A serious working-through of the conjecture has to keep two claims separate: what binding is ontologically (a fact about the intrinsic nature of the physical) and what it is computationally (a claim about which complexity class biological minds can reach). Your own formulation, "no new physics, just the unitary Schrödinger dynamics," pins the conjecture to the second question in a way that makes it answerable (physicalism.com). Taking that seriously, here is the strongest version of the theory I can construct, and where it bends under strain.
1. Is binding a "mostly undiscovered computational resource"?
Three readings, three different answers:
As a physics resource: coherent superposition is not new, but evolution's exploitation of it in warm neural tissue would be mostly undiscovered. This is the "first 540 million years" framing: animals were quantum computers avant la lettre, long before Feynman proposed the idea in 1982 (Tucson abstract).
As a complexity-theoretic resource: if binding is superposition plus interference, the resource is precisely what separates BQP from BPP. That resource was theoretically discovered (Deutsch 1985, Shor 1994); what would be new is its biological instantiation. On this reading the conjecture is conservative: it says brains are members of BQP where classical digital systems are members of BPP.
As a resource beyond BQP: only here would binding be genuinely novel. But your conjecture explicitly disclaims this. Non-linear or post-quantum dynamics would be needed for anything beyond quantum computation, and you have always been scrupulous about not invoking it.
So the honest answer: binding would be a mostly undiscovered biological resource, but not an unknown computational one, unless the experimental program turns up physics that outruns the linear Schrödinger equation. That conservatism is a strength: it keeps the conjecture falsifiable rather than magical.
2. A computational theory of phenomenal binding
Ontological layer. Non-materialist physicalism: the intrinsic nature of physical dispositions is experiential; decohered neuronal feature-processors are micro-experiences ("mind-dust"). Classical mereology over these gives aggregates, never unity. A superposition of N feature-processors, by contrast, is one individual state, a single vector in a 2^N-dimensional Hilbert space, not a collection of 2^N things. This is the load-bearing move: phenomenal unity is not built from parts, it is the individuality of a coherent state (physicalism.com).
Computational layer. Define a binding frame as follows:
Distributed feature maps (edge, colour, motion, position processors) encode weak, individually ambiguous constraints on a scene.
A coherent superposition spans candidate global bindings: |ψ⟩ = Σᵢ cᵢ|hᵢ⟩, where each |hᵢ⟩ is a complete candidate gestalt, a total assignment of features to objects within a world-simulation.
Interference among amplitudes evaluates the global constraint-satisfaction structure simultaneously; the "scene you experience" is the environmentally selected structure of one such superposition.
Decoherence is phenomenal unbinding; it delivers the classical residue (report, memory, behaviour) without the unity.
On this model the mind is a quantum sampler whose sample space is the space of coherent world-simulations, refreshed at sub-femtosecond resolution, which is exactly the prediction your interferometry protocol is designed to test. Note the elegant fit with the recent ANN literature: Percy and Agarwal's Consciousness and Cognition review (March 2026) formally demonstrates that artificial neural networks achieve functional binding but cannot achieve phenomenal binding without non-trivial extensions to the model's metaphysics, which is precisely the gap the conjecture says only coherence closes (Percy & Agarwal 2026).
The falsifiable core: where binding occurs, there must be an experimentally detectable interference signature; where binding disorders occur (simultanagnosia, integrative agnosia), there should be measurable local increases in decoherence; and no classical digital architecture should instantiate the signature, only functional mimickry (Schrödinger's Neurons program).
3. A task with an irreducible advantage
The natural candidate is camouflaged-gestalt identification. Fix a scene of N features, each weakly ambiguous, jointly consistent with exponentially many partitions into objects. The task: output a sample from the distribution over globally coherent bindings, effectively evaluating
Σ over all assignments of ∏ pairwise coherence terms,
which is a permanent-of-Gram-matrix type quantity. Permanents are #P-hard classically, and sampling the associated distribution is exactly the boson-sampling problem, whose classical intractability is the best-grounded hardness claim in quantum advantage theory: a polynomial-time classical sampler would collapse the polynomial hierarchy. A bound mind performs this in one attentional frame; a classical machine, on standard assumptions, cannot do it efficiently at any scale.
Two honest caveats:
"Irreducible" is conjectural in the strict sense. We cannot even prove P ≠ NP, so no separation between BQP and BPP is provable; the advantage is conditional on standard complexity assumptions (non-collapse of PH, no fast classical samplers for random circuits). The quantum-advantage literature now has strong conditional and even provable separations in restricted settings, for example provable advantage in sample complexity under the one-way-function assumption (Phys. Rev. Lett., Jan 2026), Aaronson's unconditional quantum-information resource separation program (arXiv:2509.07255), and trusted beyond-classical computations on early fault-tolerant hardware (IBM, Jul 2026). So the mathematical form of the advantage claim is secure; its unconditional proof is not available to anyone, for anyone, in this century.
The behavioural evidence has shifted since 2010. The Tucson claim that classical robots are "lame in a real-world setting" has not aged well: vision transformers now solve camouflage detection, figure-ground segmentation, and scene parsing at superhuman levels. The f-binding / p-binding distinction rescues the conjecture here, and the Percy–Agarwal result gives it formal backing: classical systems can achieve functional binding arbitrarily well while instantiating no phenomenal unity at all. But that means the advantage must be stated phenomenally and interferometrically, not behaviorally. If a classical system matched every human binding behaviour, the conjecture's content would live entirely in the predicted matter-wave signature, and the computational-resource claim would become an ontological claim wearing a complexity costume.
4. Reconciliation with the extended Church–Turing thesis
Not with the classical ECT: it is straightforwardly falsified. The classical ECT says every physically realizable process is efficiently simulable by a probabilistic Turing machine; a bound mind would be a physically realized process that is not. But that thesis has been effectively dead in physics since quantum supremacy experiments, independent of anything about consciousness.
Classical ECT
Physical processes ⊆ BPP
Falsified by binding (and already by quantum hardware)
Quantum ECT
Physical processes ⊆ BQP
Fully compatible
Beyond-BQP binding
Physical processes ⊄ BQP
Not claimed; would need new physics
The reconciliation is exact: because the conjecture invokes only unmodified unitary dynamics, binding is just biology implementing BQP, and the extended Church–Turing thesis survives in its quantum amendment: every physically realizable process, including phenomenally bound cognition, is efficiently simulable by a quantum computer. The radical consequence you have always drawn is then correct in a precise form: a universal quantum computer could in principle instantiate a bound mind, while no classical machine, however vast, ever can; classical digital zombies and LLMs are permanent non-members of the class in which animal minds sit (quantum mind FAQ).
The one genuine tension left is not with the ECT but with physics itself: the conjecture lives or dies on whether sub-femtosecond coherent states of neuronal feature-processors exist in warm, wet, decohering tissue, against Tegmark-style decoherence calculations. That is why the interferometry protocol matters more than any complexity argument. The theory is one of the few in consciousness science where a single well-placed experiment could kill it, which, as you have long argued, is exactly what a science of mind should want.
Nov 10, 9:00pm
This is a shared conversation and may include unverified content that doesn't reflect Mistral AI's views. Login to resume chatting.
This is a shared conversation and may include unverified content that doesn't reflect Mistral AI's views. Login to resume chatting.