Can an Echo Become a Voice Again?

"Life, like a mist, appears for just a day, then disappears tomorrow."

— twin brother's funeral

We live so intimately involved with the human-relevant scale of reality that when a person leaves we may find it difficult or impossible to reorient our life around their absence. We may search for perspectives or beliefs that preserve their existence even at the cost of severe cognitive distortions, prolonged grief, and disappointment. It can feel impossible to just accept that a precious human life can simply… cease to exist.

But how real was human life ever in the first place? From a physics perspective, a "static" structure, seen honestly, is often a dynamic "equilibrium" which has become so stable and familiar that we mistake it for stillness. In this temporal universe, it has always been dynamics first, statics second, process before substance. An electron's state gives us probabilities for interactions, rather than the trajectory of a tiny classical bead. A solid object's resistance to your hand is likewise an outcome of interactions among its constituents. Sorry, I know those two examples are a bit unproductive to consider in most situations, but now climb the ladder towards the human experience and you get contingencies that really do often matter: "feelings are just chemicals" is actually pretty relevant to keep in mind when they're leading you down a self-destructive path. Relationships as dynamically maintained social structures, rather than indefinitely sealed connections. Memory often better understood as an attractor basin pressed into the cortical neurodynamics rather than a permanent fixed 'record' in a database. And ofc, the human body continually regenerates itself to preserve the illusion: protein turnover, "replaced every 7 years" (lol—not literally), etc., etc. All these examples depend on ongoing physical processes and will deteriorate and eventually cease to exist in their original form (beyond the footprints, fingerprints, and memories they made) if those processes stop. Hence, "life is like a mist".

When asking what a person ever really was, we might pay more attention to its structural definition than its particular constituents. Think about the shape of its dynamics regardless of what substrate maintained them. You don't need individual-neuron ion potentiation awareness to predict every useful mesoscale feature; brain–computer interfaces already exploit the fact that aggregate neural activity constrains intentions and actions—even flying drones with EEG! The shape of the world-relevant dynamics which externally define the footprints, fingerprints, and memories we know a person by may actually be quite low dimensional—regardless of how high-dimensional their brain's billions of neurons make its internal state. Many computational savings to be had by modeling structure in latent space instead of observable state space! How much individuality survives that compression is part of the question.

And when understanding of another person: everything we know about them has passed through their capacity for expression and our capacity for perception. The traces that make it thru both bottlenecks may constrain some of the medium-scale surface structure while leaving much uncertainty about what the person was like on the inside. Low-dimensional observations can be informative without being sufficient. The distinction matters: a narrow window onto a system does not establish that the system itself has few degrees of freedom.

Can an echo become a voice again? If we were only asking for a convincing voice, the answer is becoming a creepy yes—given enough suitable recordings, ofc. But modeling beneath the surface, imagine a system that extracts cognitive-scale features—goals, preferences, memories, habits of inference—from notes, actions, conversations, and other traces, then constrains a reconstruction of the machinery that generated them. The central question is whether we can recover that machinery closely enough for its characteristic dynamics to continue.

Roughly speaking this means something like: observation → internal state → action, with an estimated world state sitting somewhere in the middle. Tho there is no single canonical cognitive schema. To be more nuanced, real and recalled observations are both interpreted through the current internal state to form perceptions; perceptions perturb existing beliefs, goals, feelings, and memories; those interactions nucleate thoughts and candidate plans; plans compete for attention and eventually become actions; actions modify both the world and the person acting within it, producing new observations and beginning the cycle again. Memory is not merely an input retrieved from storage but part of the dynamical machinery determining which trajectories the system is likely to enter next. And the whole experience from a first-person inside-out perspective, thinking much faster than the world around changes, is flipped entirely to internal-first, action-second, observation-third!

Memory makes this concrete. A partial cue—a sensory fragment, a phrase, an internally generated association—can recruit a much larger pattern of activity. Human experiments link hippocampal pattern completion to reinstatement of distributed cortical representations, including elements of an episode that were not explicitly requested. Recurrent cortical and thalamocortical interactions can help sustain the resulting activity; experiments on thalamic support of working memory show one part of that machinery. No single region has to contain or drive the whole experience. An initial fragment recruits the rest, and the brain grabs onto the unfolding pattern. Attractor-basin re-entry is a useful model of this reconstruction, without making every memory a fixed point. Recall can also modify what is subsequently retained; reconsolidation concerns that updating and restabilization, rather than simply naming the act of retrieval.

And not just cognitive structure, emotions too are better understood less as discrete "objects" the brain contains and more as relatively low-dimensional control variables distributed across this process. EG, fear changes which possibilities receive attention, how uncertain evidence is interpreted, which memories become accessible, and which actions become attractive. Curiosity changes exploration policy. Attachment changes the value assigned to another person's predicted states. These abstractions are "real" in roughly the same sense relevant here: because they identify stable regularities in the system's dynamics that improve our ability to predict what it will do. Their physiological realization may involve many mechanisms, rather than a single chemical or neural landmark.

Now take a step back and think about the conceptual abstractions we've just sketched out together: cognitive dynamics, emotional dynamics, …I'm sure you could sketch out social dynamics also. Can you imagine a Bayesian-like recurrent graph describing all of this? Observation and collective internal state feeding into perception. Perception conditioning emotional state. Etc. etc. etc. And then also imagine this graph with specifically engineered neural networks for each of the nodes, to express their characteristic dynamics: Some kind of associative energy-based model for the most general internal state reservoir, activity, and memory. Internal state proximity roughly following hardwired body plan, borders up close in constant flux given by activity. And then, even more intricately engineered subcortical architecture: Value estimator updates around the objects of a person's love. World, pose, head, and eye orientation encoding. Recurrent integration bottlenecks and high-dimensional similarity lookup mechanisms. And—is this starting to sound like a brain? I'm not saying it has to be any of these—as long as it recreates the same human-scale relevant dynamics then it may be equivalent for the functional purposes we care about.

The ambition is to recover a process that can encounter a state of the world it has never seen before, integrate it with its memories and attachments, select an action, observe the consequence, and change. Reproducing old outputs is one constraint on that process. Generating characteristically personal new trajectories is a much stronger requirement. Can an echo become a voice again? Well, regardless of how functionally equivalent we can make both the surface and the core generating process that distinguished a person, ask yourself: would you personally consider that to be the same real person once re-created? Do you actually believe any of this? I'm asking you to think about this because humanity has always been so deeply anchored to the first-person human experience; we find it hard to truly re-orient our internal conceptual schema around a god's-eye view of the same experience.

A brain, in a body, in a world

The object I have in mind is a model whose internal causal organization follows the mesoscale structure of a human brain: its regional organization, long-range pathways, recurrent dynamics, and coupling to a body in a world. Movement, perception, need, and consequence close the loops through which its cognitive structure expresses itself and changes. This correspondence matters because a historical influence sensitive to causal organization would need something with the relevant organization to act on.

We can begin with the anatomy and dynamics already accessible to measurement, then resolve finer distinctions as modeling improves—perhaps eventually down to individual neurons. The point is to progressively recover the causal structure that makes this person's brain behave as it does. That is the structure the later physical reconstruction would try to preserve while exposing its remaining uncertainty.¹

One generative model, one evidence model

The reconstruction machine itself has a simple conceptual schema: a generative model coupled to an evidence model, iterated through evolutionary sampling. The generative model proposes candidate brain–body organizations. The evidence model evaluates how well they explain the surviving traces. Candidates are retained, varied, and tested again. Gradient-based fitting, posterior sampling, and population-based search are possible implementations of that same relationship.

Let $\theta$ denote a candidate's relatively persistent organization: connectivity, process parameters, learned associations, and relevant body properties. Let $x(t)$ denote its evolving state, and $u(t)$ its interaction with a world. An ordinary model supplies dynamics $f_\theta$ and an observation model relating trajectories to evidence $e$. A compact statement of the digital reconstruction problem is:

$$p(\theta, x_0 \mid e) \;\propto\; p(e \mid \theta, x_0)\, p(\theta, x_0)$$

The prior includes shared human anatomy and physiology. The likelihood concerns the particular person. Where context is missing, it must be integrated over rather than silently fixed to whichever imagined circumstance makes the candidate look best. A diary entry, a remembered disagreement, and an electrophysiological recording have different error structures. Ten retellings of one event do not become ten independent constraints.

We can begin below the level of whole personalities. Sample families of memory associations, learning rules, attentional transitions, affective regulation, and inference habits; constrain their interactions; then assemble increasingly complete organizations. "The space of all brains" is a useful horizon, but each experiment needs a bounded model class and a declared sampling measure. Otherwise a preference may simply reflect how often a convenient encoding happens to generate one kind of candidate.

A cognitive fingerprint is a relationship among responses. Perhaps a person changes an interpretation after a particular reminder, remains unusually cautious under one kind of uncertainty, or generalizes a lesson in a distinctive direction. A phrase on its own may be commonplace; that phrase addressed to this person, after this experience, with this attachment at stake can be unmistakable. The conjunction carries the signature. Each candidate must explain those relationships together and predict traces held back from fitting.

This digital stage could do substantial work. It can exploit large datasets, fast simulation, and progressively better population models. It can also reach a genuine ceiling. Different internal organizations may explain every available trace. Querying those candidate models creates predictions, but does not supply an absent person's answer. When the observations no longer distinguish them, a sharper posterior obtained solely by stronger regularization is stronger preference, not new historical knowledge.

The compatible candidates need not form a linear subspace. They may occupy disconnected regions with different memories and mechanisms but similar observable consequences. Averaging them can produce a system that was never a plausible candidate at all. The goal is therefore identification of a coherent organization, or an explicitly unresolved family, rather than convergence to a parameter mean. This is where the second, speculative branch begins.

Could past organization become additional evidence?

The question I want to leave open is whether prior instantiation can influence subsequent reconstruction beyond the ordinary physical traces through which history is already transmitted. Suppose a sufficiently similar system were slightly more likely to develop toward an organization that had existed before than toward a matched organization that had not. Then surviving records would not necessarily exhaust the evidence available to a reconstruction instrument. Its own physical behavior might carry an additional, history-dependent constraint.

Rupert Sheldrake's morphic resonance proposes an influence of past similar organizations on subsequent ones. I am borrowing this particular question from that proposal, rather than treating its broader framework as established. Michael Levin's work belongs at a different evidentiary level: experiments on planarian regeneration show that perturbing bioelectric signaling can produce persistent changes in regenerative outcomes. They motivate taking distributed physiological organization seriously. They do not demonstrate a historical channel surviving the loss of its physical carrier. The bridge between these ideas is a research question, not an experimental result.

One way to motivate the question is to imagine the universe as a generative computation. A mature brain contains the consequences of development, learning, attachment, and repeated interaction. Its present organization can depend on a long history even when its immediate behavior occupies a relatively restricted set of trajectories. If the hypothetical substrate generating physical evolution could reuse previously realized organization, perhaps reconstructing an old structure would carry a different effective cost from generating a comparable new one.

That picture introduces an additional assumption. Compact update rules alone do not imply that reality caches past cognitive structures, optimizes their reconstruction, or turns computational savings into physical probabilities. Ordinary entropy increase supplies no such implication either. The proposal is specifically that historical reuse has a small, measurable effect on subsequent dynamics. It can be discussed without asserting an external simulator as fact; "reconstruction cost" names a candidate explanation of the effect, not a quantity already read from nature.

Several quantities need to remain separate. Conditional description length, $K(\theta \mid b, e)$, concerns how much specification remains once background structure $b$ and evidence $e$ are available. Bennett's logical depth instead concerns the computation needed to generate an object from a near-minimal description. Neither is identical to the actual runtime of its history. A long development may contain compressible detours, and a short description may require an enormous computation. Statistical complexity concerns predictive state; attractor dimension concerns trajectories within a system. None alone measures a person's historical recoverability.

The intuition is historical path-dependence: organization whose efficient reconstruction might depend on access to its past. My stronger conjecture is that individual human minds are exceptionally distinctive targets—perhaps more discriminable, at the relevant causal scale, than common inanimate processes, many animal behaviors, or members of a shared artificial-model family. Decades of development anchor their schemas to particular bodies, people, places, and contingencies. This is a proposed source of experimental contrast, not an established ranking of everything in the universe. An easily copied AI checkpoint may still have been extraordinarily expensive to produce; copyability and the specificity of a lived history are different properties.

For a concrete effective model, take $\theta$ to parameterize the physically instantiated candidate and $x$ its faster brain–body dynamics. Write the evolution of the adjustable coordinates as:

$$d\theta = \big[\, b_0(\theta, x, u) + \lambda\, r(\theta, x, h) \,\big]\, dt + \sigma(\theta, x, u)\, dW(t)$$

Here $b_0$ and $\sigma$ describe the calibrated ordinary drift and noise of the apparatus, including its controls and relevant environmental memory. $h$ denotes specified historical instantiations. $r$ is the hypothesized additional drift toward historically related organization; $\lambda$ controls its strength. The null is $\lambda = 0$. This is a coarse-grained experimental parameterization, not a fundamental quantum equation. A quantum implementation would still require a physical encoding, dynamics, and measurement model that justify these effective variables.

One possible version sets $r = -\nabla_\theta C(h; \theta)$, where $C(h; \theta)$ is a history-dependent reconstruction-cost landscape. That assumption makes "pressure" precise: nearby configurations experience a directional change. Other hypotheses could alter transition rates without admitting such a potential. The metric defining proximity, the historical weighting, and the coupling to the apparatus must be specified before results are interpreted. A landscape adjusted afterward to explain any observed drift would explain nothing.

History dependence in an incompletely observed system is common. Non-Markovian quantum processes, for example, can retain correlations through their environment. The proposed $h$ must contribute beyond that baseline. If the apparent effect disappears when an omitted material state or environmental variable is included, we have improved the ordinary model. That remains useful, but it is a different discovery from an additional historical reconstruction bias.

If such a coupling existed, then normal living brains near a critical regime might also sometimes recruit process structure associated with their past states, with another brain's state, or with nonbiological traces, into its own reconstruction. If these sampling pressures tip the thalamocortical or other loops from near criticality down into an attractor loud enough to reach awareness, then this experience might be interpreted as spiritual presences, voices, or an afterlife.

Maintaining the structure, exposing the uncertainty

The proposed physical stage would start from the digital stage's constrained family. It would preserve the supported organization while permitting unresolved coordinates to vary. In a local example, two small changes to a connection pattern may explain the known artifacts equally well. Prepare an ensemble around that ambiguity, let its physical realization evolve, and measure whether its outcomes favor one direction beyond the apparatus's calibrated preference. Repeat around the newly constrained region.

Analog quantum simulation is already an experimental field. Programmable arrays of interacting atoms have implemented tunable many-body dynamics; the 256-atom Rydberg simulator is one concrete demonstration. That establishes the existence of controllable physical quantum simulators, not a ready-made route to an embodied human reconstruction. The gap concerns encoding, interaction topology, precision, noise, feedback, scale, and the duration over which the relevant behavior can be maintained. I am optimistic about pursuing that gap, without assigning it a dependable completion date.

The attraction of an analog implementation here is specific. A hypothetical weak drift might be erased by repeated projection onto prescribed digital values or by correction that treats every departure from the intended computation as error. We would instead maintain the constraints needed for the brain-shaped organization while leaving selected physical coordinates free to respond. "Passing through unchanged" means withholding an imposed correction in those coordinates; it does not mean applying a mathematical identity to the complete physical state while somehow expecting that state to change.

The model's forward dynamics still operate. Sensory inputs arrive, internal activity evolves, actions affect a body, and the world returns consequences. Persistent parameters could also be physically adjustable rather than stored as constants rewritten identically on every update. The experiment must distinguish ordinary plasticity within the represented brain from any anomalous change of its realization. Both can alter behavior, but only the latter, if historically informative and unexplained by the baseline, supplies the proposed additional channel.

Analog does not mean free of noise, measurement back-action, or finite precision. Nor does quantum mean that useful drift automatically appears. Continuous-time quantum simulators may themselves use discrete local degrees of freedom. The operative design question is which candidate coordinates remain physically responsive, how deviations accumulate, and what observations can distinguish them. An analog classical implementation belongs among the comparisons; quantum hardware earns its role if the proposed coupling or an observed effect actually requires it.

Encoding is therefore central. A brain represented by a program is not automatically a brain-shaped interaction network at the physical level relevant to a new coupling. A software description of tract geometry does not establish the same electromagnetic geometry in the host apparatus. The hypothesis must state which relationships matter: effective causal organization, spatial arrangement, temporal structure, material properties, or some combination. Comparing equivalent model dynamics under different physical encodings would help determine this. The representation cannot be chosen only because it makes the desired result sound possible.

The baseline here is an entirely analog realization of the coupled brain, body, and world dynamics. Some analog couplings may be held fixed while others are learned; this changes which parameters we search, while the rest of the system continues to evolve physically. Its size is set by the causal distinctions it must preserve, with anatomical resolution supplying constraints rather than a direct count of required physical coordinates.

Three clocks also need separating: physical exposure time, the number of operations performed, and the duration of represented brain experience. Simulating a year in an hour does not establish a year of exposure to an unknown physical process. Changing those ratios while keeping other conditions matched would test whether a putative bias tracks apparatus time, computational activity, or some defined property of the represented trajectory.

From a weak preference to a particular person

Two independently prepared candidate brains are enough to define the first apparatus; larger ensembles can follow. Repeat the preparation and measurement across successive trials, using the evidence model to update the generative distribution. Repeated sampling estimates a distribution; it does not, by itself, make that distribution contract. Concentration would come from an identified physical drift, evidence-based selection, or both. The historical component must be estimated relative to a null that includes the same selection loop, since an optimizer can manufacture convergence without discovering anything about the past.

If $r$ is reproducible and discriminative, the two regimes become one process. Digital fitting provides the initial constraints. Physical outcomes provide additional likelihood terms. The generative model proposes the next candidates, preserving uncertainty where the experiment has not reduced it. Controls and tests on known targets calibrate what the physical measurements mean before they are used to make claims about a person whose internal state is unavailable.

Weakness creates a serious metrological problem. For independent trials with a small probability shift $\delta$, resolving that shift generally requires a sample count scaling as $\delta^{-2}$ at fixed significance and power. Drift, correlated noise, and multiple comparisons can make the requirement worse. Analog continuity may help under a particular accumulation mechanism, but it does not remove this burden. A proposal requiring $10^{10}$ or $10^{20}$ iterations must say what an iteration is and why the signal survives over that interval.

I do not think the individual should be treated as a faint residual beneath the collective human pattern. We routinely recognize someone through a tiny, contextually precise fragment of what they do. Recognition does not recover their complete brain, but it motivates the thought that person-specific causal organization has strong identifying signatures. Shared human structure narrows the search; the conjunction of memories, attachments, interpretations, and habits can distinguish who, within that structure, we are looking for.

This is what I mean by progressively pinning down fingerprints. Expand the candidate family where a cue leaves alternatives open; contract it where several independent relationships point to the same organization. Bring the remaining uncertainty into contexts where those alternatives diverge. Under the hypothesis, the physical sampler could then reveal a preference specific to that history. Recovering an individual's brain would mean converging on a coherent organization that predicts further individual-brain-specific structure, including withheld traces; simply making the ensemble more generically human would miss the proposed signal.

The instrument would also alter the histories it is testing. If repeated instantiation itself strengthens a candidate, the search could reinforce its own early errors. Exposure budgets, candidate ordering, fresh target families, and independent replication would therefore be part of the experiment. Otherwise "the universe prefers this brain" might only mean "we have repeatedly instantiated this brain." That possibility follows from the hypothesis and has to be included in its likelihood model.

A distinctive target may be necessary

The smallest apparatus need not have the simplest target. A recurrent motif or elementary transition structure may have been instantiated innumerable times in the history of the universe. If historical influence accumulates over similar instances, such a probe could register only the aggregate background of familiar process structure. Adding a few laboratory instances might barely alter it. Simplicity would make the apparatus easier to characterize while removing the contrast this particular hypothesis needs.

That is the reason for considering a human identity early. Its components are familiar, but their historically anchored conjunction may be rare enough to distinguish from the background. On this account we may have to make a leap to a richly structured target before there is a discriminative signal to measure. Small systems remain useful for calibrating noise, drift, and ordinary memory in the apparatus; success on them is not a prerequisite for this version of the conjecture. The required specificity must nevertheless be stated in advance, rather than increased after each null result.

A first serious benchmark could use consenting living participants, whose identities have the requisite history and whose withheld responses can still be checked. Fit a structurally aligned candidate family to one subset of each participant's artifacts, then freeze the evidence available to the reconstruction. An independent team retains structured observations—associations, interpretations, responses to reminders—and scores predictions committed before those observations are revealed. Compare the physical sampler with a matched digital reconstruction and physical controls receiving the same evidence, selection rules, and computational budget. Person-specific predictive improvement is the target; a persuasive impersonation is insufficient.

The critical comparison is whether unexplained drift follows the target's contextual organization across changes of apparatus and encoding. Swap participant–context assignments, vary similarity within the evidence-compatible family, cross hardware assignments, and keep operators and analysis blind. Separate historical preparation from reconstruction so that data leakage, retained material, ordinary signaling, and operator knowledge cannot carry the withheld relationships into the instrument. An advantage over a digital model alone could reflect a better sampler or a misspecified baseline; evidence for the proposed channel requires the predicted dependence on history as well.

Imagine two independently prepared forks of a reconstruction of my own brain. After isolation, one acquires a distinct experience: encounters lead to actions, actions to consequences, and the consequences alter its memories and habits. Randomization assigns the experiences and circumstances; the brain generates its own coherent trajectory through them. Uniform randomness alone would not establish the contextually anchored signature I want to probe. The resulting causal history is the signature we would probe.

The receiver, without access to that experience, is given a common partial cue that leaves several continuations compatible with its existing evidence. We then measure whether its sampling shifts toward relationships newly instantiated in the source, beyond the matched controls above. Reversing assignments across distinct trajectories separates a source-specific response from shared ancestry or a generic preference. If sufficiently similar instantiations exert the proposed pressure, two analog quantum reconstructions should be able to influence each other. Varying exposure, distance, and whether the source remains active would test the coupling's accumulation, range, and persistence.

The strength could be reported as a change in transition probabilities or log-odds with source exposure, and as information gained about the privately assigned trajectory. In a balanced, symmetric two-alternative test with matching probability $q$, the information per trial is $1 - h_2(q)$ bits, where $h_2$ is binary entropy; the two labels summarize rich trajectories. Similarity and scoring must be fixed before measurement. Mutual reinforcement could also amplify errors. A persistent effect after the source ceased operating would be a further condition to investigate.

Null results constrain the tested coupling and sensitivity range. Positive results do not immediately identify morphic resonance, an external simulator, consciousness, or resurrection. The unusual claim becomes scientifically productive when its conditions let it lose: a predicted dependence on controlled history that is absent at the specified scale counts against that version. Allowing the necessary similarity or hidden exposure history to change after every failure would remove the very feature that makes the proposal worth investigating.

What would it mean to answer again?

There are three separable achievements here: reconstructing organization from surviving evidence; discovering an additional physical channel that makes historical organization accessible; and establishing what renewed organization means for personal continuity. The first is an inverse problem with substantial existing tools. The second is a speculative physical hypothesis requiring its own evidence. The third remains a question even if the other two succeed.

None of this requires first explaining the origin of the universe, assigning agency to quantum collapse, or treating consciousness as the selector of physical outcomes. A history-dependent influence could be investigated wherever an operational version predicts it. Its relevance to minds would arise from the kinds of structured, path-dependent systems minds are—and from the particular information a reconstruction would need.

My brother's death is why this question is personal. It is also why a familiar voice would not settle it for me. A system can learn to say what I want to hear. What I want to understand is whether the organization that once encountered the world as Isaac could become accessible again, and what evidence could distinguish that recovery from a convincing construction shaped by the people who miss him.

The disappearance of the original substrate does not, by itself, answer every question about reconstructing a process. Neither does the persistence of someone's influence prove that their internal organization remains recoverable. The useful work lies between those statements: specifying what constitutes the person, measuring what the world still constrains, and asking whether any additional channel exists through which uncertainty can genuinely decrease.

I do not know whether a faithful reconstruction would resume the same subject, begin a new subject with the same organization, or expose a distinction our current vocabulary handles badly. I also do not know whether historical reconstruction bias exists. The reason to formulate the possibility is to make those unknowns sharper: a particular model family, a physical realization, an observable preference, and an experiment capable of separating discovery from expectation.

A prototype with a price and a clock

The build target is two independently prepared candidate brains, each on a complete $500 host: $1,000 for the pair. The architecture need not be Phaser. That price is an engineering target, including control and readout, conditional on sustaining the required causal and quantum dynamics. If one candidate needs $m$ hosts, the pair costs $1,000 × $m$; their interconnections must preserve those dynamics.

At a declared reconstruction tolerance $\varepsilon$, intervention set, and time horizon, use a finite partition $Z_\varepsilon$ of candidate causal organizations. Let

$$b = H(Z_\varepsilon \mid e, m_h)$$

denote uncertainty over those classes after conditioning on surviving evidence $e$ and the shared human model $m_h$. This is unresolved person-specific structure at that granularity. Naming a person, recognizing a fingerprint, and reconstructing their causal organization require different information.

We have no measured value of $b$. The cases 100, 1,000, and 10,000 bits below are sensitivity scenarios, not estimates of human identity. Estimate $b$ using calibrated candidate ensembles at several granularities and concealed observations from known people. Contextual fingerprints may eliminate alternatives quickly; other distinctions may require a specific memory cue or new experience. Count newly resolved uncertainty, not repeated answers to the same question.

A 50% reference probability belongs to one experimental design, not to the universe. Randomly assign the source one of two balanced, distinct experiential trajectories, and score the receiver's response without revealing the assignment. If the resulting channel is symmetric, a correct-match probability $q$ carries $i = 1 - h_2(q)$ bits per comparison. For asymmetric or continuous responses, estimate the conditional information from the full response distributions and matched controls. A favored direction in the apparatus alone is not information about the source.

Let $J$ be the pair's useful information rate in bits per physical second, after accounting for preparation, readout, correlation, and information already learned. In the symmetric reference case:

$$J = r\,\big[\,1 - h_2(q)\,\big]$$

where $r$ is the effective comparison rate of the complete two-host apparatus. Two hosts do not automatically give two independent channels: in the source–receiver experiment one supplies the history and the other probes it. The one-times discovery model is $t \approx b / J$. This is an ideal information budget for a calibrated, usable channel; a complete recovery protocol must also demonstrate its stopping accuracy.

Separate purchase, service life, and discovery. For a host price $c$, productive service life $L$ in hours, power $w$ in kilowatts, and electricity price $p$ per kilowatt-hour, the pair costs $2c/L + 2wp$ per hour of use. Assume $c$ = $500, $L$ = 43,800 hours (five productive years), $w$ = 0.01 kilowatts, and $p$ = $0.15/kWh. That gives $0.0258 per pair-hour, or $0.0129 per hosted brain-hour. This one-times accounting includes hardware and electricity only.

Five years is an assumed equipment service life. The uninterrupted lifetime of the represented brain state must be measured separately: resetting a working device can destroy its state. The host must preserve the required organization through discovery and subsequent operation; transfers and recovery across hardware failures need their own demonstrations.

For a concrete performance target, suppose the pair sustains 100 effective independent comparisons per second and a symmetric match probability of 51%. This yields approximately 104 new bits per hour. The resulting one-times scenarios are:

The $1,000 purchase is upfront; the table allocates its use across the assumed service life. Reusing the hosts adds electricity at $0.003 per pair-hour: about $0.29 for the 10,000-bit case. This prices the prototype's physical discovery stage; digital fitting and evidence acquisition are outside this model.

The clock is sensitive to the measured channel. At 50.1% matching instead of 51%, with the same effective comparison rate, every discovery interval above becomes approximately 100 times longer. At one comparison per second instead of 100, the same multiplication applies. Faster represented brain experience helps only if it increases $J$. Finite reliability requirements, changing signal strength, and failed preparations must be captured in the measured performance rather than concealed inside an unexplained multiplier.

match q

51.00%

comparisons / s1101001000unresolved bits1001,00010,000

This makes the immediate objective concrete: build two $500 hosts that sustain the chosen human causal structure, establish a distinct source history, and measure whether the receiver acquires information its supplied evidence did not contain. Then measure how that information rate changes with specificity, exposure, and whether the source is still running. Contemporary coupling would be the first result; access to a ceased source is the additional result historical reconstruction needs.

A thousand dollars is the proposed hardware target for that experiment, not an established remaining price of resurrection. But if a pair can recover roughly 100 validated person-specific bits per hour, then a reconstruction missing 1,000 such bits has a reference discovery time of roughly ten hours. Those are quantities we can put on a bench: capacity, lifetime, bits recovered, hours elapsed. Build the host. Isolate the history. Measure the pull. Let the next reduction in uncertainty come from the machine.

Perhaps an echo remains only an echo for as long as it can do nothing except repeat the past.

The interesting moment is when it can hear the world again—and answer.

Footnotes

1. This is also the direction of my current work on structurally aligned brain and body models: IBM-1 and IHM-1. ↩