A while back I listened to Lex Fridman’s podcast with Michael Levin (episode #325 Oct 1, 2022). I immediately wrote down some notes and recently was queried by my old friend Ben Barnett—my college roommate and teammate—about what I thought on Levin’s ideas. Therefore, I assembled my notes here. I have focused my attention on some of Levin’s broader speculations about evolution, in which he argues that evolution works for ultimate goals. For example, Levin argued that evolution does not adapt a frog to its immediate “froggy” environment, but that it works towards “long-term attractors” like “optimizing for biomass.” Fridman said, “It’s almost going to be hilarious a few centuries from now when they look back how dumb we were.” Levin agreed. The problem is that Levin’s speculations already look pretty dumb.
Here are my specific critiques of Levin’s arguments:
(1) I agree with Levin that we can analyze complex systems at multiple levels, and that it is best to choose the levels that allow the simplest models yielding useful predictions. In doing so, however, it is important to keep in mind what sorts of questions that we are asking. For example, if we want to know why a plasmodial slime mold fuses with itself following fragmentation, rather than proceeding to the nearest food source, the best approach is to consider the biology and evolutionary history of this organism. In its natural environment, if two Physarum polycephalum plasmodia come into contact, they will begin to fuse initially and then one of two different things will usually happen. First, if they are different clones, they will undergo an immediate rejection reaction, halting fusion, and thereafter will typically fight with each other until one either retreats or is dead. Second, if they are the same clone, they will typically fuse. A third and less common event is that the two clones are unrelated and share the outer plasmodial compatibility fusion loci. They will then fuse initially but still undergo an aggressive internal rejection reaction after fusion, and one clone will kill the other.
To understand this behavior requires an evolutionary model capable of explaining the complex genetic kin recognition system that allows plasmodial compatibility. Such a model predicts a historical sequence of events including the evolution of fusion, the evolution of differential treatment, the evolution of fusion compatibility, a switch in cues used, and the elaboration of polymorphism (Gilbert 2015, 2017). Such a model suggests that the reason that fragments of the same clone of P. polycephalum fuse is that the benefits of fusion outweigh the costs and there is some positive encounter rate with clonal fragments. Encounters with clonal fragments occur in part because plasmodia do sometimes spontaneously fragment, and in part because they can be broken apart by external physical forces. Benefits of fusion with the same clone can include size advantages, competitive ability with other clones of the same species or different species, benefits of resource sharing, and avoidance of competition with physically discrete fragments of the same clone.
(2) In Levin’s analysis of plasmodial slime molds, he viewed a new fragment of the same clone as a separate “agent.” He then asked whether this new entity should try to capture the food itself or fuse first. In his paper on the subject, he said, “prior to a join, a selfish agent should want to keep the food to itself, but after the join, it will not exist, so there is no sense in which it can compete or share resources with ‘the other’.” The authors then hypothesized that “after cutting, the small piece would proceed to exploit the food.” They then “disproved” the hypothesis, by showing that the small fragment fuses back to the original first. They therefore concluded that Physarum prefers the long-term benefit of fusing over the short-term benefit of monopolizing the resources. This seemed to be in line with Levin’s greater teleological outlook on evolution, in which “agents” behave with foresight to future effects rather than immediate environmental demands.
The problem with Levin’s conclusion is that from an evolutionary standpoint, there really is no such thing as “separate interests” when individuals share the exact same genomes. Rather, the most likely the reason that the P. polycephalum clone fused first with a clonemate is that it is not really a separate evolutionary “agent.” This is a consequence of a history of evolution in which natural selection has favored fusion with clonemates, which typically has immediate benefits of increased size, ability to share resources, and so on. Therefore, there is no sense in which this experiment suggests that agents behave in a teleological way with ultimate goals in mind.
(3) As much Levin talks about the origin of a “self,” he does not mention the existence of self/not-self recognition systems (or genetic kin recognition systems) common to many organisms that have the potential to fuse or aggregate with unrelated conspecifics. He seems to be completely ignorant of plasmodial compatibility systems. Levin seems to think that “self” has more to do with control and agency, rather than genetics. He also overlooks evolutionary theory grounded in genetics that explains why genetic kin recognition systems evolve. On the subject, he says, “You can imagine a kind of game theory in which the number of agents is not fixed and it is not just cooperate or defect but merge and whatever…I am not an expert in economic game theory…but maybe this idea that the actions you take not only change your payoff but they change who or what you are….you could take an action after which you are merged with somebody else…as far as I know we are still missing a formalism for even knowing how to model any of that.” On this end, Levin totally ignores my work (Gilbert 2015), which specifically addresses how to model fusion compatibility systems from a game-theory framework that allows for separate cooperate-defect and fusion-rejection type strategies. On the bright side, he at least recognized that this type of theoretical work would be valuable.
(4) Levin uses teleology to mean the study of design or purpose in nature. In contrast, teleology often refers to the doctrine that final causes exist or that apparent design is evidence of final causes for existence. These latter definitions of teleology reflect the philosophy guiding Aristotle’s Parts of Animals and Paley’s Natural Theology. It is also a common cognitive bias of humans. Even children tend to assign “purpose” to organismal traits and assume that the cause for existence is manifest by apparent design. Many evolutionists even fall victim to this same logical fallacy.
(5) There is a good reason why some biologists, including myself, express disdain for teleology. The reason is that the more complex an organismal trait is, and the more interactive subparts it contains, the less likely they all originated for their final apparent goal as part of a complex whole. In contrast, anyone adopting a teleological view will take increasing complexity of design as evidence of adaptive value, thus being led to the exactly wrong conclusion. Often, such researchers will then invoke models of saltational evolution, in which multiple sub-traits co-evolve simultaneously. In the history of evolutionary biology, however, saltational models of evolution have proven much less fruitful in general than those based on gradualism. This is because under an assumption of gradual evolution, researchers are forced to break complex traits into simpler components and derive stepwise historical models to explain how they might have originated. These models are often tested through the comparative approach, which draws on phylogenetics and taxonomy of trait distributions to infer sequences of evolutionary steps leading to complex traits. Particular hypotheses for the origin of each subtrait can then be tested experimentally.
(6) Levin defines teleology in a way that produces a “defensible” version of the term. He then makes arguments employing an indefensible use of the term. Particularly, Levin answers why questions with how questions, which assumes that the causes for existence are found in apparent design. Therefore, his rhetoric leads him to the same erroneous conclusions that have arisen periodically throughout the history of evolutionary thought—from orthogenesis to group selection.
(7) One of the main assumptions of scientific investigation is that the causes of events are antecedent. If a helium atom forms, it is because earlier two hydrogen atoms fused, not to facilitate the ultimate formation of galaxies and life. As soon as we begin to assume that the causes of events are in the future, we begin to leave the realms of science, and enter the realms of metaphysics and religion. We sometimes might take “short cuts” that are useful for understanding complex systems, for example assuming that a trait has a simple genetic basis and evolved in a single step. This is sometimes called “the phenotypic gambit” paradigm. The paradigm is popular in social theory, but it has severe limitations. Those who employ it have sometimes failed to provide adequate explanations of natural phenomena that are more genetically complex than they originally assumed. More generally, Darwinian theory is based on the very opposite of a teleological worldview. Optimization logic only works in general only when it is applied to a small enough genetic unit. The safest bet is to restrict one’s attention to the smallest replicating unit possible, the allele, and compare its fitness to alternative alleles.
(8) Dawkins once said that “Just as one can put oneself in the position of an imaginary light beam, intelligently choosing the optimal route through a cascade of lenses and prisms, or an imaginary gene choosing an optimal route through the generations, so one can postulate an individual lioness, calculating an optimal behavioural strategy for the long-term future survival of her genes.” The fallacy in this comparison is obvious. While it might be useful to assume that a light beam intelligently chooses an optimal route through a cascade of prisms, it is not useful to assume that a gene chooses an optimal route through the generations. Instead, we must focus on why a particular allele—a variant of a gene—is favored at a particular moment in time, given its genetic background and environment. Only by focusing on the immediate context of adaptation can we understand how evolution builds complex traits over time. In biology, if we are to assume that some trait exists because of its final apparent purpose, we are led into the pre-Darwinian paradigms of Aristotle and Paley. Darwin revolutionized this teleological worldview.
(9) Levin speculates about long-term goals of evolution. He claims that evolution does not produce solutions to particular problems. Levin argues that evolution does not adapt a frog to its immediate “froggy” environment. But then why are there no frogs in saltwater? Why would evolution produce a dodo bird if evolution has goals and anticipates the future? Why are there are so many examples of evolutionary dead ends? Why are over 99% of species extinct?
(10) Levin also throws in some odd claims, like that evolution “optimizes for biomass” but not for “intelligence” or “fairness.” Why, then, has evolution produced cooperative and intelligent organisms? What does it mean to “optimize for biomass”? Is the contention that there are trends towards large organisms or small organisms? If so, then why does evolution sometimes produce larger organisms and sometimes smaller organisms? Why have some species of multinucleate fungi later become uninucleate (yeasts)? Why are parasites often the first novel forms to originate in computer simulations of populations of replicating codes?
(11) Levin claims that evolution requires resource limitation but that it does not produce solutions to particular problems. This makes absolutely no sense, because it is the assumption of resource limitation that tends to produce context-dependent adaptation that adapts species to their immediate environments. Darwin conceptualized natural selection as occurring when species are checked from increasing, when limitations imposed by food, water, and predators and parasites keep populations steady. The consequence is that more are born than can survive, and that natural selection operates on small trait variations that better adapt species to their immediate situations. A frog becomes adapted to its immediate freshwater pond environment, not the oceans and deserts of the world. Natural selection works only for immediate benefits, not ultimate goals.
(12) I agree with Levin that science has much to learn about the broad questions of evolution. However, some of what Levin says about emergence and multilevel causality is not as novel as it sounds. For example, Levin emphasized that biological systems take advantage of emergent laws that cannot be predicted from lower levels. He contends that evolutionists have tended to ignore this fact. But if that were true, then why have evolutionists bothered to derive a theory based on properties at levels higher than physics and chemistry, based on the details of organismal reproduction, variation, and inheritance? Why have evolutionists bothered doing experiments on actual living organisms, rather than just predicting everything from physics and chemistry? The simple fact is that the theory of natural selection is itself based on emergent properties of living organisms, and every detailed model of biology incorporates a myriad of emergent properties of populations and genetics.
(13) Levin does well to draw attention to planaria, which is a very interesting oddball organism because it can reproduce by fragmentation. It reminds me of Trichoplax or a honeybee colony. To explain such modes of reproduction, Levin treats the entire organism as an “agent” or unit of reproduction. However, the fallacy of this approach can be shown by studies of honeybees. Like planaria or Trichoplax, honeybees reproduce by fission of groups. Honeybee colonies have much genetic heterogeneity within them, because the queen mates with many males and the various workers are mostly half-sisters. Honeybee colonies appear organismal or “agential” because of their strong coordination of parts and intelligent behavior. However, the best models showing how that this organismal behavior originates focus on how that the individuals within colonies interact. Such models view the individuals within colonies as “agents,” and they predict an evolutionary sequence of events including the origin of multiple queen mating, the origin of worker reproduction, the origin of worker policing, and the origin of self-restraint on worker reproduction because of the threat of worker policing. The end result was a system that functions in a cohesive and unified manner. However, nobody would ever be led discovery the cause of honeybee “organismality” by models that view the colony as a whole as an “agent” as Levin does.
(14) I am not ready to buy Levin’s argument that the reason hierarchical cognitive systems are not simpler is that organisms have evolved to prevent themselves from being hijacked. Again, this appeals to a final cause without first examining how evolution could get there.
(15) Computer scientists and engineers like Lex Fridman are very enamored with biologists who study the functional design of biological systems but not their evolution. This seems quite natural, because computer scientists and engineers build things with goals in mind. But does it follow therefore that evolution has goals, or that we can understand evolution better by assuming that it works with ultimate goals in mind? This question can be turned on its head by imagining that we ask an engineer to build a honeybee colony. Applying Levin’s “agential” thinking to honeybee colonies as a whole, it would be very difficult to know where to start. Would the colony as a whole be assumed to be free of internal conflict? With knowledge of evolution, we would be ignorant of the fact that honeybee colonies are composed of interacting agents with different evolutionary interests. We would have trouble reconstructing a honeybee colony and programing the agents to behave like really honeybees.
(16) In the Genealogy of Morals, Nietzsche recognized that the confusion between historical origin and current utility of punishment obscured an understanding of why punishment exists. Nietzsche said that authors analyzing the subject often highlight some apparent purpose, manifest in its current utility, and naively take this purpose as an explanation for the origin of punishment. Nietzsche concluded that, “the origin and emergence of a thing and its ultimate usefulness, its practical application and incorporation into a system of ends, are toto coelo separate.” Nietzsche highlighted this distinction as the “major point of the historical method.” Indeed, it is also the primary point of Dawin’s work, as emphasized in The Origin.
(17) In the example of honeybees, the teleological fallacy might conclude that honeybees police each other “for the good of the colony.” The reality is that they are less related to their half-sister’s daughters than to the queens’ daughters. Therefore, workers kill the eggs laid by other workers, but not those laid by the queen, not because it is “good for the colony” (even though it is), but because it benefits their own selfish-genetic interests. In some other bees in the genus Melipona, queens and workers are the same size and workers cannot tell which developing cells contain developing queens. Because it is in the larvae’s genetic interests to develop as queens, the colonies produce a great excess of queens, most of which are then swiftly executed by workers. In that case, the selfish-genetic behavior of interacting agents does not produce an outcome that is “good for the colony.” In other words, evolution by natural selection does not have “ultimate goals” in mind.
(18) The reality is that evolution by natural selection optimizes only in a very local sense. In my 2020 paper, I went into some detail showing examples of complex traits that appear designed for a purpose, but which did not evolve for that purpose. An example of bacterial citrate metabolism that evolved in the laboratory. The historical record showed that most selective steps leading to the novel metabolism ability had nothing to do with the final apparent goal of metabolizing citrate. Nevertheless, once in existence, the metabolic ability influenced ecological expansion. Another example comes from my own work on the evolution of kin recognition, which also showed a series of selective steps that had nothing to do with the final apparent purpose of the trait. These examples show that teleology should be abandoned as a guiding framework for understanding complex traits.
(19) Levin argues for what he calls “changes in software.” Davidson and Erwin identified changes at different levels of the genetic hierarchy, like gene regulation or core “kernels” of regulatory networks. Natural selection typically acts through allele frequency change and does not usually alter core “software.” But such changes have occurred and must be incorporated into a broader non-teleological theory of macroevolution.
Conclusion
Ernst Mayr observed, “It is remarkable how often a person who is trying to solve a particular evolutionary problem goes through the same sequence of unsuccessful attempts to find the solution, as has the whole field of evolutionary biology in its long history.” In this case, the problem is why the history of life is characterized by a number of major macroevolutionary trends, and the same teleological theories have reappeared through the years to explain them. We first had orthogenesis, then aristogenesis, then Fisher’s theory of organismal fitness increase, then Wynne Edward’s group selection, then Wright’s fitness maximization, then Hamilton’s inclusive fitness maximization, and finally Dawkins’s appeal to long-term gene survival. Each new theory has attempted to justify teleology, the doctrine of taken apparent design as the cause for existence. Such theories have frequently led to saltational models of evolution, pseudoscientific research programs, and the idea that evolution works towards ultimate goals.
Although I disagree with Levin’s teleological worldview, I agree that evolution has a sort of long-term attractor, which is a tendency towards increased innovativeness. Indeed, the trends in life’s history have been increased command over resources, energy flux, biomass, and species diversity and complexity. But this is not because evolution has “ultimate goals.” Instead, as I have argued, there are two evolutionary forces—natural selection, which works like nature’s inventor, and natural reward, which acts like nature’s entrepreneur. Darwin’s theory ignored the latter force because it assumed that species are checked from increasing in population size. Under an expanded theory, novel inventions allow species to escape normal Darwinian “checks to increase” and the resulting transient population increases result in a different type of competition. Just as the first company to invent “Amazon” or “YouTube” gains an incumbent advantage in market space, so the first organisms that radiate into an environment are protected from future competitors by the mere fact of their ecological dominance. Their underlying genetic system is then naturally rewarded with an incumbent advantage. Over vast time frames, through repeated bouts of invention-conquest macroevolution and extinction-replacement megaevolution, natural reward leads to build up of innovation-enhancing factors that advance life with time.
The primary fallacy of teleological theories has been to confuse the roles of natural selection and natural reward. As soon as we try to make natural selection into the force that rewards innovation, we begin to use natural selection incorrectly. We begin to assume the complex traits are optimized for ultimate goals. The reality is that complex traits often originate randomly with respect to their final effects on ecological expansion, and natural reward operates as a second non-random force, which operates on the random “variation of invention” produced by natural selection. Natural selection can just as easily produce a duck-billed platypus, with is webbed feet, duck bill and stingers, as a human, with its fingers, larynx and large brain. Natural selection did not produce fingers, larynxes, and large brains so that humans would invent capitalistic economic systems. Nevertheless, appearance of capitalistic economic systems triggered the largest expansion in human population in the history of the genus Homo, leading to the origin of the geological epoch most correctly labeled the Capitalocene.
In summary, the history of life on earth has produced trends that give the appearance of teleological foresight. However, a proper understanding of the history of life requires understanding the different roles of natural selection as a force of inventiveness and natural reward as a force of entrepreneurship (the non-inventive part of innovation or spreading the inventions to the “markets” they happen to be able to exploit). The overuse of optimization reasoning, borrowed from engineering, and teleological “final causes” has produced a myriad of teleological theories of macroevolution, of which Levin’s is the latest edition.
References
Gilbert, O. M. (2015). Histocompatibility as adaptive response to discriminatory within-organism conflict: A historical model. The American Naturalist, 185(2), 228-242.
Gilbert, O. M. (2017). Association theory: a new framework for analyzing social evolution. bioRxiv, 197632.
Gilbert, O. M. (2020). Natural reward drives the advancement of life. Rethinking Ecology, (5).
Levin, S. P., & Levin, M. (2021). Physarum polycephalum: Establishing an Assay for Testing Decision-making Under Shifting Somatic Boundaries. bioRxiv, 2021-10.