True Tales of Evolutionary Geometry

The model, stated as a model

The previous chapter left smoothness as a matter of the scale at which one system couples to another. Nothing in that statement installs a privileged observer, and the model does not want one: there is no exalted vantage from which the world is seen as it really is, and no seat that reality is staged for. What there are, instead, are physical systems coupled to other physical systems — a photodiode, a rock, a retina — each registering only what its coupling admits. A nervous system is one such coupled system, distinguished not by standing but by history: it was built by selection, it runs on a metabolic budget, and every stage of it discards information. That is why its couplings are worth describing here alongside the rest of the physics.

The model has three movements, and they are not equally secured, so they are separated here rather than fused into one thesis. First: perception evolved for fitness, not for fidelity, so the smooth world an organism inhabits is a functional summary rather than a transcript. Second: biology exploits quantum processes for survival advantage, so it is not absurd to ask whether neural processing does too. Third: neural aggregation performs a biological coarse-graining that converts a jagged substrate into a continuous experience. The first is a well-supported reading of sensory physiology. The second is true in specific documented cases and remains a hopeful question for neurons. The third is a figure — a way of seeing — and is labelled as one.

None of this asks anything of a spectator. The compression happens whether or not anyone is attending to it, and it happens the same way in a bird as in a person, because the mechanism is coupling and budget rather than witness. The model is a sheet draped over the unknowns: a vibrational placeholder, beautiful because it lets the primes, the zeros, and the geometry breathe together without pretending the mechanism is already known.

Selection pays for fitness, not for fidelity

Donald Hoffman's interface theory of perception, developed with Chetan Prakash, gives the model one of its entry points. In evolutionary-game models where organisms with veridical perceptual strategies compete against organisms tuned directly to a fitness function, the veridical strategies are driven toward extinction as the state space grows. The theorems are real and the mathematics is checkable. What they establish is narrower than the headline: in a class of models with fitness functions that are generic and not monotonically related to the world's structure, tracking fitness outperforms tracking truth. Where fitness is a monotone function of a physical quantity — distance to a cliff edge, size of a predator — tracking the quantity is tracking fitness. The model does not need the strong claim that veridical perception is impossible; it needs only the softer claim, which the theorems support, that veridical perception is not automatically favoured.

The softer claim is also textbook sensory physiology. The optic nerve carries roughly a million fibres out of about a hundred million photoreceptors; the retina throws away most of what it absorbs before the signal leaves the eye. Human colour vision samples three overlapping bands of a continuous spectrum. Flicker fusion sets a temporal cut, and above it discrete events read as continuous. The brain costs about a fifth of the body's resting energy, which is an enormous standing bill for an organ that already discards most of its input — and a decisive argument that no additional metabolic capacity was ever available for computing raw microphysics. A system that spent energy resolving what it did not need to act on would be outcompeted by one that did not. That argument is sufficient for the model and does not require the game-theoretic result to do any work.

The desktop-icon analogy is useful and should be held lightly. An icon is a compressed, actionable handle on machinery it does not resemble, and perception is like that. The analogy imports a designer and a user, and there is neither here: no one chose the interface and no one is looking at it. Stripped of the design language, what remains is the model's working claim — perception is lossy compression under a fitness objective — and that claim is enough.

What quantum biology has actually established

The field is real and the standard dismissal — warm, wet, noisy, therefore decoherent, therefore nothing to see — has not survived. The evidence is uneven across the usual examples, and running them together as a single trend is how this material loses its credibility with the people best placed to evaluate it. The model is stronger if it keeps the cases separate.

Strongest: avian magnetoreception. The radical-pair mechanism in cryptochrome, proposed by Schulten in the 1970s, has accumulated genuine support — magnetic-field effects on cryptochrome-4 from migratory songbirds measured in vitro, behavioural disruption by radiofrequency fields at the predicted resonances, and a spin-coherence lifetime long enough to matter. This is a documented case of a spin-correlated quantum state being used for a survival task. It is also not computation; it is a magnetically sensitive chemical yield. For the model, it is enough: biology has already shown that a quantum spin process can be recruited for navigation.

Substantially revised: photosynthetic energy transfer. The 2007 two-dimensional spectroscopy results on the Fenna–Matthews–Olson complex were widely read as long-lived electronic coherence enabling wavelike search of transfer pathways. Subsequent work through the 2010s reassigned most of the long-lived oscillations to vibrational rather than electronic coherence, and the current consensus is that vibronic coupling shapes efficient transfer while coherence lifetimes at physiological temperature are short. Quantum effects are involved; the version where a photon simultaneously explores every path and thereby achieves near-perfect efficiency is not what the data now support. The model can borrow the revised version without embarrassment.

Open: quantum processing in neurons. Orchestrated objective reduction in microtubules — Penrose and Hameroff — remains a live proposal with no confirming experiment and long-standing decoherence objections from Tegmark and others; the anaesthetic and terahertz-vibration results are suggestive and not decisive. The Fisher proposal that Posner molecules could protect nuclear spin coherence in phosphate chemistry is a well-specified hypothesis with a stated experimental programme, and the lithium-isotope result that motivated it has not settled the question. Recent superradiance work in tryptophan networks is interesting and early. None of this licenses the sentence that the brain uses quantum operations to compress the subatomic matrix, and the model does not need it to. Koch's remark — that anything not forbidden by physics can be exploited by evolution — is a good reason to keep looking and is not evidence that anything was found.

Neural aggregation and the shape of the blur

The aggregation movement of the model is the part that stands on established ground, and it does not need quantum mechanics. A single cortical neuron integrates thousands of synaptic inputs; a percept is the state of populations numbering in the millions; a psychophysical judgement pools across time windows tens of milliseconds wide. The stochastic detail of individual events is averaged away by construction, and the averaging is the mechanism by which a discrete, noisy substrate yields a stable, continuous report. Ion-channel opening is stochastic and single-photon absorption is quantised, so the fine grain really is discrete — the smoothing is doing real work on real jaggedness, which is the strongest form of this argument and stays entirely inside neurophysiology.

Two qualifications keep the central-limit framing honest. Neural noise is emphatically not independent — correlated variability across populations is one of the most studied facts in systems neuroscience, and correlations set a ceiling on how much averaging buys, which is why pooling a million neurons does not reduce noise by a factor of a thousand. And the resulting distributions are frequently not Gaussian: neural firing statistics, interval distributions, and perceptual judgement distributions are often skewed or heavy-tailed. The theorem's hypotheses are exactly what fails first in a nervous system, so the correct statement within the model is that aggregation smooths to the extent that correlations permit, which is measurable and has been measured.

Sensory compression also has a known functional form, and it is a better fit to the volume's interests than the smoothing story. Weber–Fechner scaling makes the just-noticeable difference proportional to the stimulus, so perceived intensity grows roughly logarithmically; Stevens's power law fits many modalities better and the two coexist as competing descriptions with a long literature. Either way the transformation is compressive and its consequence is worth stating precisely: a logarithmic front end converts multiplicative structure into additive structure. That is the same operation that turns the Euler product into a sum over log p, and it is why the volume keeps meeting log-spaced quantities. The parallel is structural and it is not evidence that perception is reading primes. The model treats it as a rhyme, not a derivation.

Where the model is currently mute

The model asks for no privileged observer and grants no one a special seat. It requires only that any evolved nervous system, coupled to a field at its own scale and paying its own metabolic bill, compresses a high-dimensional substrate into a low-dimensional, actionable summary. What is compressed, and how, is a question the model opens rather than answers.

There is no exhibited map from the stair-stepped primes or the non-trivial zeros to any sensory quantity, so the sentence that neurology averages the upwellings of prime frequencies into a solid rock is a figure with no operator behind it. The earlier chapters were explicit that the prime-and-zero language awaits an operator; nothing about adding a nervous system supplies one. The correspondence table — micro-matter bedrock to quantum coherence to bounded objects, stair-stepped primes to Weber–Fechner to smooth perception, torsional shear to eigenvalue bounds to a stable spacetime canvas — is a way of seeing, not three established mappings. The middle column is real psychophysics and physiology; the left column is this volume's own unexhibited proposal; and lining them up in a table makes an alignment look like a derivation only if the reader forgets the model's status.

Also outside the model: the claim that evolution built the brain to hide the harshness of quantum reality. Selection has no such intention and the model does not need the world to be harsh. What is true is simpler: an organism that resolved detail it could not act on would pay for it and lose. Nothing needed hiding; the resolution was simply never purchased.

A symmetry worth naming, because it cuts across the whole volume. If perception is an evolved, fitness-tuned compression, then so is the mathematics done by brains — including this volume's mathematics, and including the intuition that a jagged substrate underlies a smooth appearance. The model does not exempt the arguer. That is not a reason to abandon the line of thought; it is a reason to keep the falsifiers attached, because a compressed instrument checking its own compression has only external tests to rely on.

What would sharpen or replace the model

The model survives as a way of seeing: perception as lossy compression under a fitness objective, argued from bandwidth and metabolic cost; radical-pair magnetoreception as a documented case of biology using a quantum spin state for a survival task; vibronic coupling in photosynthetic transfer as a real quantum contribution with revised claims; population averaging as the mechanism producing continuous report from discrete substrate, bounded by correlated variability; and logarithmic or power-law sensory compression as the established functional form of the front end.

What remains open: quantum computation in neurons, which is a hypothesis; the probability-zero reading of Hoffman's results as a statement about human vision; the long-lived electronic coherence version of the photosynthesis story; and any identification of the perceptual substrate with primes or zeros.

Three tests would sharpen or retire parts of the model. The Fisher programme is the sharpest: measure nuclear-spin coherence times in Posner molecules in physiologically realistic conditions, with a stated threshold below which the neural proposal is dead. For microtubules, a decoherence-lifetime measurement in situ, against the Tegmark estimates, decides whether the mechanism has room to exist at all. For the aggregation claim, the prediction is already testable and partly tested: measure the correlation structure of a sensory population and show that the perceptual discrimination threshold matches what the measured correlations permit, rather than what independent pooling would predict. And the whole model would move from figure to result only on the exhibition of the missing operator — some quantity in a nervous system that is a function of a number-theoretic spectrum, stated with units. Until then this is a chapter about how an evolved instrument produces a smooth reading, which is worth having, and not a chapter about what it is reading.