The Structure of Evolutionary Theory

by Stephen Jay Gould

  The first assertion holds that natural selection, operating primarily by biotic competition under ecological plenitude, will indeed generate a bias towards “progress” by granting a statistical edge to general mental and biomechanical improvement. The second assertion then holds that this micro-evolutionary edge should accumulate smoothly, through time and up lev­els, to yield the general vector of life's progress that Darwin described in his geological chapters (see quotes on pp. 467–479). Thus, refutation at the first end-member accepts full fractality and extrapolation from microevolutionary [Page 1322] generality to macroevolutionary pattern, but denies that microevolution works in Darwin's required manner. The opposite refutation of the second end-member fully accepts Darwin's argument for the operation of microevolution, but denies the extrapolationist premise of scaling in continu­ity to impart the same theme of progress to higher levels, and thus to life's broadest history.

  If only for the obvious reason that Darwinian selection has been so overwhelmingly validated, both empirically and theoretically, as a dominant mechanism of evolutionary change in populations at generational time scales, the first end-member refutation has not garnered much consideration or sup­port, although Raup's principled exploration (1991b, 1992, 1996), to be dis­cussed just below, deserves considerable respect and attention, if only to re­mind us that apparently absurd propositions may hold far more plausibility than our knee-jerk reactions allow. But the second end-member, based on the opposite premise of nonextrapolatability for a Darwinian mechanism fully valid at its primary, smallest scale, tier of time, has been — correctly in my judgment — the standard, if usually inchoate or inarticulated, view of paleon­tologists uncomfortable with the full sufficiency of microevolutionary princi­ples to explain the entire history of life. The scaling of time's tiers, in this sec­ond position (and to cite a pair of metaphors), is neither fractal nor isometric.

  Before defending this second position as the key to solving the paradox of the first tier, I should say that, despite some initial enthusiasm as our profes­sion first embraced the renewed respectability of catastrophism, I doubt that any paleontologist would now defend a dichotomous division of time into two tiers — an ordinary or “background” world, granted entirely to Darwin­ian mechanisms, between catastrophic episodes; and a few, but markedly ef­fective, momentary disruptions of this generality in episodes of mass extinc­tion. This model of an alternation between background and wipeout regimes presents far too simple a picture, while admittedly capturing the central prin­ciple of higher and rarer modes that must be titrated with ordinary Darwin­ism to generate the full pattern. As with the cognate theme of hierarchical lev­els of selection, explored and defended throughout this book, we must try to render time as a series of rising tiers, each featuring distinctive modes of evo­lution, and each functioning as a gatekeeper to bar full passage, a ringmaster to add new acts to the mix, and a facilitator to alter, in interesting ways, the expression of conventional Darwinism in its domain.

  Time's higher tiers, in other words, introduce causes and phenomena to ex­pand the modalities of evolution, not to restrict or refute the powerful Dar­winian forces that rise from the organismal level in ecological time, but do not maintain their pervasive sway in these broader realms. Again, as with ris­ing hierarchical levels of structure, our increasing understanding of evolution at time's upper tiers establishes the architecture of a larger, sturdier, and inter­estingly different Darwinian edifice, and does not operate as a demolition team for razing old domiciles and then building some hurried heap of super­ficial appeal, thrown up without a foundation, and therefore destined to top­ple as soon as the inevitable winds of fashion change their capricious course. [Page 1323]

  Fractal iconoclasm scaled down

  In evolutionary theory, the canonical observation of inverse relationship be­tween frequency of occurrence and intensity of effect has generally been used to deny the force, or even the existence, of the largest pulses as too rare to matter, even in the amplitude of geological time. (Fisher's citation, invoked to privilege small events by their overwhelming frequency, opens the general ar­gument in his Genetical Theory of Natural Selection (see pp. 508–514), and remains the classical example in Darwinian literature.) Interestingly, the ap­plication of this inverse relationship to assess the range of catastrophic mod­els had proceeded in precisely the opposite manner — from accepting the mag­nitude of a proven single event (the K-T extinction) at maximal scale, and then extrapolating down to ask whether much more common small events of the same type could yield enough oomph and frequency to provide an alter­native account for our Darwinian preferences at the ecological scale of natu­ral selection's supposed and undoubted domination.

  My dearest paleontological colleague David M. Raup has delighted me throughout my career, and kept an entire profession on its collective toes, by acting as Peck's Bad Boy to express outrageous and unthinkable ideas in the form of testable hypotheses. I confess that I have never quite figured out whether Dave believes in, or would place more than a minuscule probability upon, the hypotheses behind several of his tests; or whether he just loves to play the role that the Church wished to assign to Galileo — that is, to pre­sent the almost surely untrue as a hypothetical claim in mathematical form, thereby to sharpen our empirical and logical skills in finding best arguments for truthful propositions. Only once did I ever win an argument against one of his null hypotheses, or “beans in a bag” models, for random worlds com­posed of identical objects. Raup held for several years (before the Alvarez data convinced him of the reality of a K-T event) that mass extinctions might be entirely artifactual, representing only an occasional extreme in sampling from an actual set of equally sized extinction pulses. “But Dave,” I would say in frustration, “perhaps the Permo-Triassic granddaddy of all extinctions can be rendered statistically as no more than an extreme sample from a uniform pool. Still, you can't deny that, on Earth, the Triassic organisms that actually reappear are so different from their Permian forebears. Something 'real' must have happened then.” I think that he finally acquiesced to this point!

  Raup developed his extreme model as a thought experiment because mass extinction by bolide impact might, at least in principle, be regarded as ran­dom in both of Eble's (1999) senses previously called, in this particular con­text, the “random” and “different rules” models (see p. 1314) — that is, either truly so in the formal statistical sense, or only so in the vernacular sense that reasons for differential success in such catastrophically altered moments must be exaptive with respect to Darwinian bases for evolving the relevant features in the first place, and in background times. Raup therefore posed the follow­ing sly question: if this broad sense of randomness applies to the largest event, and if the famous inverse curve of frequency vs. effect implies a continuity in causality as well, then maybe we should extrapolate down and reconceptualize [Page 1324] the smaller and much more common extinctions as equally random in their raison d'etre (in opposition to the knee jerk view that local Darwin­ian determinism rules for ecological moments in competition by wedging, whereas randomness can only enter at higher levels, where the speed and in­tensity of an input can catch a Bauplane unawares).

  Such a model of fractal continuity in extinction, triggered by sudden im­pact at all scales and levels, might be conceptualized as a “field of bullets” (Raup, 1991a) — with agents of destruction raining from the sky and death as a random consequence of residence in the wrong place at the wrong time (when each member of the population expresses exactly the same properties as any other, and with each independent of all others). One might conceptual­ize the agents of catastrophic destruction (the field of bullets) in either of two ways:

  First, the random shooter in the sky may, for each episode of the game, re­lease a varying number of simultaneous bullets of identical form, with contin­ually decreasing probability of a larger number (following the inverse curve of frequency and magnitude). Thus, as a lazy or compassionate ch
aracter, he hurls only one bullet most often, but must occasionally release such a dense load that few inhabitants can escape annihilation. Thus, the vast majority of moments feature none, one, or just a few extinctions, easily equated with our usual idea of a “background,” but with causes just as random in their “selec­tion” of targets, and just as sudden in their effects, as in the largest event of mass extinction. Once in a great while, following the dictates of the same dis­tribution and its implied continuity in causality, bullets reach the extirpating density of the nearly continuous sheet of arrows launched by the English longbowmen at Agincourt in 1415, where the French suffered some 6000 deaths to an English handful. Second, the random shooter might always re­lease the same number of projectiles, this time following the inverse curve by using smaller bullets (covering a tiny percentage of territory) most of the time, and large bombs (flattening most of life's field) only rarely, at the much lower frequency of mass extinctions.

  In practice, Raup (1991b, 1992, 1996) derived a “kill curve” (his chosen term) from the empirical compendium of generic level extinctions per geolog­ical stage developed by J. J. Sepkoski, and widely used by the entire “taxon counting” school of modern paleobiology (see Figure 12-1). The frequency distribution, based on Sepkoski's data, assumes the expected inverse form, monotonic and strongly right skewed, with about half the 106 geological units (with their average duration of 6 million years) plotting in the leftmost interval, and showing less than 10 percent extinction of genera.

  Raup's kill curve (Figure 12-2) follows the familiar form (the inverse relationship of frequency and magnitude again) that generates such vernacular concepts as the “100 year flood” — so often, and so tragically, misunderstood by so many people who, for lack of education to undo one of the most stub­born of our inherent mental foibles, do not grasp the basic meaning of proba­bility and assume, for example, that they may safely build their house on the floodplain because the 100 year deluge swept through the region five [Page 1325] years ago and therefore cannot recur for almost another century. Raup freely admits that the orderly and monotonic form of the kill curve does not spec­ify any style of causality by itself — and, most relevantly and especially, does not permit an inference about random effects at lowest levels merely be­cause we can advance a powerful case for this style in one event at the highest level.

  Nonetheless, the numerical specifics of Earth's particular curve does impose

  12-1. From Raup, 1996. Typical negative correlation of intensity of extinction and frequency of occurrence, with small extinctions common and large events rare.

  12-2. From Raup, 1996. The “kill curve” derived from Figure 12-1, showing expected waiting time between extinction events of various magnitudes — with longer waiting times for larger events.

  [Page 1326]

  some restriction of possibilities, and does suggest some causal infer­ences. Raup notes (1996), in particular, that if each species, following Dar­win's explicit claims (cited at the beginning of this chapter), pursued its own independent history of origin, expansion, reduction and death in terms of its own special competitive prowess vs. other unique species in singular faunas — thus implying that no general cause can impact all species at once (or at least that such general causes only nudge, and do not set or determine, the overall pattern) — then the kill curve, while continuing to obey its predictable mono-tonic form, would never reach such high percentages of death in its rarest up­per episodes. Such concentration in mass extinction does imply a coordinated cause of some sort. Raup writes (1996, p. 422):

  If each genus died out independently of the others, the probability of producing a range of from near zero to 52 percent extinction in 106 six million year stages would be negligible. This means that pulses of extinc­tions of genera must be connected in some way . . . because of common factors, such as ecological interdependence or shared physical stress. We thus see a picture of episodic extinction wherein the more intense an ex­tinction episode, the rarer it is. To describe extinction only as back­ground or mass extinction, as is commonly done, is to hide much of the structure of the extinction phenomenon.

  The nonfractal tiering of time

  Strict Darwinism implies continuity in the style and causal structure of change from successive generations in populations to the waxing and waning of faunas across geological eras. An alternate construction of time as a series of discrete tiers, or at least of rising “regions of coagulation” that pull phe­nomena away from boundaries and towards more central nucleating places — with each tier then featuring different weights and styles, or even truly dis­tinct modes, of causality — would seriously challenge the crucial extrapolationist premise of Darwinian logic (the third leg on my tripod of support). Al­though I know the quibbles and inconsistencies, and I recognize that many of my neontological colleagues regard such problems as central flaws worth a lifetime's research, I have no personal quarrel with Darwin's argument that a vector of general progress would pervade the history of life if all scales of time record the competitive styles of natural selection supposedly prevailing at the first tier of anagenetic change within the history of single populations. We should therefore take firm notice, and regard as highly paradoxical, the fail­ure of life's history to feature such a vector as an obvious organizing principle and predominant signal of phylogeny.

  I reject, for the most part, the Raupian solution of fractality in time, with Darwinian inefficacy throughout. Instead, I favor the alternative view that Darwinism basically works as advertised in its own realm at the first tier of time, but cannot “push through” to impose its characteristic signal upon pro­cesses and phenomena of higher tiers. Such a proposal implies a very different attitude towards time and change — an attitude inspired and encouraged by [Page 1327] two controversial topics of the last quarter of the 20th century: punctuated equilibrium and catastrophic mass extinction, our two most notorious hy­potheses about the non-homogeneity of time's principal modes.

  In proposing that we conceptualize time as a rising set of tiers, I do not argue (thus hoping to forestall, by this explicit statement, the same mis­understanding kindled by the debate about hierarchical levels of selection) that entirely new, and truly anti-Darwinian, forces emerge at each higher tier. I am quite content to allow that no fundamental laws of nature, and any en­tirely novel causes or phenomena, make their first appearance in larger slices of time. But, at these broader scales and intervals, the known principles of genetics, and the documented mechanisms of selection, may operate by dis­tinct and emergent rules that, as a consequence of time's tiering, cannot be fully predicted from the operation of the same kinds of causes at lower levels. The logic of this critique flows from potential fallacies in Darwinian assump­tions about extrapolation across time's putative smoothness and causality's supposed invariance. (Selection on species-individuals, for example, follows all the abstract and general principles required by Darwinism for this central mechanism of the general theory, but the modes and regularities of selection at this higher level differ strongly, and cannot be predicted, from our canoni­cal understanding of Darwinian organismal selection within populations — see Chapter 8.)

  The dilemma, and eventual insufficiency of the Modern Synthesis for paleontology lay in this third crucial Darwinian claim that all theory could be extrapolated from the first tier, thus converting macroevolution from a source of theory into a pure phenomenology — a body of information to document and to render consistent with a theoretical edifice derived elsewhere. But if the tiers of time, and the hierarchy of life's structure, create pattern by emer­gent rules not predictable from processes and activities at lower tiers, then pa­leontology will add insights, and augment theory, without contradicting the principles of lower tiers.

  We need to distinguish between the stability and continuity of causal prin­ciples (the spatiotemporal invariance of natural law in our usual jargon) and the potentially discontinuous (and disparate) expression of these principles across a spectr
um of time that may be strongly stepped (as nucleating points attract surrounding events, and as nature's inherent structure clears out space at the unstable positions between), rather than fully and smoothly continu­ous. I began my original article (Gould, 1985a, p. 2) on the tiering of time with an admittedly humble analogy that may still help to clarify the impor­tant principle of noncontradiction between structural tiering and a unitary order of underlying entities and causes:

  In the glory days of Victoria's reign, when a pound bought more than five American dollars, the English economy operated on two distinct tiers. The working man, paid weekly for his labor and without benefit of banking or hope of accumulation, might pass his entire life without ever seeing a pound note, for he received his wage in shillings and pence and [Page 1328] its total never reached a full pound. Meanwhile, bankers in the City of London transacted the world's imperial business in pounds. India today operates on two similar and largely noninteracting tiers — the 100-rupee notes (about ten dollars) of the hotel shops and the bustling economy of the bazaars, where 10 rupees buy at least one of anything and no one ever sees (or could cash) a 100-rupee note.

  Our world of times and amounts is not always continuous. Its metrics usually extend smoothly from one end to the other (shillings did grade to pounds and rupees are rupees), but its activities are often sharply con­centrated in definite regions of a potential spectrum, with large open spaces between. Systems often drive in opposite directions away from break points; location on one or another side of a threshold inevitably pushes toward an equilibrium far above or below.