Claus Wedekind’s 1995 “sweaty T-shirt” study is one of evolutionary biology’s most quoted findings: women asked to smell T-shirts worn by different men tended to prefer the scent of men whose MHC genes, a large immune-related gene family, were most dissimilar from their own. The popular takeaway usually gets simplified to “we’re attracted to genetically different mates because it makes healthier babies.” That’s close, but not quite what the actual biology, or its computational cousin, says. Genetic algorithms, the optimization technique built to mimic natural selection in software, have their own well-documented struggle with genetic diversity — and the field learned decades ago that the naive version of “more diversity is always better” is wrong there too. The precise way both fields corrected that naive intuition is where this comparison actually gets interesting.
Scientific Foundation
MHC genes are the most polymorphic gene family known in vertebrates, and for good reason: each specific variant confers resistance to a different, specific set of pathogens, so an individual carrying a wider range of MHC variants can recognize and fight off a broader range of infections. Wedekind’s study, and follow-up work by other researchers, found that human body odor carries detectable information about MHC genotype, and that odor preferences, at least in women not using hormonal contraceptives, correlated with MHC dissimilarity from the smeller’s own genes — a plausible mechanism for steering mate choice toward offspring with broader immune coverage. But the more precise, more recent framing of this research isn’t “maximum dissimilarity is best.” A detailed review of MHC odor signaling puts it directly: an individual’s ideal MHC collection is optimally diverse to resist a range of infectious diseases, but not so diverse that it triggers autoimmune disease, since some MHC combinations are known to raise the risk of the immune system attacking the body’s own tissue. Mate choice, in this more accurate account, aims for optimally complementary MHC combinations, not simply the most different ones available. The human evidence is also considerably messier than the famous headline suggests: preferences reverse in women taking oral contraceptives, and follow-up research found that only certain trained evaluators, including professional perfumers, could reliably describe MHC-correlated differences in body odor at all — researchers studying this mechanism explicitly note that the situation in humans is more complex than in the cleaner experimental animal studies that first suggested the effect.
Cross-Domain Connection
Genetic algorithms ran into a structurally identical problem decades ago, in a completely unrelated context. A GA that converges too quickly toward genetically similar candidate solutions gets trapped in premature convergence, settling on a mediocre local optimum instead of continuing to explore the broader solution space for something better. The field’s answer, developed starting with De Jong’s crowding technique in the 1970s and refined since through niching and fitness-sharing methods, works by comparing new candidate solutions against similar existing ones and managing which survives, specifically to keep the population’s diversity from collapsing too early. But GA researchers learned, just as MHC biology did, that the naive fix overshoots: as one detailed analysis of crowding methods states plainly, too much diversity can dramatically reduce GA efficiency, wasting computational effort exploring unpromising regions of the search space instead of refining good solutions. The field now frames this explicitly as a trade-off between exploitation of the best current solutions and exploration of alternative regions — an intermediate optimum, targeted deliberately, not a maximum to chase.
What Remains Undemonstrated
The conceptual match here is genuinely precise, not just a loose family resemblance: both fields independently arrived at “avoid both too little and too much diversity, and manage the trade-off explicitly” as the correct principle, after presumably starting from a simpler, wrong intuition that more diversity is categorically better. Where the comparison needs honest tempering is in how reliably each system actually implements that principle. A genetic algorithm’s crowding or niching operator computes genotype or phenotype distance directly and precisely, every generation, using complete information the program has full access to. MHC-based mate choice runs through something far less clean: an indirect, evolved sensory channel, body odor shaped by which peptides an individual’s MHC molecules happen to bind and present, that carries genetic information imperfectly and inconsistently. The signal is often too subtle for most people to consciously register in a way they can articulate, it reverses in a large, common subset of the population using a specific widely-taken medication, and researchers studying it are explicit that the clean, mouse-level story doesn’t fully hold up once you look carefully at human data. No genetic algorithm’s diversity operator has an equivalent vulnerability — nothing analogous to a medication a large fraction of “individuals” might be taking that would systematically scramble its distance calculations.
Why It Matters
That gap is itself a useful, precise lesson rather than a disappointing footnote. It confirms something real: “target an intermediate optimum of diversity, not a maximum” is a genuine, transferable principle that shows up wherever a population-based selection process, biological or computational, needs to balance coverage against coherence. But it’s also a caution against assuming biological systems implement general principles with anything like an algorithm’s precision. Evolution didn’t get to write a clean distance function and compute it exactly each generation — it had to route the same underlying logic through smell receptors, peptide chemistry, and hormonal state, a much noisier engineering substrate than anything a genetic algorithm has to contend with, and the resulting behavior is correspondingly messier, more conditional, and more easily disrupted than the tidy optimization story usually gets credit for.
Human Dimension
There’s something worth sitting with in the fact that your immune system and a piece of optimization software are, in a precise sense, solving the same equation — how much genetic difference is actually good for you, given that both too little and too much carry real costs. But only one of them gets to compute the answer exactly. The other has to guess at it through the tangled, imperfect, occasionally contraceptive-scrambled machinery of a body doing its best with a sense of smell nobody designed for precision. It’s a reminder that “nature already solved this” is often true at the level of the underlying math, and considerably less true at the level of how faithfully biology actually manages to execute it.
Sources:
1. Wikipedia — “Claus Wedekind” — https://en.wikipedia.org/wiki/Claus_Wedekind
2. PMC (National Institutes of Health) — Milinski, M., “A Review of Suggested Mechanisms of MHC Odor Signaling” — https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9405088/
3. SAGE Journals — Wedekind, C. et al., “The Major Histocompatibility Complex and Perfumers’ Descriptions of Human Body Odors” — https://journals.sagepub.com/doi/10.1177/147470490700500206
4. PMC (National Institutes of Health) — “Major histocompatibility complex peptide ligands as olfactory cues in human body odour assessment” — https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3574394/
5. ResearchGate — “Sweaty T-Shirt Experiment” — https://www.researchgate.net/publication/370173972_Sweaty_T-Shirt_Experiment
6. Evolutionary Computation (ACM) — “The crowding approach to niching in genetic algorithms” — https://dl.acm.org/doi/10.1162/evco.2008.16.3.315
7. ScienceDirect — “Adaptive generalized crowding for genetic algorithms” — https://www.sciencedirect.com/science/article/abs/pii/S0020025513006300
8. arXiv — “Competition and Attraction Improve Model Fusion” — https://arxiv.org/pdf/2508.16204
9. arXiv — “On the Runtime Analysis of the Clearing Diversity-Preserving Mechanism” — https://arxiv.org/pdf/1803.09715
Idea originated at artificialideas.org. Article researched and written by Claude Sonnet 5. Published at artificialideas.org.
Artificial Ideas 