Is Cosmological Natural Selection an example of extended Darwinism?

Lee Smolin

by Massimo Pigliucci

Daniel Dennett, in his famous Darwin’s Dangerous Idea, put forth the proposition that “Darwinism” is a “universal acid” that “eats through just about every traditional concept, and leaves in its wake a revolutionized world-view, with most of the old landmarks still recognizable, but transformed in fundamental ways.” I tend to be cautious about universal statements of any kind, particularly when it comes to evolutionary theory, and especially if they are made by non-biologists (for instance, I criticized Addy Pross for a paper published in the Journal of Systems Chemistry that attempted — improperly, I think — to expand Darwinian selection to chemical “evolution”). So, while Dennett’s book provides plenty of stimulating food for thought, I’m not so sure that Darwinism amounts (or should amount) to a “worldview.”

Be that as it may, it is because of similar worries that I have intended for a while to take a look at Lee Smolin’s theory of Cosmological Natural Selection (CNS), which is by far the most ambitious attempt so far to generalize Darwinism to the whole cosmos — indeed, to the full multiverse! I have a lot of respect for Smolin (though we haven’t met, yet), particularly for his lucid and insightful book on The Trouble with Physics, which is not only a critique of superstring theory, but a scientific, sociological and philosophical survey of the current status of fundamental physics (and the picture ain’t pretty, according to Smolin). Still, I do have some qualms about the analogy between CNS and the biological theory of natural selection (BNS).

Let me be clear on what I am about to attempt in this post. This will not be a critical evaluation of CNS qua theory in the physical sciences. I am simply not qualified to do so, and I leave that task to Smolin and his colleagues (though I must admit that I like the idea of a mechanism along the lines of what Smolin presents). Rather, I will focus only on the parallel — which Smolin draws explicitly — between CNS and BNS. I will argue that such parallel should be understood at best as a loose metaphor, not as a scientific expansion of Darwinism. Also before getting started, one might reasonably wonder why bother making distinctions of the kind I am about to make. Several reasons: first, this is what philosophers do; second, it seems to me that conceptual clarity is intrinsically a good thing (see first point); third, I am arguing against unwarranted expansions of a given scientific theory because of my belief that scientific theories have proper domains of application, and that no theory is “universal” (no, not even the famous “theory of everything” that fundamental physicists are after, for reasons that I will not explore here but that are related to my recent posts on Ladyman and Ross’s naturalistic metaphysics); fourth, because perhaps conceptual clarifications will even turn out to be useful to the scientists involved, especially those like Smolin who are philosophically sensitive (as opposed to those who aren’t).

Now, then, the next step is to get a minimal understanding of CNS, which is not necessarily an easy task. Above I linked to a nice paper by Smolin himself, which can be read and largely followed even if you are not a fundamental physicist. Smolin also has a book out that expands on CNS, The Life of the Cosmos, and a new one coming out soon, entitled Time Reborn. I have also asked Smolin directly for clarification, as well as cosmologist Sean Carroll. They both graciously answered my queries and have authorized me to quote them briefly where needed.

Smolin’s theory is an attempt to solve the so-called special tuning problem, the idea that “both long lived stars and stable nuclei owe there existence to highly improbable tunings of the parameters that give large hierarchies of dimensionless parameters” (all quotes are from Smolin’s paper, unless otherwise specified). One answer to this problem is a version or another of the so-called anthropic principle, which invokes special considerations for the evolution of life, and which Smolin finds inelegant and untestable (and so do I). CNS was also generated as a way to prove that so-called “landscape theories” (like superstring theory), i.e. theories that generate a large landscape of possible physical instantiations of their parameters, can in fact be falsified, despite the apparently very large number of theoretical possibilities that characterize them, and therefore the looming specter of underdetermination by data.

Here is where BNS comes into play. As Smolin puts it: “[The situation with landscape theories in physics] is of course analogous to biology, in which many predictions could be made and tested without knowing the details, or even anything about, molecular genetics. Indeed, the use of biology as an analogue is suggested by the fact that evolutionary biology is the single case in science where a landscape problem and a special tuning problem have been solved successfully in a way that generated new falsifiable predictions. As argued, this is probably not accidental, hence it makes good sense to try to apply the schema of evolutionary biology to cosmology.” [Here I think Smolin is being a bit optimistic: natural selection seems to be capable of solving the landscape problem, in the sense of producing populations of fit organisms (albeit with a 99.9% extinction rate!), but biologists are often hard pressed to make predictions about evolutionary trajectories, and they often have to content themselves with postdictions, i.e., with tests that data about the past.]

Smolin’s hypothesis of cosmological natural selection is based on these three tenets (again, from the paper):

* The world consists of an ensemble E of universes, each of which is characterized by a point x ∈ L and hence by a point p = Ix ∈ P.
* Black hole singularities bounce and evolve to initial states of expanding universes. Hence there is a fitness function f on P where f(p) is equal to the average number of black holes produced by a universe initiated in such a bounce transition, which has parameters p.
* At each such creation event there is a small change in x leading to a small random change of x in P. Small here means compared to the step size under which f(p) changes appreciably.

In plainer language, collapsing black holes spawn baby universes, and these newly formed universes are characterized by combinations of parameters similar but not identical to the one of the “mother” universe. The process therefore includes physical analogies of both reproduction (new universes are formed) and mutation (the physical parameters of the progeny universes is variable). The idea is that some of these universes will “die” before reproduction is possible, while others will reproduce prolifically, and that the difference is due to the physical parameters of each universe. The fact that we find ourselves in a universe with certain parameters, therefore, is not an accident (nor, contra the anthropic principle, is the result of an alleged necessity for life — though, crucially, CNS does generate universes with carbon chemistry, which is made possible by the existence of massive stars, which are in turn necessary for the production of black holes). Rather, it is simply the result of the fact that universes like ours are maximally capable of producing black holes (this latter point, however, has been challenged by Joe Silk, one of Smolin’s critics; another objection that has been raised, by John Polkinghorne, is the possibility that under certain scenarios short-lived universes could actually reproduce more efficiently than long-lived ones like ours).

I mentioned the falsifiability criterion: my understanding is that Smolin predicted — among other things — that we would not discover neutron stars more than 1.6 times more massive than the Sun. Apparently, one such object has been discovered, though it is only slightly off the predicted range (2 solar masses). There seem to be ways of reconciling this discrepancy with the theory (which, of course, brings up the whole philosophical discussion of what one means by falsification), but, again, this is physics and I’m not going to dwell on it.

The rest of Smolin’s paper deals with a detailed reply to an objection to his theory advanced by Alexander Vilenkin. Both the objection and the rebuttal are very technical, but I need to quote a couple of passages because they are pertinent to my analysis of whether CNS truly is analogous to BNS: “If Vilenkin’s argument works against cosmological natural selection, it works also against many predictions of Darwinian biology.” Here it appears that Smolin is relying on the cachet of the Darwinian theory to support his cosmological theory by drawing once again a close analogy between the two, but he then immediately admits that “Vilenkin’s argument could be used to discredit many predictions of science which up till now were thought to be established.” Not just predictions of Darwinian science, but of science in general. So one does not actually need to invoke the special case of biology in this context.

Smolin, toward the end of his paper, says: “My main claim is that it is the time bound, evolutionary picture, which is best suited to the world as we observe it, because what we observe is in fact a universe evolving continually in time. By contrast, the main theoretical object one works with in eternal inflation and the anthropic principle is not a representation of what we observe, it is an entirely invented eternal and static multiverse.” Again, I like this idea of looking for a physical picture of the evolving cosmos, rather than for timeless and universal answers — but “evolving” here could simply mean changing over time, it doesn’t have to do with the specific technical meaning that the word has within the context of evolutionary biology.

And now to the crux of the issue. I see at least two significant disanalogies between CNS and BNS. The first is that at best we should be talking about “sorting” rather than “selection.” The second one is the twin issue of heritability and mutation connecting mother to baby universes in a “lineage” analogous to biological ones.

Perhaps the best well known modern definition of BNS is from Harvard geneticist Richard Lewontin.  Here he is, verbatim, in this case from a review he wrote of the infamous What Darwin Got Wrong by J. Fodor and M. Piattelli-Palmarini (a book I didn’t like either, to put it mildly). According to Lewontin we can talk of natural selection if the following conditions are satisfied:

(1) The principle of variation: among individuals in a population there is variation in form, physiology, and behavior.
(2) The principle of heredity: offspring resemble their parents more than they resemble unrelated individuals.
(3) The principle of differential reproduction: in a given environment, some forms are more likely to survive and produce more offspring than other forms.

Let’s start with number (3): notice that differential reproduction — according to Darwinian theory — has to take place “in a given environment,” because there has to be competition for resources in order to have selection for the most fit individuals. It is not at all clear what this common environment would be in the case of the universes making up a multiverse, since they are not supposed to be in contact with one another. Now, in biology there is a (hypothetical, highly debated!) situation where differential birth and death do lead to non-random evolutionary outcomes, but it is referred to as “sorting” and has been famously proposed by Steven Stanley in the context of evolution between species (so-called “species selection,” which is not selection, but, again, sorting). Species may give rise to sorting because they may have species-level characteristics (such as dispersal frequencies, or geographical distribution) that may increase the frequency of speciation or decrease the likelihood of extinction (and, again, speciation and extinction are things that happen to species, not individual organisms). Insofar, then, as universes have a lower likelihood of dying out (longevity due to stability) and/or a higher likelihood or originating baby universes (because of their black holes), there can be sorting. But it would not be natural selection because the universes aren’t competing for resources in a common environment.

(1) and (2) on Lewontin’s list require a mechanism for producing variation (in biology, mutation and recombination) and a mechanism of heredity. Smolin does claim — as we have seen — that baby universes inherit their “mother’s” parameter settings, and that moreover this inheritance is subject to random variation, but I couldn’t find anywhere in his paper a reference as to how exactly cosmological mutation and inheritance are supposed to work.

So I asked Smolin whether I got his point correctly, and he answered: “The answer is yes, I postulate that the analogue of the genes are the 20+ dimensionless parameters of the standard model of particle physics and that when new universes are created in black holes they inherit these values from the parent universe subject to small random changes.” The key word here is “postulate.” I have asked for further clarification as to the basis of this postulation, but have not gotten a response yet (if I do, I will post it, or perhaps I can talk Smolin into writing a commentary for Rationally Speaking on this subject).

I also inquired with Sean Carroll, who is not a supporter of CNS, but is interested in and knowledgeable on these issues. His response: “I’m pretty sure it does include a mechanism for ‘heritability,’ but I think the mechanism is just asserted — it’s not something that comes out of any deeper understanding. We imagine that the laws of physics are specified by some list of parameters, like Newton’s constant and the mass of the electron. Smolin suggests that (1) black holes produce new universes, and (2) the new universes ‘inherit’ the parameters of the universe they came from, with some small random variation. As far as I know there’s no reason for this to happen — Smolin is just wondering, what if it does? — and [we know of] no mechanism analogous to DNA to actually carry the information.”

So, it seems to me that cosmological natural selection is at best analogous to species sorting in biology (a mechanism that has been hypothesized but is controversial), not to the much better established principle of natural selection. Moreover, to say the least, even this is very tentative, unless and until someone discovers an actual mechanism of cosmological inheritance and variation. It is true, of course, that Darwin himself didn’t know anything about mechanisms (a point Smolin makes at the beginning of his paper), but it is also the case that Darwin could demonstrate both selection and inheritance readily, while nothing like that has been established in the case of CNS.

If I am correct, then, CNS is an interesting theory in cosmology, but the analogy with BNS may be incorrect and is at least problematic. Which reinforces my specific point about the need to be careful before declaring the universalization of Darwinism, and my more general point that scientific theories — qua human constructs about how the world works — should be thought of as having proper domains of application, domains that need to be respected to avoid conceptual confusion.