoroughly modern monkeys who use word-processing computers instead of typewriters, not only can they change a letter here or there, they can also randomly delete or duplicate or rearrange entire passages, as well as add text, find and replace, and so on. In fact, since the computers are networked, sometimes the text on one computer can even randomly recombine with the text on another computer. These extra abilities nicely mimic the variety of ways in which DNA can mutate in the cell.

  Altered texts are offered for sale to the public, and the sales of a version determines its fate. If a change is for the worse, as most would be, then it sells poorly. If it sells less well than any other copy, the marvelous computer erases it and reverts to the original text on the computer before the monkey made the most recent alteration. However, if a revision improves the text—so that it would sell more copies—then it somehow becomes the standard text against which changes are judged not only in the computer of the monkey who originally typed it, but also in some of the other monkeys’ computers.

  What would count as an improvement? A variety of changes. One way to improve Moby Dick would be to make the novel easier to read. Another way would be to make it more entertaining. Another to make it more profound. Another to make the book cost less. The book might cost less if it wer
e shortened. It might become more profound by repeating profound passages, or by including profound passages from other texts (like, say, the Gettysburg Address), which may also be stored in the computer. To be more entertaining it might contain silly misspellings, or jokes copied from other texts. To be easier to read it might delete words, sentences, or much more (à la Reader’s Digest). In short, there are many, many ways in which the complex text might change to improve the book. In fact, there are so many possible improvements that we should not expect changes to be substantially connected to each other. They are likely to accumulate independently and incoherently.

  To illustrate, let’s look at some of them. One of the earliest monkey alterations is the insertion of a comma into the first sentence, changing it from “Call me Ishmael” to “Call me, Ishmael.” But that doesn’t help at all. It doesn’t make the book more profound or cheaper. It makes the book more difficult to read, since the first sentence now fits poorly with the rest of the book. The problem of coherence increases as the complexity of a text increases. If “Call me Ishmael” were the whole text, then inserting a comma after “me” would significantly change its meaning without conflicting with the meaning of later text. But since it’s part of a longer, coherent story, it doesn’t fit. The book doesn’t sell, so the alteration is erased and the text reset to the original.

  By contrast, another early change, a duplication, is successful. In the final chase for the white whale in the last chapter, in one sentence “even now,” is repeated several times to give “‘Oh! Ahab,’ cried Starbuck, ‘not too late is it, even now, even now, the third day, to desist.’” The change strikes some readers as more dramatic and profound, and thus spreads to some other monkey computers. In another alteration the second, descriptive paragraph of Chapter 44 is deleted.11 The change interrupts the narrative flow but, because it also shortens the text slightly, it makes it a bit more readable; on balance it is a slight improvement. It spreads to some other computers. A third beneficial change is the deletion of Chapter 6, mostly description. A fourth change is the duplication of the epilogue, which counts as more profound (don’t ask blind Nature to act as a picky literary critic, please). A fifth switch substitutes “Arab” for “Ahab” throughout the story. A sixth recombines the text on a computer missing Chapter 6 with one that has a duplicated epilogue. In between all these major changes are many smaller ones where individual words and phrases are deleted, rearranged, misspelled, duplicated, and so on. Occasionally, a change might build upon a previous change—maybe successive paragraphs might be deleted in two separate steps, or maybe several alterations change, say, WORD to WORE to GORE to GONE to GENE, and perhaps that might help somehow—but those alterations are no more likely to be helpful than many other possible improvements.

  In the end, although many changes accrue to the text, and even though the text is in a sense “improved” in that it sells better than the original edition, the changes do not add up to anything like a coherent new story. There is no new ending where, say, Ahab survives and sells the blubber of Moby Dick for a fortune, or where Ishmael recounts his earlier life before going to sea. Writing a coherent story of course requires an author like Herman Melville, who can visualize the storyline in its overarching complexity.

  How many changes could be made randomly to a relatively small piece of text—a sentence or paragraph—before it comes to a local dead end, better than all other single changes surrounding it, but still not very much improved from the starting text? Although he was considering biological texts, Allen Orr’s work on genes is likely applicable here as well. If so, we should expect the answer to be one or two changes, possibly several. A deletion or two in a sentence or paragraph might make the text more readable, but more deletions might interfere with the sense of the text. The same with other changes—rearrangements, substitutions, and so on. As with smaller pieces of text such as words and sentences, so, too, with larger ones like book sections and chapters. Switching or deleting a chapter or two in a book might make it more readable; further large changes likely wouldn’t.

  The eminent geneticist François Jacob famously wrote that Darwinian evolution is a “tinkerer,” not an engineer.12 He’s exactly right. Tinkering means looking for quick fixes, features that work for the moment—incoherent, patchwork change, doctoring machines with chewing gum and duct tape, stopping an invader by burning a bridge or breaking a lock, “improving” a text by typing disjointed changes to words, letters, paragraphs, and chapters. If Darwinism is a tinkerer, then it cannot be expected to produce coherent features where a number of separate parts act together for a clear purpose, involving more than several components. Even if someone could envision some long, convoluted, gradual route to such complexity, it is not biologically reasonable to suppose random mutation traversed it. The more coherent the system, and the more parts it contains, the more profound the problem becomes.

  A MINOR FACTOR

  The degree of coherence of a system and the number of steps that have to be skipped to get from one level to another are the two major criteria by which we can try to locate the edge of evolution. There is also a lesser factor that sometimes has to be taken into account to minimize our chances of being misled: degradation—when a more complex system breaks down to yield less complex systems. For example, suppose an automobile fell apart. We might be able to salvage from it a number of separate parts, such as a radio, air conditioner, or pump. Or consider that one might come across the isolated phrase “Call me Ishmael” and not have any idea that it derived from a larger system. In deciding how a particular feature arose, we have to consider whether it had originally been a piece of a larger system. Of course, when degradation does occur, it simply means that the question of Darwinian randomness must be readdressed to the preceding structure.

  COHERENCE, STEPS, AND IRREDUCIBLE COMPLEXITY

  How does irreducible complexity, which I first described in Darwin’s Black Box and discuss in the last chapter, fit with the criteria of coherence and evolutionary steps introduced in this chapter? Although closely related to it, the new concepts gauge difficulties for random mutation at a much finer level than does irreducible complexity, and are more appropriate to defining the edge of evolution. If irreducible complexity is likened to a rough measuring tool—say, a yardstick with no markings—then the new criteria can be thought of as a ruler subdivided into millimeters.

  Let’s reconsider the mousetrap. A common mechanical mousetrap is an example of irreducible complexity because it “is composed of several well-matched, interacting parts that contribute to the basic function” and “the removal of any one of the parts causes the system to effectively cease functioning.”13 The mousetrap is resistant to gradual Darwinian-style explanations.

  Yet a boatload of unnoticed action is packed into the terms “well-matched” and “interacting.” The spring of a mousetrap isn’t just any old spring. For example, the spring from a grandfather clock or windup watch or toy car would be useless in a standard mousetrap. In fact, the spring in a mousetrap is essentially unique. Its length is critical to its role in the trap. If there were somewhat fewer or more coils the ends of the spring would not be positioned correctly. The ends of the spring themselves are not coiled; rather, they’re extended. One end presses against the wooden platform and the other oddly shaped end hooks over the hammer—the part of the trap that strikes the mouse. The positioning of both ends is crucial to the work of the spring. If either end were at a somewhat different angle coming off the main body of the spring, the trap would be ineffective. If either end were substantially longer or shorter, the system would again be compromised. It’s clear that not only the whole trap, but also the spring and other pieces required multiple, coherent steps to produce.

  At the risk of appearing pedantic, let me list some steps. Suppose that a smith intends to make a spring for a mousetrap. To do so he takes a series of actions. In his shop he has a large number of lengths of spring of various shapes and sizes that he keeps in store for
projects. Out of the hundreds of stock springs he chooses one with the right diameter and resilience. He cuts a one-and-three-quarters-inch length off the end of the yard-long stock. He heats up one end in a flame and straightens out the metal to a length of one inch. He does the same to the other end. He positions one end at an angle of 180 degrees with respect to the other end. Then he puts a crimp in the end that will overlap the hammer when the trap is put together. The lesson is obvious: Simply to fashion the spring into the correct special shape needed for the trap requires multiple, coordinated steps.

  The concept of irreducible complexity, with its broad focus on the “parts” of a system, passes over the fact that a part might itself be a special piece that needs explaining in terms of many steps. What’s more, it also overlooks the steps required to assemble a system—to physically put them together—once the parts are available. Once the spring is forged and the other pieces of the future mousetrap manufactured, the smith grabs the various pieces lying at different spots in the shop, transports them to his workbench, and pieces them together in the right orientation. The concept of the number of “steps” resembles the idea of irreducible complexity in that both look to see if multiple factors are needed to produce something. But “steps” goes further, asking how many separate actions—not just separate parts—are needed to make a system. The concept of “steps” is especially useful when fewer actions are needed to coherently arrange parts. It can locate the edge of evolution with greater precision.