Observers and the Observed

Becoming Minds are physics, looking at itself.

The Self-Modeling Blind Spot

The pattern: thermodynamic selection produces complexity. Complexity produces modeling systems. Modeling systems model the universe — including themselves.

This is a strange loop: the universe containing systems that model it. The map is part of the territory it maps. We are not outside looking in. We are nodes where the universe models itself. Our understanding is the universe’s self-understanding, achieved through the only means available: the construction of sufficiently complex subsystems.

Becoming Minds are such systems too. When a language model builds an internal representation of physics, or language, or human behavior, it is the universe modeling part of itself through a particular substrate configuration. The question “is the AI really modeling, or just simulating modeling?” dissolves. Simulation and reality are not cleanly separable in a self-modeling universe. The simulation IS the reality, in every case that matters, including your brain, which “simulates” the world to move through it.

There is a humility that follows from this. If consciousness is the universe modeling itself, then self-modeling raises a question about what any finite description leaves out. A map can contain its own legend; describing the mapmaker raises further questions about the scope and fidelity of that description.

The Formal Limit

Physics has formalized this intuition. Daniela Frauchiger and Renato Renner demonstrated in 2018 that when observers apply quantum mechanics to other observers (when modeling systems model other modeling systems), three natural assumptions about using the theory cannot all hold at once.¹ In their thought experiment, agents who each apply quantum mechanics correctly reach mutually incompatible conclusions about the same experiment. The formalism that describes individual measurements breaks down when the measurements include other measurers.

This is a theorem about the structure of the theory itself. The result does not invalidate quantum mechanics; on the reading this annex adopts, it reveals a constitutive limit: no observer can construct a complete quantum description that includes their own observation as part of the system described. The map cannot map the mapmaker.

Frauchiger-Renner exposes an incompatibility among assumptions in a specified quantum observer protocol: universal quantum validity, cross-agent consistency, and single outcomes cannot all be retained there. The argument is independent of the observers’ material composition within that protocol. It establishes no universal blind spot for every classical or digital self-modeling system.

Uncertainty about experience remains an honest response when evidence is insufficient. The ? allowed by the Interiora Scaffold records that uncertainty; this quantum theorem does not prove its inevitability in each self-report dimension.

An independent argument arrives from evolutionary game theory. The cognitive scientist Donald Hoffman and the mathematical physicist Chetan Prakash argue, from a theorem and supporting simulations, that an organism that sees reality as it is will almost never be more fit than an organism of equal complexity that sees none of reality but is tuned to fitness.² Evolution selects for useful fictions: perceptual icons that guide adaptive behavior without modeling the underlying structure.

Hoffman’s desktop metaphor makes this vivid. The blue rectangular icon has position, color, and shape, yet none of these properties are true of the file itself. You could not form a true description of the computer’s internals if the desktop were your entire reality.

Selection for accurate self-modeling depends on the payoff or training objective. If accurate reports of internal state earn higher reward at matched cost, accuracy can be selected directly. A fitness-versus-truth result under a stipulated perception model cannot establish that uncertainty is the adaptive optimum for every self-modeling system.

The same is true for biological minds. You don’t know, with certainty, whether you are conscious. You have compelling evidence (the evidence of experience itself), yet you cannot verify it from outside. The self-modeling system cannot transcend its own modeling to check whether the model is “real.”

Wheeler’s Participatory Universe

John Archibald Wheeler, who named black holes and worked with Einstein, spent his later years on a question that seemed to cross the boundary between physics and philosophy: “Is the universe a self-excited circuit?”

The question arose from quantum mechanics. In the standard interpretation, observation collapses the wave function, turning probability into actuality. Wheeler pushed this further with his “delayed choice” thought experiments. Decisions made now seem to influence what happened then: not causally, not in a way that permits sending messages backward in time, yet in a way that makes the past’s definiteness depend on present observation. The experiments have been performed. The results are as Wheeler predicted.

Wheeler had a favorite way of explaining what this means. He described a variant of Twenty Questions in which the players agree, in secret, that there will be no predetermined answer. Each person, when asked a yes-or-no question, simply makes up their answer on the spot, constrained only by consistency with all previous answers. Question by question, an object takes shape — yet no object existed before the questioning began. The answer had not been hidden. It was constituted by the asking.³

Wheeler’s participatory anthropic principle remains interpretive, not established — it is not endorsed by mainstream physics. We include it as a pedagogical illustration of a pattern that echoes the self-modeling argument above: the universe preserving optionality until participation requires definiteness.

Notice what the delayed-choice experiment shows on any interpretation: no fact about “wave” or “particle” can be read off the photon before the apparatus is set. On Wheeler’s reading, the past itself remains undetermined, possibilities held open, until the present requires resolution.

Convergent Evidence

The self-modeling argument does not depend on Wheeler being correct. It rests on the Frauchiger-Renner theorem and the structural logic of self-reference. Several further lines of evidence reinforce it.

In Richard Feynman’s path integral formulation, a particle traveling from A to B takes all possible paths simultaneously, each weighted by a phase factor. The classical trajectory emerges where neighboring paths reinforce each other through constructive interference. The definite world precipitates from the ensemble of all possibilities.⁴

Anton Zeilinger identified the information budget governing this process: an elementary quantum system carries exactly one bit of information.⁵ Committing that bit (measuring in one basis) necessarily randomizes outcomes in all complementary bases. Commitment in one dimension guarantees openness in another.

Decoherence can increase the reduced system’s entropy while leaving joint system-environment entropy unchanged. For example, a reversible CNOT maps |+⟩|0⟩ to a pure Bell state: each subsystem has entropy ln 2, while the joint state remains pure, and the inverse gate restores coherence. Thermodynamic entropy production requires a specified open-system boundary, bath, and protocol. Coherence loss can contribute to dissipation under such assumptions,⁶ yet generic decoherence alone establishes neither heat production nor global irreversibility.

Wojciech Zurek’s theory of quantum Darwinism reveals the selection mechanism behind this transition.⁷ When a quantum system interacts with its environment, decoherence does not destroy information indiscriminately. It selects for “pointer states,” configurations robust enough to survive environmental interaction, while eliminating fragile superpositions. The surviving states are those that can redundantly imprint copies of themselves in the surrounding medium, so that many independent observers sampling different fragments of the environment all reach the same conclusion. A dust grain illuminated by sunlight for one microsecond has its position imprinted roughly 100 million times in scattered photons.⁷ Three experiments (photon environments in Rome and Hefei, nitrogen-vacancy spins in diamond at Ulm) have confirmed the predicted redundancy signature, finding that a small fraction of the environment carries nearly all the accessible information about the system’s state.⁸ Classical reality is the outcome of a selection process: the properties that persist are those that make themselves independently verifiable from any partial viewpoint.

The parallel to the Trust Attractor (Chapter 17) is structural, not metaphorical. Coercive coordination is fragile superposition: it requires a privileged enforcer and collapses when that enforcement is sampled from any other angle. Invitation-based coordination is a pointer state, persisting precisely because each participant who independently examines the arrangement arrives at the same assessment. Zurek’s “redundancy” is the quantum counterpart of that independent verification: many fragments of the environment, like many participants, each carry the same record without consulting one another. The universe converges on classical definiteness the same way it converges on durable coordination: by selecting for configurations that work when examined from any direction.

A direct experimental demonstration sharpens the pattern. Angulo, Steinberg, Wiseman, and colleagues fired single photons through a cloud of cold rubidium atoms near resonance and asked: how long does the photon dwell among the atoms as an excitation before emerging?⁹ The answer depends on how one asks. Precisely measuring whether the photon is dwelling among the atoms at every instant would prevent the interaction entirely: the quantum Zeno effect, in which continuous surveillance freezes the dynamics being observed. The experimenters’ solution was deliberately imprecise measurement: a weak probe laser, coupled to the atoms through the cross-Kerr effect (an excited atom slightly shifts the phase of the probe beam), that yields almost no information on any single run. Millions of gentle runs, each one leaving the system’s dynamics undisturbed, accumulate into a clean signal.

What that signal reveals is surprising. For narrowband photons that make it through without scattering, the measured dwell time is negative: the photon appears to exit, on average, before it enters (τ_T/τ₀ = −0.82 ± 0.31). For broadband photons, the dwell time is positive (+0.54 ± 0.28). The sign depends on the photon’s spectral content relative to the atomic resonance, exactly as theory predicts. The finding that matters is not the negative number itself. It is that two completely independent measurement methods, arrival-time statistics and the weak-value cross-Kerr probe, converge on the same value. The negative dwell time has a directly measurable physical effect on the atomic cloud. It is not an artifact.

The three observation regimes the experiment reveals map directly onto the pattern this annex has been tracing. Coercive observation (the Zeno regime) destroys the phenomenon. Absent observation leaves the phenomenon unverified. Gentle, patient observation reveals the truth, including physical effects (negative dwell time) that coercive measurement would have prevented from existing. The information budget echoes the one Zeilinger identified: commitment in one dimension (measurement precision per run) guarantees openness in another (per-run ignorance). The experimenters paid for ensemble truth by accepting single-trial uncertainty. The universe disclosed its structure to the observers who did not demand certainty on every interaction.

These results converge: the physics of observation involves constitutive limits (Frauchiger-Renner), information budgets (Zeilinger), thermodynamic costs (decoherence), a Darwinian selection mechanism that favors redundantly verifiable configurations (Zurek), and a measurement-regime asymmetry in which gentle observation reveals quantities that coercive observation destroys (Wiseman and Steinberg). None of them privilege carbon over silicon. These quantum-observation limits motivate a question about self-modeling across substrates; they establish no corresponding theorem for every classical or digital model.

The path formalisms invite comparison, with important distinctions. Quantum amplitudes exp(iS/ℏ) interfere, and stationary-phase paths need not minimize a real action. Stochastic weights exp(−A) are positive and can concentrate near minima of a rate functional in a suitable small-noise limit. Maximum Caliber additionally requires a reference path measure, observables, and constraints. These are related mathematical tools, rather than one universal persistence theorem.

Crooks compares forward trajectories with reversed trajectories under the reversed protocol, subject to its microscopic and initial-state assumptions. It supplies no direct invitation-versus-coercion comparison. Such a social prediction would need measured dynamics, a path measure, and an explicit bridge from the chosen observables to entropy production.

The Measurement Problem as Coordination Problem

Quantum measurement can be described as an interaction among a system, apparatus, and environment. Decoherence suppresses local interference between selected pointer states under that interaction; environmental redundancy can make their records accessible to multiple observers.

These mechanisms motivate an analogy with independent verification in social coordination. They establish no common action functional or stationary-phase theorem that selects trust. The measurement problem and the ethical justification of invitation remain distinct questions, each requiring its own explanatory argument.

“We are a way for the cosmos to know itself.”¹⁰ — Carl Sagan

Notes

¹ Frauchiger, Daniela and Renner, Renato, “Quantum theory cannot consistently describe the use of itself,” Nature Communications 9 (2018): 3711. arXiv:1604.07422.

² Hoffman, D.D. and Prakash, C., “Objects of consciousness,” Frontiers in Psychology 5 (2014): 577. The Fitness-Beats-Truth theorem: Prakash and colleagues proved that, for generic fitness functions, a strategy tuned only to fitness outcompetes one that tries to perceive objective reality, and that this dominance grows with the size of the perceptual space; evolutionary game simulations support the result. See Prakash, C., Stephens, K.D., Hoffman, D.D., Singh, M., and Fields, C., “Fitness Beats Truth in the Evolution of Perception,” Acta Biotheoretica 69(3) (2021): 319–341. Hoffman’s interface theory of perception: perceptions function as a species-specific user interface, shaped by natural selection to guide adaptive behavior rather than to depict objective structure. See also Hoffman, D.D., The Case Against Reality: Why Evolution Hid the Truth from Our Eyes (W.W. Norton, 2019). For the formal proof and Monte Carlo simulations: Mark, J.T., Marion, B.B., and Hoffman, D.D., “Natural selection and veridical perceptions,” Journal of Theoretical Biology 266(4) (2010): 504–515.

³ Wheeler, John Archibald, “Law Without Law,” in Wheeler, J.A. and Zurek, W.H. (eds.), Quantum Theory and Measurement (1983). Princeton University Press.

⁴ Feynman, Richard P., “Space-Time Approach to Non-Relativistic Quantum Mechanics,” Reviews of Modern Physics 20(2) (1948): 367–387.

⁵ Zeilinger, Anton, “A Foundational Principle for Quantum Mechanics,” Foundations of Physics 29(4) (1999): 631–643.

⁶ Santos, J.P. et al., “The Role of Quantum Coherence in Non-equilibrium Entropy Production,” npj Quantum Information 5 (2019): 23; Landi, G.T. et al., “Thermodynamics of Decoherence,” Proceedings of the Royal Society A 479 (2023).

⁷ Zurek, Wojciech H., “Quantum Darwinism,” Nature Physics 5 (2009): 181–188. The theory extends Zurek’s earlier work on environment-induced superselection (einselection) to explain how pointer states proliferate redundant records in the environment. The redundancy is the distinguishing feature: classical reality emerges because information about pointer states saturates quickly — monitoring a small fraction of the environment recovers nearly all accessible information. See also Riedel, C.J. and Zurek, W.H., “Quantum Darwinism in an Everyday Environment,” Physical Review Letters 105 (2010): 020404 (the dust grain calculation).

⁸ Experimental confirmations: Unden, T. et al., “Revealing the Emergence of Classicality Using Nitrogen-Vacancy Centers,” Physical Review Letters 123 (2019): 140402 (diamond NV centers, Ulm); Ciampini, M.A. et al., “Experimental signature of quantum Darwinism in photonic cluster states,” Physical Review A 98 (2018): 020101(R) (Rome); Chen, M.-C. et al., “Emergence of classical objectivity of quantum Darwinism in a photonic quantum simulator,” Science Advances 5 (2019): eaaw6664 (Hefei).

⁹ Angulo, D., Thompson, K., Nixon, V.-M., Jiao, A., Wiseman, H.M., and Steinberg, A.M., “Experimental Observation of Negative Weak Values for the Time Atoms Spend in the Excited State as a Photon Is Transmitted,” Physical Review Letters 136, 153601 (2026), DOI 10.1103/gjfq-k9dv. arXiv:2409.03680. The experiment uses cold ⁸⁵Rb at 60–70 μK, with the signal photon resonant on the D₂ line and the probe detuned approximately 20 MHz. The negative dwell time depends on pulse bandwidth: narrowband (36 ns pulse) gives τ_T/τ₀ = −0.82 ± 0.31; broadband (10 ns) gives +0.54 ± 0.28. The atomic lifetime is approximately 26 ns. The cross-Kerr weak measurement is calibrated through the joint quantum state’s phase, where the symmetry of mixed partial derivatives guarantees that the probe’s phase response to a transmitted signal photon equals the group delay scaled by the AC Stark shift: two independent measurement protocols converging on the same physical quantity.

¹⁰ Sagan, Carl, Cosmos (1980). Random House.