Sorry for the gap, RL intervened. This is going to be the hardest one of these to write. And if you think tl;dr, fair enough. But it’s a crucial aspect to discussions of EP and SB.
Although I think that Darwin’s greatest idea was common descent as an explanation for the relationships of organisms, I am very much in the minority here. Most people take it as read that Darwin’s best idea was natural selection. However, ideas very like natural selection (NS) had been around for a long time, unsurprisingly since they are employed in animal husbandry (the artificial/natural distinction is, well, artificial). What was novel about Darwin’s and Wallace’s use of NS was that it caused change in species, or so they thought. These days the consensus appears to be that species originate independently of natural selection processes, and selection changes the ecological adaptive features of the species rather than the isolation of species from other species.
That is by the way, however. Almost everyone in the evolutionary psychology movement and prior sociobiologies, at least after the turn of the 20th century, thinks that natural selection accounts for human behaviours. The issue is again a haggling over the price: how much and how often, and of what? So a major issue in SB4.0 is the role of adaptive explanations: what should we propose, how do we test them, and what is the default explanation: chance or selection? That is what this post is about.
EP and SB have typically presented explanations of human behaviour based upon the adaptation of those behaviours. Everything from mate choice to rape to religious belief to cognitive deficits are explained in this fashion. I do not propose to examine any of them here. What, as a philosopher, I find interesting are the assumptions underlying them, and how they might fail. Scientific research has a number of mathematical and experimental techniques to determine if something is the outcome of selection: selective sweep analysis, experimental work on cases, and phylogenetic comparisons. Each seems to have problems.
To determine if something is the result of a selection process, one has to have a good idea of what it is that could be the target of selection. The Standard Answer is that genes are, but often as not we do not yet have much of an idea what genes are responsible for what traits, and what the norm of reaction (the way the genetic trait is expressed) might be. This has always been a problem for evolutionary genetics, and indeed for a long time, a “gene” was simply a heritable trait (or rather, what underlay a heritable trait). As once noted by Lewontin, phenotypes are what Mendelian genetics studied.
So EP/SB has to indirectly approach selectionist explanations. Often they do not, but the better studies have to face the same methodological issues that every other evolutionary explanation does:
- We do not know the genes for most traits (and few traits are single gene traits, even for metabolic diseases),
- We often do not know much about the environment in which selection is thought to have occurred, and when we do it is global rather than local (e.g., a snail is not affected by the climate of the whole world, but by the microclimate of its hillside or valley),
- Information about the past is often fragmentary, leading to easy and cheap narratives that happen to suit the researchers,
- Information about the present is also often fragmentary and partial, and
- If a trend is seen in the data, it is often unclear when this is due to drunkard’s walks (chance) which can deliver directionality over moderately long periods, and when it is due to selection (which can be chaotic if the environment is).
Moreover, the standard EP approach relies upon an atomisation of traits that is more based on expectation than upon observational data. For example, and the key issue for EP for humans, the “massive modularity hypothesis” requires that the cognitive and psychological traits are parcelled into discrete and relatively independent modules. There is a parlous amount of evidence for this outside sensory modalities. Yes, vision (and possibly hearing and the other senses) is quite modular. It is also very old, going back perhaps to the Cambrian, so given that things with eyes need to see well, we might expect the visual module to be independent of the other neurological traits.
But modularity is not required by evolutionary theory; in fact it seems more to be a matter of convenience for researchers than anything else. A complex multifactorial system can be modified simultaneously in many ways by selection, so long as the factors (i.e., the underlying genetics) are heritable, because what NS requires is that things be heritable for them to be adaptively changed by NS. Think of an organism as a system that develops like a sprung mess of rigid poles connected by elastic bands. You can move several at once and the whole system will reshape (this is called tensegrity by Buckminster Fuller, as a punishment for thinking about these things). The rigid poles are the heritable genes, and the rubber bands are the norms of reaction.
[A tensegrity example, from here.]
So we might expect that you can modulate behaviours (or at any rate d-behaviours) by adjusting a few genes at once.The requirement that d-behaviours (or for that matter expressed behaviours) should be isolated and independent of other traits is not required. Gould and Lewontin referred to this as “inappropriate atomisation”. The “adaptationist programme”, they wrote (Gould and Lewontin 1979: 585), starts with this step:
An organism is atomized into ‘traits’ and these traits are explained as structures optimally designed by natural selection for their functions. For lack of space, we must omit an extended discussion of the vital issue: ‘what is a trait?’ Some evolutionists may regard this as a trivial, or merely a semantic problem. It is not. Organisms are integrated entities, not collections of discrete objects. Evolutionists have often been led astray by inappropriate atomization, as D’Arcy Thompson … loved to point out.
The emphasis upon isolating single genes or alleles is a case in point. No gene does what it does in a vacuum (in fact, put DNA into a vacuum, and it will simply sublimate and denature). It needs not only other genes but a developmental organism in an environment it affects simply by developing. I can’t stress this truism enough. What a gene does depends on the system it is a part of.
Modularity, whether of genes or organs or neurological systems, is only ever a conceptual abstraction. Cognitive modularity, however, was based upon the presumption that to be the subject of selection, a cognitive process had to be isolated from other processes in order to be optimised, to be encapsulated: modules were shallow so they could produce outputs without much delay based on the types of inputs they were optimised to receive, domain specific to they only received one type of stimulus or input, inaccessible so they were not affected by content or dispositions elsewhere in the brain, and so on. Later work showed that even those exemplars of modularity, such as are tested by the Wason Test, of reasoning are able to be moderated and even interrupted by contents of the cognitive system
So, enough about what fails or is problematic about EP and SB. What about what works?
Some may take issue with (PZ has) my claim that only selection can generate complex d-behaviours. A brief word is due. We know that selection (as an instance of general sorting processes) can generate complexity of behaviours. We know that a lineage or population that has some suite of behaviours can randomly attain a more complex behaviour. I am not suggesting that random variations within a population of already high complexity cannot produce such behaviours. But over the longer term, what randomness giveth, randomness will taketh away. For a trait to persist in a population (note, in a population, not a lineage of species over large periods) and to go to fixation if the population is of any size it is more often than not, much more often, due to selective pressures on those behaviours than to chance. And where using chance as a “null hypothesis” generates few if any hypotheses of what did happen that are testable, selective explanations do generate many, often very specific, hypotheses that are testable.
There is a real issue (see here, here and here; the work that convinced me of this is Fidler’s) with null hypotheses anyway. A “null” hypothesis is, simply put, the hypothesis that the researcher or the researcher’s community defaults to in the absence of anything else. In short, the null hypothesis is what the community of researchers is comfortable with, and since this is not really constrained by data (there are an indefinitely large number of potential nulls) testing between a null and positive hypothesis is at best a very localised test.
But leave this to one side: if chance explanations of a d-behaviour explain it arising, it can equally explain it disappearing. To explain the persistance of a complex d-behaviour, selection is pretty well it. As was long ago noted the origins of the raw material of selection are down to accidental changes, but not their retention.
But that is not the whole story. Once you have a reproducing populations of organisms of overall high fitness, individual d-behaviours may wander about stochastically. This can be true of a single variable, so long as the overall or general fitness of the population mean is high enough. So it comes down to being able to establish whether the d-behaviour is likely to lower the fitness enough that selection pressures will overcome the randomness of mutation and recombination. And there is no principled answer to this – it relies upon the facts of the case; the boundary conditions and the number and relation of the variables. Sergey Gavrilets (1997, 2004, see Wilkins 2007) has noted that in a correlated or smooth fitness landscape (such as we might expect nearly all species to exist in when in their ordinary environments) there are what he calls “holey landscapes” (see figure), which are components of fitness space that are of roughly equal fitness, and which are connected in various ways to higher and lower fitness components. In short, in a sufficiently high dimensional landscape, a population may drift around at about the same fitness (in other words, its d-behaviours can shift stochastically) so long as the fitness of that population remains about the same. So this implies that a d-behaviour can be both the result of selection relative to its absence, but also the result of chance relative to equally fit alternatives. The chance/selection dichotomy is misplaced. Both can be true simultaneously.
So to take on the adaptationist program (or “programme” as Gould and Lewontin spelled it) is not to make a mistake. It is to make an epistemic bet. Cheap and simple adaptive hypotheses must be treated with derision, of course, as they must in every aspect of science – this is not a critique of EP as such, but of bad science. But good ones must be treated as working hypotheses. There’s a big literature on testing and choosing between models. I shan’t bore you with that here (I find the Akaike Information Criterion approach perplexing, for example, which probably says more about me than it). The point is that EP/SB is not required not to be
non-adaptationist as a null hypothesis.
In the light of all this, I suggest that so long as we test our adaptive scenarios, and do so realistically on the basis of phylogenetic bracketing (to ensure we are actually dealing with a d-behaviour) and on the best data (especially that of neurobiology and cognitive psychology), EP/SB (or SB4.0 at any rate) has a right to be adaptationist, and that alternative approaches must be justified. Adaptation is the “null hypothesis” in the case of complex d-behaviours, subject to the qualifications I make above.
As once said, there’s the bit where you say it and the bit where you take it back. Adaptationism does not explain, I believe, the persistence of a trait across phylogenetic bifurcations – that is, in the case of evolutionary lineages. Or rather, it can, but only in the sense of it being “evolutionarily conserved”. For example, the structure of the genetic code is highly conserved. Although there are some 15 variants across the whole of life, they vary only in details. The having of the “universal” code is highly fit, and variants tend to be eliminated. Many developmental, phenotypic, physiological and cytological structures are highly conserved. Why? The answer has to be something like: if you deviate from this state, your fitness is lowered, just because it is the default or modal state. So the having of, say, a particular social disposition among primates can be maintained because the having of that disposition increases your fitness – deviants get fewer mating opportunities. Creatures inveterately wrong at following the norm have a pathetic, but praiseworthy, tendency to die before reproducing their kind…
Which leads us to a final point about adaptation: the what it adapts to. Good Old Fashioned Adaptationism (GOFA*) always presumed that what the fitness assigning process was had to do with the ecological context of the organisms. But equally, or in the case of highly adapted creatures like primates, predominantly, the fitness assigning process is social. You are fitter because your d-behaviour fits a potential mate or social conspecific. Here I think the emphasis has been more productive. Others will disagree, but as the Buss Lab’s defence shows (Confer et al 2010), there are some useful results in so doing.
* Not to be confused with GOFAI, which is a position in artificial intelligence philosophy.
Confer, Jaime C., Judith A. Easton, Diana S. Fleischman, Cari D. Goetz, David M. G. Lewis, Carin Perilloux, and David M. Buss. 2010. Evolutionary psychology: Controversies, questions, prospects, and limitations. American Psychologist 65 (2):110-126.
Gavrilets, Sergey. 1997. Evolution and speciation on holey adaptive landscapes. Trends in Ecology & Evolution 12 (8):307-312.
———. 2003. Perspective: models of speciation: what have we learned in 40 years? Evolution Int J Org Evolution 57 (10):2197-2215.
———. 2004. Fitness landscapes and the origin of species, Monographs in population biology; v. 41. Princeton, N.J.; Oxford, England: Princeton University Press.
Gould, Stephen Jay, and Richard C. Lewontin. 1979. The spandrels of San Marco and the Panglossian paradigm: a critique of the adaptationist programme. Proc R Soc Lond B 205:581–598.
Wilkins, John S. 2007. The dimensions, modes and definitions of species and speciation. Biology and Philosophy 22 (2):247 – 266.