The Structure of Evolutionary Theory

by Stephen Jay Gould

  But can such a sequence represent the history of life, or even stand as a surrogate either for the fundamental feature of that history, or for the central causal processes of evolutionary change? When we shift our focus to the full range of life's diversity, rather than the upper terminus alone, we immediately grasp the treacherous limitations imposed by misreading the history of an ex­treme value as the epitome of an entire system. The sequence of increasingly right-skewed distributions with a constant modal value firmly centered on bacteria throughout the history of life (Fig. 9-29) represents only a cartoon or

  9-29. This cartoon of changing form and range in the histogram of complexity through life's phylogeny illustrates how we fall into error when we treat extreme values as surrogates or epitomes of entire systems. A view that emphasizes speciation and diversity might recognize the constancy of the bacterial mode as the outstanding feature of life's history. From Gould, 1996a.

  [Page 898]

  an icon for an argument, not a quantification. The full vernacular under­standing of complexity cannot be represented as a linear scale, although meaningful and operational surrogates for certain isolated aspects of the ver­nacular concept have been successfully designated for particular cases — see McShea, 1994.

  I do not see how anyone could mistake the extreme value of a small tail in an increasingly skewed distribution through time for the evident essence, or even the most important feature, of the entire system. The error of construing this conventional trend of extremes as the essential feature of life's history be­comes more apparent when we switch to the more adequate iconography of entire ranges of diversity through time. Consider just three implications of the full view, but rendered invisible when the sequential featuring of extremes falsely fronts for the history of the whole.

  1. The salience of the bacterial mode. Although any designation of most salient features must reflect the interests of the observer, I challenge anyone with professional training in evolutionary theory to defend the extending tip of the right tail as more definitive or more portentous than the persistence in place, and constant growth in height, of the bacterial mode. The recorded history of life began with bacteria 3.5 billion years ago, continued as a tale of prokaryotic unicells alone for probably more than a billion years, and has never expe­rienced a shift in the modal position of complexity. We do not live in what older books called “the age of man” (1 species), or “the age of mammals” (4000 species among more than a million for the animal kingdom alone), or even in “the age of arthropods” (a proper designation if we restrict our focus to the Metazoa, but surely not appropriate if we include all life on earth). We live, if we must designate an exemplar at all, in a persisting “age of bacte­ria” — the organisms that were in the beginning, are now, and probably ever shall be (until the sun runs out of fuel) the dominant creatures on earth by any standard evolutionary criterion of biochemical diversity, range of habi­tats, resistance to extinction, and perhaps, if the “deep hot biosphere” (Gold, 1999) of bacteria within subsurface rocks matches the upper estimates for spread and abundance, even in biomass (see Gould, 1996a, for a full develop­ment of this argument). I will only remind colleagues of Woese's “three do­main” model for life's full genealogy (see Fig. 9-30), a previously surprising but now fully accepted, and genetically documented, scheme displaying the phylogenetic triviality of all multicellular existence (a different issue, I fully admit, from ecological importance). Life's tree is, effectively, a bacterial bush. Two of the three domains belong to prokaryotes alone, while the three king­doms of multicellular eukaryotes (plants, animals, and fungi) appear as three twigs at the terminus of the third domain.

  2. The cause of the bacterial mode. “Bacteria,” as a general term for the grade of prokaryotic unicells lacking a complex internal architecture of organelles, represent an almost ineluctable starting point for a recognizable fos­sil record of preservable anatomy. As a consequence of the basic physics of self-organizing systems and the chemistry of living matter — and under any [Page 899] popular model for life's origins, from the old primordial soup of Haldane and Oparin, to Cairns-Smith's clay templates (1971), to preferences for deep-sea vents as a primary locale — life can hardly begin in any other morphological status than just adjacent to what I have called (see Fig. 9-29) the “left wall” of minimal conceivable preservable complexity, that is, effectively, as bacteria (at least in terms of entities that might be preserved as fossils). I can hardly imagine a scenario that could begin with the precipitation of a hippopotamus from the primordial soup.

  Once life originates, by physicochemical necessity, in a location adjacent to this left wall (see Kauffman, 1993), the subsequent history of right-skewed expansion arises predictably as a fundamental geometric constraint of this initial condition combined with the principles of Darwinian evolution — that is, so long as the most genuine trend of life's history then prevails: “success” measured variationally, in true Darwinian fashion, as expansion in diversity and range through time.

  If life continues to add taxa and habitats, then structural constraints of the system virtually guarantee that a right tail of complexity will develop and in­crease in skew through time as a geometric inevitability, and not necessarily for any overall advantage conferred by complexity. As noted above, life must begin, for physicochemical reasons, next to the left wall of minimal com­plexity. Little or no “space” therefore exists between the initial bacterial mode and this natural lower limit; variation can expand only into the “open” domain of greater complexity. The vaunted trend to life's increasing complex­ity must be reconceived, therefore, as a drift of a small percentage of species from the constant mode of life's central tendency towards the only open direction

  9-30. In life's full genealogy, all three multicellular kingdoms grow as twigs at the terminus of just one branch among the three great domains of life's history. The other two domains are entirely prokaryotic. From work of Woese and col­leagues, as presented in Gould, 1996a.

  [Page 900]

  for expansion. To be able to formulate this alternative view at all, we must reconceive the history of life as expansion and contraction of a full range of taxa under constraints of systems and environments, rather than as a flux of central tendencies, valued extremes, or salient features.

  3. The right tail as predictable, but passively generated. A critic might respond that he accepts the reformulation but still wishes to assert a vector of progress as life's central feature in the following, admittedly downgraded, way: yes, the vector of progress must be construed as the expanding right tail of a distribution with a constant mode, not as a general thrust of the whole. But this expanding tail still arises as a predictable feature of the system, even if we must interpret its origin and intensification as the drift of a minority away from a constraining wall, rather than the active trending of a totality. The right tail had to expand so long as life grew in variety. This tail therefore originated and extended for a reason; and humans now reside at its present terminus. Such a formulation may not capture the full glory of Psalm 8 (“Thou hast made him a little lower than the angels”), but a dedicated anthropocentrist could still live with this version of human excellence and dom­ination.

  But the variational reformulation of life's system suggests a further implication that may not sit well with this expression of human vanity. Yes, the right tail arises predictably, but random systems generate predictable consequences for passive reasons — so the necessity of the right tail does not imply active construction based on overt Darwinian virtues of complexity. Of course the right tail might be driven by adaptive evolution, but the same configuration will also arise in a fully random system with a constraining boundary. The issue of proper explanations must be resolved empirically.

  By “random” in this context, I only mean to assert the hypothesis of no overall preference for increasing complexity among items added to the distribution — that is, a system in which each speciation event has an equal probability of
leading either to greater or to lesser complexity from the ancestral design. I do not deny, of course, that individual lineages in such systems may develop increasing complexity for conventional adaptive reasons, from the benefits of sharp claws to the virtues of human cognition. I only hold that the entire system (all of life, that is) need not display any overall bias — for just as many individual lineages may become less complex for equally adaptive rea­sons. In a world where so many parasitic species usually exhibit less complex­ity than their free living ancestors, and where no obvious argument exists for a contrary trend in any equally large guild, why should we target increasing complexity as a favored hypothesis for a general pattern in the history of life?

  The location of an initial mode next to a constraining wall guarantees a temporal drift away from the wall in random systems of this kind. This sit­uation corresponds to the standard paradigm of the “drunkard's walk” (Fig. 9-31), used by generations of statistics teachers to illustrate the canonical ran­dom process of coin tossing. A drunkard exits from a bar and staggers, en­tirely at random, along a line extending from the bar wall to the gutter (where he passes out and ends the “experiment”). He winds up in the gutter on every [Page 901] iteration of this sequence (and with a predictable distribution of arrival times) simply because he cannot penetrate the bar wall and must eventually “reflect off” whenever he hits this boundary. (Of course, he will also end up in the gutter even if he moves preferentially towards the bar wall; in this case, the average time of arrival will be longer, but the result just as inevitable.)

  The issue of active drive (a small bias in relative frequency fueled by the general Darwinian advantages of complexity) vs. passive drift (predictable movement in a random system based on the model of the drunkard's walk) for the expansion of the right tail must be resolved empirically. But the macroevolutionary reformulation of life's history in variational terms estab­lishes a conceptual framework that permits this question to be asked, or even conceived at all, for the first time. Initial studies on mammalian vertebrae and teeth, foraminiferal sizes, and ammonite sutures have been summarized in Gould, 1996a, based on pioneering studies of McShea, Boyajian, Arnold, and Gingerich. This initial research has found no departure from the random model, and no overall preference for increase in complexity in studies that tabulate all events of speciation.

  General rules. The older literature of paleobiology focused on the recognition and explanation of supposedly general “rules” or “laws” regulat­ing the overt phenomenology of life's macroevolutionary pattern. As the modern synthesis developed its core of Darwinian explanation, several lead­ing theorists (see especially Haldane, 1932, and Rensch, 1947, 1960) tried to render these laws as large-scale expressions of evolution's control by adaptive anagenesis in populations under Darwinian natural selection.

  This subject fell out of favor for several reasons, but in large part because non-adaptationist explanations deemed less interesting (and certainly less coordinating) than accounts based on natural selection, provided an ade­quate compass for most of these “laws.” Thus, for example, Dollo's law of irreversibility

  9-31. The standard statistical model of the “drunkard's walk” shows that even the expanding right tail of life's right-skewed histogram of complexity may arise within a random system with equal probabilities for the movement of any descendant towards either greater or lesser complexity. From Gould, 1966a.

  [Page 902]

  (see Gould, 1970b) only restates the general principles of mathe­matical probability for the specific case of temporal changes based on large numbers of relatively independent components. And Williston's law of reduc­tion and specialization in modular segments may only record a structural constraint in random systems, thus following the same principles as my previ­ous argument about the expanding right tail of complexity for life's total­ity. Suppose that, in overall frequency within the arthropod clade, modu­lar species (with large numbers of similar segments) and tagmatized species (with fewer fused and specialized groupings of former segments) always en­joy equal status in the sense that 50 percent of habitats favor one design, and 50 percent the other. (I am, of course, only presenting an abstract “thought experiment,” not an operational possibility for research. Niches don't exist independent of species.) But suppose also that, for structural reasons, modu­lar designs can evolve toward tagmatization, but tagmatized species cannot revert to their original modularity — an entirely reasonable assumption under Dollo's law (founded upon the basic statements of probability theory) and generalities of biological development. Then, even though tagmatization en­joys no general selective advantage over modularity, a powerful trend to tagmatization must pervade the clade's history, ultimately running to comple­tion when the last modular species dies or transforms.

  However, one of these older general rules has retained its hold upon evolutionary theory, probably for its putative resolvability in more conventional Darwinian terms of general organismic advantage: Cope's Law, or the claim that a substantial majority of lineages undergo phyletic size increase, thus im­parting a strong bias of relative frequency to the genealogy of most clades — a vector of directionality that might establish an arrow of time for the history of life.

  A century of literature on this subject had been dominated by proposed explanations in the conventional mode of organismic adaptation fueled by nat­ural selection. Why, commentators asked almost exclusively, should larger size enjoy enough general advantage to prevail in a majority of lineages? Pro­posed explanations cited, for example, the putative benefits enjoyed by larger organisms in predatory ability, mating success, or capacity to resist extreme environmental fluctuations (Hallam, 1990; Brown and Maurer, 1986).

  The speciational reformulation of macroevolution has impacted this sub­ject perhaps more than any other, not because the theme exudes any spe­cial propensity for such rethinking — for I suspect that almost any conven­tional “truth” of macroevolution holds promise for substantial revision in this light — but because its salience as a “flagship,” but annoyingly unre­solved, issue inspired overt attention. Moreover, the conventional explana­tions in terms of organismal advantage had never seemed fully satisfactory to most paleontologists.

  The rethinking has proceeded in two interesting stages. First, Stanley (1973), in a landmark paper, proposed that Cope's Law emerges as a passive consequence of Cope's other famous, and previously unrelated, “Law of the Unspecialized” — the claim that most lineages spring from founding species with generalized anatomies, under the additional, and quite reasonable, assumption

  [Page 903]

  9-32. Cope's Law shown, under a speciational perspective, as a differential movement of speciation events towards larger size from a constraining boundary imparted by a small founding member of the lineage. Adapted from Stanley, 1973.

  that the majority of generalized species also tend to be relatively small in body size within their clades.

  These statements still suggest nothing new so long as we continue to frame Cope's Rule as anagenetic flux in an average value through time — that is, as a conventional “trend” under lingering Platonic approaches to macroevolution. But when we reformulate the problem in speciational terms — with the history of a Cope's Law clade depicted as the distribution of all its species at all times, and with novelty introduced by punctuational events of specia­tion rather than anagenetic flux — then a strikingly different hypothesis leaps forth, for we now can recognize a situation precisely analogous (at one fractal level down) to the previous construction of life's entire history: an evolving population of species (treated as stable individuals), in a system with a left wall of minimal size (for the given Bauplan), and a tendency for founding members to originate near this left wall (Fig. 9-32).

  Therefore, just as for all of life in my previous example, if the clade pros­pers with an increasing number of species, and even if new species show no directional tendency for increasing size (with as many species arising smaller
than, as larger than, their ancestors), then the mean size among species in the [Page 904] clade must drift to the right, even though the mode may not move from initial smallness, just because the space of possible change includes substantial room in the domain of larger size, and little or no space between the founding lin­eage and the left wall. Thus, as Stanley (1973) stated so incisively, Cope's Law receives a reversed interpretation as the structurally constrained and passive evolution (of an abstracted central tendency, I might add) from small size, rather than as active evolution towards large size based on the organismic ad­vantages of greater bodily bulk under natural selection.

  But we must then carry the revision one step further and ask an even more iconoclastic question: does Cope's Law hold at all? Could our impressions about its validity arise as a psychological artifact of our preferential focus upon lineages that grow larger, while we ignore those that remain in stasis or get smaller — just as we focus on fishes, then dinosaurs, then mammoths, then humans, all the while ignoring the bacteria that have always dominated the diversity of life from the pinnacle of their unchanging mode throughout geo­logical time?

  Again, we cannot even ask this question until we reformulate the entire is­sue in speciational terms. If we view a temporal vector of a single number as adequate support for Cope's Law, we will not be tempted to study all species in a monophyletic clade that includes signature lineages showing the docu­mented increase in size. But when we know, via Stanley's argument, that Cope's Law can be generated as a summary statement about passively drifting central tendencies in random systems with constraining boundaries, then we must formulate our tests in terms of the fates of all species in monophyletic groups. Jablonski (1997) has published such a study for late Cretaceous mollusks of the Gulf and Atlantic coasts (a rich and well-studied fauna of 1086 species in Jablonski's tabulation) and has, indeed, determined that, for this prominent group at least, prior assertions of Cope's Law only represent an ar­tifact of biased attention (see commentary of Gould, 1997b). Jablonski found that 27-30 percent of genera do increase in mean size through the sequence of strata. But the same percentage of genera (26-27 percent) also decrease in mean size — although no one, heretofore, had sought them out for equal ex­amination and tabulation.