The Luthier: Spandrels, Social Eigenvalues, and the Instrument Evolution Built

The paradox, stated fairly

Natural selection has no foresight, no preferences, and no interest in beauty. It is differential reproduction, iterated. And yet the organism it produced writes the Divine Comedy, carves the Pietà, sets Psalm 127 for muted strings, and occasionally walks into a burning building for a stranger. That is a real puzzle and it deserves a real answer rather than a reverent one.

The honest form of the puzzle is narrower than the rhetorical form. Nothing needs explaining about a species that likes food, sex, safety, and status. What needs explaining is (a) capacities that appear to cost fitness and (b) capacities that appear to exceed any plausible ancestral task. Art and self-sacrifice are the two standard cases. Biology has answers for both, and the answers are less mystical and more interesting than the framing usually allows.

Spandrel, said precisely

The word comes from Gould and Lewontin's 1979 paper, and it is worth using it as they defined it: a spandrel is a structural byproduct — the tapering triangular space that necessarily arises when you put arches under a dome, later decorated, but not built in order to be decorated. Their target was the habit of assuming that every trait has an adaptive story. The word is not a synonym for happy accident; it is a demand that byproduct be considered as a competing hypothesis before adaptation is asserted.

Applied here, the claim is that language, hierarchical planning, theory of mind, and pattern completion were selected for reasons that are easy to state — tracking, coalition management, tool sequences, deception and its detection — and that music, mathematics, and narrative fall out of that machinery without being separately selected for. Pinker's provocation that music is auditory cheesecake is the strong version of this. It has been contested rather than settled: adaptationist accounts of music through coalition signalling, infant-directed vocalisation, or credible signalling of group cohesion are live, and the cross-cultural regularities catalogued by Mehr, Singh and colleagues in their 2019 Science survey are real data that any byproduct account has to accommodate.

So the volume can assert the byproduct hypothesis as a hypothesis and should not assert it as the finding. What it can assert flatly is the weaker and still substantial point: a general-purpose capacity, once built, is usable outside the domain that paid for it. Reading is the cleanest example — no ancestral selection for it, a few thousand years old, and it recruits object-recognition cortex wholesale. Dehaene's neuronal recycling account gives the mechanism, and it is the mechanism the chapter actually needs.

Where the overflow image holds, and where it breaks

The generative figure of this volume — a played layer driving a sympathetic layer beneath it — was corrected in Chapter 38 in a way that matters here. A passive linear resonator adds no frequency that was not already in the drive. It selects and sustains; it does not invent. Genius, read through that instrument, is therefore not the brain producing more than it was given. It is the brain answering, with enormous gain and long ring-down, a narrow set of structures it happens to be tuned to.

That is a smaller claim than transcendent excess and it survives contact with the evidence better. Expertise research says the same thing from the other end: what looks like overflow is usually a highly trained selection filter. The chess master's board perception, the mathematician's sense of which lemma will close, the musician's hearing of a modulation two bars before it lands — all are chunking, all are learned, all are domain-bound. Nobody is generally brilliant. People are catastrophically well tuned to one thing.

The phrase we are vibrating at a frequency higher than our survival requires should be retired as physics and kept, if at all, as prose. There is no frequency here. There is a capacity margin, and margins are ordinary in engineered and evolved systems alike: a bone strong enough for the worst load it will ever meet is stronger than it needs to be on almost every day of its life.

Compassion without the eigenvalues

The claim that altruism stabilises groups is not speculative, and it does not need borrowed spectral vocabulary. Hamilton's 1964 inclusive-fitness result explains care directed at kin. Trivers' 1971 reciprocal altruism explains care returned over time. Axelrod and Hamilton's 1981 tournament work showed that conditional cooperation is robust against defection in repeated interaction. Fehr and Gächter's 2002 experiments showed that human cooperation is sustained by costly punishment of free riders, which is to say by a mechanism people find unpleasant and use anyway. Multilevel selection remains contested in its strong form; the weaker statement — that cooperative groups outcompete fractious ones under some conditions — is not.

None of that requires the word eigenvalue, and this chapter declines it. Chapter 34's eigenvalues were real: solutions of a characteristic equation for a specified operator, with units, computable, and constrained by theorems. Calling empathy a social eigenvalue names no operator, no matrix, and no spectrum, and therefore cannot be wrong — which is exactly the failure mode this volume was built to avoid. The figure is allowed only in the form the mathematics actually supports: cooperative institutions act as negative feedback on individual defection, and negative feedback is what pulls a dynamical system back toward a basin. That sentence is checkable. The eigenvalue sentence is decoration.

The stronger correction is to the moral of the story. Empathy is not the ultimate survival strategy. In-group compassion and out-group hostility are, empirically, closely coupled, and the same machinery that makes a stranger's pain legible makes a rival's pain satisfying. Any account of compassion that omits that coupling has not explained human beings; it has flattered them. The interesting fact is that the aperture of the in-group has widened historically — kin, band, city, nation, species — and that widening is cultural and reversible, not a biological ratchet.

Two claims this chapter drops

First: quantum-coherent mirror neurons. There is no evidence that mirror neurons operate through sustained quantum coherence, and the standard decoherence estimates for warm, wet, ion-dense neural tissue run many orders of magnitude below the milliseconds that cognition requires. The volume has kept a live file on warm quantum biology — avian magnetoreception, photosynthetic transport — and those cases are real and narrow. Extending them to the felt experience of another person's suffering is not supported by anything, and the sentence reads as physics while carrying none. It goes.

Second: mirror neurons as the seat of empathy. Mirror neurons are a genuine finding in macaque premotor cortex from the Parma group in the 1990s, and human homologous activity in action-observation networks is well replicated. What is not established is the interpretive layer built on top: Hickok's critique documents that the action-understanding-by-mirroring account fails on several of its own predictions, and human single-neuron evidence remains thin. Empathy in humans is better described by overlapping affective networks — anterior insula, anterior cingulate — active both when a person is in pain and when they watch someone in pain, as in Singer's 2004 work. That is a real, measured overlap, and it is enough for the point being made.

With those gone, the etymological observation stands on its own and is worth keeping: compassion means to suffer with, and the neuroscience says something close to that literally — shared representational territory, not shared substance. The listener resonating with the Gloria Patri and the bystander wincing at an injury are running the same trick: a model of a state, built from the inside, driven from the outside.

The luthier, and what the metaphor is allowed to claim

Evolution as master luthier is the chapter's best image and its most dangerous one, because a luthier intends. Selection carves without intending, which is why the instrument has an appendix, a blind spot, a birth canal in tension with a skull, and a spine that was not designed for standing. A luthier who worked like this would be unemployable. The image is admissible only as the observation that a brutal process can produce a precisely tuned object, which is true and is the whole content of the Blind Watchmaker argument.

What survives, then, is this. The capacities were paid for by ordinary selection. Their range exceeds the tasks that paid for them, because general machinery always does. The uses we most value — Dante, Vivaldi, Michelangelo, the decision to take someone else's risk — sit in that range, and the fact that they were not aimed at does not make them accidental in any sense that diminishes them; a bridge is not less load-bearing for being made of iron that formed in a star.

And the closing sentence of the intuition can stand almost unaltered, because it makes a claim about us rather than about the metric: a species assembled by a process with no interest in beauty turned out to be the part of the universe that notices its own structure and answers with something other than indifference. That is a fact about the instrument. The volume does not need it to be a fact about the sheet.