To summarize, it now appears that over the last 25 years, anywhere from several hundred to over a thousand scientific papers were published, in reputable, peer-reviewed journals, based around presumed “genes for depression” (5-HTTLPR being perhaps the best-known) that are now thought to be completely bogus. The dramatic and abrupt discrediting of these genetic linkages is mostly due to an immense, 600,000-subject study by Border et al., released just last March, which investigated a collection of the most famous “depression genes” and came away finding no statistical support for any of them.
Alexander’s discussion of this, the scientific equivalent of a 500-car-pileup, is especially punchy and concentrated; he is, understandably, much shaken that such a huge body of seemingly reputable confirmatory research on 5-HTTLPR could have turned out, apparently, to be pure phantasm. He throws the absurdity of the whole situation—where perhaps hundreds of academic research groups all managed to convince each other for decades of the rock-solid validity of a host of nonexistent effects—into sharp relief, using a barrage of vivid analogies like the following:
“…This isn’t just an explorer coming back from the Orient and claiming there are unicorns there. It’s the explorer describing the life cycle of unicorns, what unicorns eat, all the different subspecies of unicorn, which cuts of unicorn meat are tastiest, and a blow-by-blow account of a wrestling match between unicorns and Bigfoot.”
Yet even this evocative comparison doesn’t quite capture the bizarreness of the “depression genes” situation, for what we see here is less like one explorer, than an entire corps of hundreds of explorers, all going to the Orient and all coming back claiming they saw the same collection of extraordinary, unicorn-themed hijinks.
One obvious possibility, which Alexander gives perhaps too little credence, is that the allegedly dispositive 600,000-subject study, despite being larger, broader, and more modern than all the previous ones put together, may have nonetheless missed something in dismissing the other results. Certainly it seems easier to believe that one study, however large, might be inaccurate, than that hundreds of smaller but independent ones might be. That said, no flaw is yet apparent in the new study, and as Alexander points out, it is not the first work to cast serious doubt on 5-HTTLPR.
But what really haunts me while reading about this latest scientific mess, is a wilder, more Sheldrakean possibility: could it be that 5-HTTLPR and the other gene variants actually were associated with depression for a decade or two—during which these hundreds of studies simply reflected reality—but then simply ceased to be associated with depression, which was then also correctly reported by the new study? If one pauses to consider, it’s not obvious that this possibility is much crazier than the notion that 1,000 studies were carried out, written up, and accepted, year after year, about a completely nonexistent effect.
These are desperate epistemic times, indeed.
* * *
Such snowballing interpretations and re-interpretations, often involving hypotheses of increasingly surreal strangeness, are suggestive of a far more sinister epistemological breakdown at the heart of at least some branches of science.
The issues are by no means limited to 5-HTTLPR. For example, in another of Alexander’s posts, where he reflects on the growing difficulties in establishing reliable scientific truth through research. Strikingly, he recounts findings indicating that parapsychology—the study of such problematic, nay “unscientific” phenomena as clairvoyance, telepathy, telekinesis, etc.—actually now manages to justify its results at a level of rigor equivalent to that required of “normal” scientific publications, and at about the same rate at normal scientific fields do. He sums the situation up this way:
“…with enough energy focused on a subject, you can always produce ‘experimental evidence’ for it that meets the usual scientific standards.”
This is a remarkable statement. Of course by it Alexander means to drive home something like, “real science’s evidentiary standards are in trouble, because even the parapsychologists, whom we know produce only unscientific nonsense, are now about equally able to meet these same standards”.
Yet as with 5-HTTLPR, we find that Alexander has again inadvertently set us face-to-face with still more Sheldrakean alternate interpretations. First and most obvious of these is that parapsychology’s success in meeting scientific standards of knowledge may actually imply that it is not wholly unscientific nonsense after all, and hence that Alexander’s working epistemic assumptions about it are no more solid than the mainstream wisdom about 5-HTTLPR apparently was. But secondly, and even more strikingly, Alexander’s remark aimed at dismissing parapsychology seems in itself to concede a kind of parapsychological effect: to wit, that “focused energy”, in the form of mass intentions and expectations, can in some way directly influence, even reverse, scientific outcomes, thus fanning the flames of the replication crisis.
* * *
At any rate, even if we do not dare propound such dangerous notions as a fundamental ambiguity in the distinction between parapsychology and “real” psychology, or a direct effect of mass mentality on the aetiology of psychological conditions, we may at least point out certain other, no less remarkable aspects of the current situation.
In particular, at the core of the 5-HTTLPR disaster—standing over its spent body, one might even say—is the now-ascendant omnigenic hypothesis, which asserts, based on a growing number of big-data GWAS studies with huge subject sizes much like Border et al., that almost none of the variability in most kinds of “complex” traits can be explained by single gene variants, or even by small clusters of genes or metabolic pathways. Instead, it appears that human characteristics of such obvious importance as height, intelligence, temperament, mental illness, and even skin color all have an important genetic component, but that this component can only be reckoned as a summation of extremely tiny effects exerted by thousands or tens of thousands of genetic variants.
So then the failure of the “depression genes” hypothesis, from an omnigenic point of view, is “simply to be expected”. In his 5-HTTLPR post, Alexander himself indulges in a bout of Whiggish, retroactive rationalization on this score, approvingly pointing to 5-HTTLPR’s downfall as reaffirming the strength of scientific progress. In science, he tells us, what was once real to us we later discover to be “really” the silly ignorance of our past selves; in the present case, we now see clearly that the whole idea of “depression genes” could never really have made sense, because it would conflict with omnigenics.
Yet why such self-congratulation or epic retconning should increase our confidence in science and truth in the context of escalating failures of knowledge is, to put it mildly, hard to to understand. What makes the modern revision more solid or stable than the old one we now mock, and what makes our present selves fundamentally different from the past selves that were so hilariously taken in, when the whole pattern really suggests our present views, too, may soon need similar retconning?
Much like his oddly parapsychological argument against parapsychology, this appears to be another case of Alexander inadvertently strengthening a point he means to persuade against (and in this we admittedly here take him as somewhat emblematic of the scientific mindset). One is reminded rather much of McGilchrist’s view of the left brain run amok, continually spinning stories, howsoever fabulous, to reassure us of the world’s tractability and our own powers of control. I would submit that 5-HTTLPR was one such story; the assurance that “failures of science actually are proof of its strength” is another. This is not to say that some stories may not have more objective validity than others, but that a situation like that now increasingly manifest in science—a rapid succession of stories without a clear way of judging whether there has been an actual improvement in explanatory or causal understanding from one to the next—is a sign of breakdown.
In light of this, the most disquieting aspect of the omnigenic model is not that it cannot be shown to explain more trait variance than models with fewer gene variants; it is that the gene variants found to account for most traits have, usually, no evident functional relation to each other. They typically do not segregate strongly by metabolic pathway, or chromosomal, physiological or cellular location, for instance. Often, the variants are not even in regions that code for protein. In this sense, the omnigenic model might best be characterized as the absence of any model, a limit on our understanding as much as it is an advance.
Consider an omnigenic accounting of depression risk. Here, the likelihood of an individual developing depression will be governed by some kind of weighted linear combination of tens of thousands of gene variants. Yet what makes this combination the “depression combination” rather than some other, equally random-seeming combination out of the trillions upon trillions possible in combinatorial space? In short, what is so special about it? In the omnigenic world, these questions have no answer—or else “answering” them will require models containing so many variables and assumptions that they give at most a hugely contingent and relatively weak statistical account, not a mechanistic one.
Yet assuming omnigenics is nonetheless “the case”, we are brought to the following fascinating idea: features that are salient and even indispensably important from the perspective of the mental, such as depression and intelligence, can yet have, from the perspective of the physical (or genetic), absolutely no discernable salience whatsoever.
Omnigenics makes much of the idea that these thousands of variants, once properly weighted and accounted for, “explain” the “missing heritability” of complex traits. Yet this depends very much what one means by “explain”. While the variants may collectively lead to better statistical estimates than those made with any one gene or few genes, they give no intuition about what is going on, and no suggestion of how to proceed towards such intuition. We must simply take the numbers and do what we can with them.
Omnigenics therefore represents a kind of abdication of causation—since the interactions between the tens of thousands of tiny genetic effects, although they may lead to a profoundly significant and totally undeniable mental consequence (depression), cannot be traced to any pathway or even to any discrete sequence of events. Complexity has not only swallowed up explanation; it has digested it.
This places us in the throes of a strange new sort of dualism. The omnigenic model suggests that the world as seen from the mental or intentional realm, as well as most of the other complex traits of life that we single out so intuitively within the realm of our perceptions, are in fact causally near-complete strangers as far as the realm of material is concerned. For, as well as we can see, the mappings between the two, while capable of being hacked out by brute statistics using huge sample sizes, are curiously arbitrary, in the sense of being governed by no determinate, comprehensible, or known law or causal factor. Analogously to the hyper-complex representations inside a trained artificial neural network, they are conceptually random; they give us no conceptual sense of how a linear combination of 20,000 genes strongly influences the development of the given mental trait, but only stipulate that it does. And if these mappings really are conceptually random/arbitrary, with no comprehensible sense of causation in which to ground them, then nothing much stands to argue against their being much more freely and rapidly changeable than evolutionary theory or mutation rates might suggest. Through omnigenics, Sheldrake (or possibly a weird dualist-postmodernist kind of irrealism) may have the last laugh.