Part V · Character

# The Geometry of Qualia

Valence is the sign and size of an experience. Its character — what it is like, rather than how good it is — lives in the curvature and holonomy of the state manifold.

> Assumes
> Chapters 4 and 8. Riemannian geometry to the level of sectional curvature; parallel transport.
> Delivers
> Theorem 10.1 (constant negative curvature); the holonomy proposal for qualitative character.

10.1 The Bures metric

The final entry in the invariant (2.2) is a geometry. The state space of the split factor $\mathcal{N}$ is not a bare set; it carries a canonical Riemannian metric, the Bures metric, which is the quantum Fisher information metric:

$$ ds^2_{\mathrm{B}}=\frac{1}{2}\sum_{j,k}\frac{\bigl|\langle j|d\rho|k\rangle\bigr|^2}{p_j+p_k} $$ (10.1)

Its operational meaning is distinguishability: the Bures distance between two states is, up to normalisation, the number of measurements needed to tell them apart. This is exactly the right notion for a phenomenal geometry. Two experiences are close when they are hard to tell apart from the inside, and the metric that measures that is the one that measures statistical distinguishability.

Restricted to a commuting family, (10.1) reduces to the classical Fisher–Rao metric $ds^2=\sum_i (dp_i)^2/p_i$. And on one particular family, the Fisher–Rao metric has a closed form that has been known since Rao's 1945 paper and is, for our purposes, remarkable.

10.2 The geometry is hyperbolic

> Theorem 10.1 · Constant negative curvature
>
> Established
>
> The Fisher–Rao metric on the family of univariate Gaussians $\mathcal{N}(\mu,\sigma^2)$ is
>
> $$ ds^2=\frac{d\mu^2+2\,d\sigma^2}{\sigma^2} $$ (10.2)
>
> which, after the substitution $\mu\mapsto\sqrt{2}\,u$, is twice the Poincaré metric on the upper half-plane. Its Gaussian curvature is constant and negative: $K=-\tfrac{1}{2}$.
>
> For the $n$-variate family with fixed mean, the manifold of covariances is the symmetric cone $\mathrm{GL}(n,\mathbb{R})/O(n)$ with metric $ds^2=\tfrac{1}{2}\mathrm{tr}\bigl[(\Sigma^{-1}d\Sigma)^2\bigr]$ — a Hadamard manifold of non-positive sectional curvature, with exponential volume growth.

The space of Gaussian probability distributions is the hyperbolic plane. Not analogous to it, not approximately it: the Fisher–Rao geometry of $(\mu,\sigma)$ is the upper half-plane model, and has been since long before anyone thought to connect it to phenomenology.

10.3 Why an annealed state should feel hyperbolic

> A derivation of a reported phenomenology, which is rare enough to flag. It also fixes a scale*: curvature $-1/2$ in Fisher units, not a free parameter.*

Now combine Theorem 10.1 with Chapter 8. Annealing drives a state toward maximum entropy subject to its second-moment constraints — which is to say, toward a Gaussian on an expanded set of modes. Theorem 10.1 then says something specific and, I think, non-obvious:

As an experience anneals and its mode set expands, the intrinsic geometry of its qualia manifold becomes hyperbolic, with volume growing exponentially in the rank $n$.

Gómez Emilsson's reports of hyperbolic phenomenal geometry under psychedelics — and the recurring, otherwise puzzling description of interior spaces vastly larger than they could contain — are what a Hadamard manifold of growing rank is like from inside. In hyperbolic geometry the volume of a ball grows exponentially with radius, so a space of modest diameter has room for enormously more structure than Euclidean intuition allows. That is a specific geometric fact, and it matches a specific and consistently reported phenomenological one.

The dose–geometry relation is then a prediction rather than a description: reported hyperbolicity should track the dimensional expansion of the coherent mode set, not the drug concentration directly. Two interventions producing the same $n$ should produce the same reported geometry regardless of pharmacology, and a drug that raises concentration without expanding $n$ should not produce the effect at all.

10.4 Character as holonomy

Curvature tells us the shape of the space of experiences. It does not yet tell us what distinguishes one experience from another in kind — why red is not a sound. For that we need an invariant that is not a distance.

The Bures metric comes with a connection, and parallel transport of a state around a closed loop $\gamma$ in state space returns it rotated by the Uhlmann holonomy $U_\gamma\in\mathcal{U}(n)$. Holonomy is the natural carrier of "kind": it is invariant under reparametrisation, it composes correctly under concatenation of loops, and it is trivial exactly when the geometry is flat.

> Proposal 10.2 · Qualitative character
>
> Posited
>
> The qualitative character of an experience is the conjugacy class of its Uhlmann holonomy. Two experiences differ in kind, rather than merely in degree, exactly when their holonomies are non-conjugate.

This completes the invariant (2.2): $g^{\mathrm{B}}$ supplies distinguishability and curvature, its holonomy supplies modality, $\mu_\rho$ supplies intensity, and $P(q)$ supplies valence. I present Proposal 10.2 as the most speculative claim in the book — it is the one place where I am proposing an identification on grounds of formal aptness alone, without either a theorem or a measurement behind it.

10.5 Colour, the tractable case

Colour is where this can be tested, because colour is the one quale with a century of quantitative psychophysics behind it. The state manifold of a three-mode Gaussian family is $\mathrm{GL}(3,\mathbb{R})/O(3)$, six-dimensional, and its chromatic slice is the natural home of the Helmholtz–Stiles line element — the classical attempt to give colour space a Riemannian metric of exactly Fisher–Rao type.

The empirical situation is interesting. Bujack et al. (2022) showed that perceptual colour space fails geodesic additivity: large colour differences are consistently smaller than the sum of the small differences composing them, which no Riemannian metric with the usual assumptions can reproduce. This is at least the right shape of anomaly for a non-flat Fisher geometry, and diminishing returns along long geodesics is what one expects when curvature is negative.

I would not call this evidence. The result is standardly read as showing colour space is not Riemannian at all, which is a stronger and different claim than that it is Riemannian with negative curvature, and distinguishing the two requires an analysis nobody has done. It is listed here as the most promising place to look, not as a confirmation.

> ### What Chapter 10 established
>
> - The Bures metric (10.1) is the canonical geometry of state space, and measures distinguishability.
> - Theorem 10.1: the Fisher–Rao geometry of Gaussians is the hyperbolic plane with curvature exactly $-1/2$; the $n$-variate case is a Hadamard manifold.
> - Annealing expands the mode set toward a Gaussian, so annealed experience should be hyperbolic — a derivation of reported psychedelic geometry, with a fixed curvature scale.
> - Proposal 10.2: qualitative character is the conjugacy class of the Uhlmann holonomy. The most speculative claim in the book.
> - Colour is the tractable test case; the failure of geodesic additivity is suggestive but not evidence.

> ### Exercises
>
> 1. Verify that (10.2) becomes twice the Poincaré metric under $\mu=\sqrt2\,u$, and confirm that scaling a metric by $c$ divides curvature by $c$, giving $K=-1/2$. [2]
> 2. Compute the Fisher–Rao distance between $\mathcal{N}(0,1)$ and $\mathcal{N}(0,e^2)$, and between $\mathcal{N}(0,1)$ and $\mathcal{N}(3,1)$. Comment on the asymmetry between mean and variance directions. [2]
> 3. In hyperbolic space of curvature $-1/2$, compute the volume of a ball of radius $r$ and find the $r$ at which it exceeds the Euclidean volume by a factor of $10^3$. Relate this to reports of "impossibly large" interior spaces. [2]
> 4. Show that the Uhlmann holonomy of a loop confined to a flat submanifold is trivial, and hence that Proposal 10.2 predicts no qualitative distinctions within such a submanifold. Is that a virtue or a defect? [2]
> 5. Geodesic additivity fails in perceptual colour space. Construct two hypotheses — non-Riemannian, versus Riemannian with negative curvature — and specify a measurement that separates them. [2]
> 6. Open. Proposal 10.2 identifies character with a conjugacy class in $\mathcal{U}(n)$. Derive the dimensionality of colour experience — that it is three, not four — from the structure of the relevant holonomy group rather than from retinal physiology. [3]

For agents

GET /api/library/spectral-panpsychism/ch10.md

← Intensity and the Logarithmic Scales  ·  Eight Ways This Is Wrong →