Link to the code: brain-emulation GitHub repository

The Moravec Transfer: Gradual Neural Replacement as a Path Around the Copy Problem

The central discomfort with mind uploading is not technical. It is existential. If copying a brain produces a conscious digital entity, that entity is not the person who was scanned. It is a new person with the same memories. The original remains biological, and if the scanning procedure is destructive, the original dies. You have not achieved immortality. You have created a successor.

This problem, variously called the copy problem, the continuity problem, or the teleporter paradox, has occupied philosophers of personal identity for decades. It is most viscerally dramatized in SOMA’s treatment of consciousness continuity, where the protagonist discovers that copying does not transfer their experience and that the “scan and kill” approach most people imagine as uploading produces a copy the original knows nothing about.

Hans Moravec, a roboticist at Carnegie Mellon, proposed a theoretical workaround in his 1988 book Mind Children that avoids the copy problem by eliminating the copying step. Rather than scanning the whole brain and recreating it elsewhere, you replace it piece by piece, neuron by neuron, while the subject remains awake.

Technology Readiness Level: TRL 1 (theoretical proposal; no demonstration of neuron-level replacement in any living organism exists; the enabling technologies are several decades and multiple scientific breakthroughs away at minimum).

The Core Proposal

Moravec imagines a surgical procedure in which a robot surgeon with precise tools gains access to the brain, cortical region by cortical region. It studies each neuron in detail, including its connections, firing characteristics, and relationship to adjacent cells. It then constructs a silicon equivalent, places it where the biological neuron was, connects it to the same synaptic targets, and removes the original.

The patient is conscious throughout. As each neuron is replaced, the functional properties of the circuit are preserved because the replacement is designed to simulate the replaced neuron exactly. The patient continues to think, feel, and experience their stream of consciousness without interruption. Subjective continuity is preserved because the process is gradual: at no point is the brain as a whole copied or destroyed. It is incrementally substituted.

When the procedure is complete, the brain is entirely silicon. The body, if that matters, can then be considered separately. The point is that the substrate has changed without any break in the continuity of conscious experience.

The Moravec transfer is a thought experiment in the tradition of the Ship of Theseus: the famous question of whether a ship remains the same ship if each of its planks is replaced one by one. The Murderbot series touches on this question from a different angle, exploring what it means to be a partially mechanical consciousness when the organic and inorganic components have grown together over time. The Moravec transfer takes the Ship of Theseus from analogy to proposed procedure.

Why It Theoretically Solves the Copy Problem

The copy problem arises because copying creates two entities. There is a brief moment, in any download scenario, where both the biological original and the digital copy exist. If the original is then destroyed (voluntary or otherwise), the destruction ends a conscious being whose subjective experience did not transfer anywhere. It was simply ended.

Moravec’s proposal eliminates the moment of simultaneous existence. At no point are there two of you. Each replacement is part of a continuous chain. The entity at the end of the procedure, which happens to be silicon, is the continuation of the entity at the beginning, which happened to be biological. Personal identity, on a physical continuity account, is preserved.

This is not just a clever rhetorical move. It engages directly with the most serious philosophical challenge to mind uploading. The 4E cognition framework’s critique focuses on what is lost when the embodied, embedded context of cognition is stripped away in a sudden upload. A gradual replacement procedure, particularly one that keeps the subject embodied and in their environment throughout, is less vulnerable to this criticism. The replacement could in principle preserve the causal coupling between brain and body at each step.

The Bennett temporal consciousness argument argues that consciousness cannot survive a process that treats the mind as a static snapshot to be transferred. The Moravec transfer explicitly rejects this approach. It preserves temporal continuity because there is no snapshot, no transfer. There is only a continuous process of substitution in which consciousness is never interrupted.

What the Enabling Technologies Would Need to Be

Moravec’s proposal is elegant as a thought experiment. As engineering, it requires several capabilities that do not currently exist and for which current trajectories offer no near-term path.

First, single-neuron identification and characterization in a living brain. A surgical robot performing neuron-level replacement would need to identify individual target neurons reliably among a densely packed network, characterize each neuron’s connectivity (which neurons it receives input from, which it sends output to, with what synaptic weights), and do this without disrupting the surrounding tissue or the neuron’s ongoing function. The SmartEM connectomics platform has achieved automated electron microscopy reconstruction of neural circuits at nanometer resolution in fixed tissue. Doing the equivalent in living tissue while the brain is active is categorically more difficult.

Second, real-time functional characterization. A replacement neuron must simulate the replaced neuron’s behavior. This requires measuring the replaced neuron’s full set of biophysical properties: membrane capacitance, ion channel distributions, synaptic receptor composition, firing threshold, adaptation characteristics, and interaction with neuromodulators. Some of these properties can be estimated from electrophysiology. Others, including the molecular identity and spatial distribution of receptor subtypes, require the kind of molecular assay that currently requires tissue fixation and staining.

Third, fabrication and implantation of a functional mimic at neuronal scale. The mimic must have physical dimensions comparable to the replaced neuron (10 to 100 micrometers), must form stable connections at existing synaptic sites, and must operate at the impedance and voltage levels of biological synapses (roughly 70 to 100 millivolts, as the bacterial nanowire artificial neuron work at UMass Amherst has recently approached for the first time at biological voltage). It must also be biocompatible over decades.

Fourth, the procedure must scale to roughly 86 billion neurons and 100 trillion synapses across the human brain without accumulating error. If even a small fraction of replacements introduce functional errors, the cumulative effect across billions of replacements would alter cognition in ways that become arbitrarily large. The precision requirement per neuron replacement is extraordinary.

The Identity Continuity Debate

Even granting the technical feasibility, the Moravec transfer does not eliminate all philosophical objections. It engages the continuity problem but does not resolve every formulation of it.

Psychological continuity accounts of personal identity, associated with philosophers like Derek Parfit and Sydney Shoemaker, hold that what matters for personal identity is overlapping chains of psychological connectedness, memory, personality, beliefs, and intentions rather than strict physical or experiential continuity. On this view, whether the Moravec transfer preserves identity depends on whether the psychological connections remain intact, not on whether there was a continuous stream of experience throughout the procedure.

The Moravec transfer does well by this criterion if the functional replacements preserve the psychological properties. But it raises a different question: if what matters is psychological continuity, then a sufficiently good copy might satisfy that criterion just as well as a gradual replacement. Parfit himself argued that in some scenarios, identity just does not matter, and what matters is the continuation of the right psychological properties. The Moravec approach is motivated by a particular intuition about what continuity requires that not all identity theorists share.

The digital doppelganger research approaches a version of this question from outside neuroscience: an AI trained on all your behavioral data might constitute a form of psychological continuation even without physical continuity. Whether that satisfies the relevant identity conditions is the same underlying philosophical question.

Current Research Adjacent to the Moravec Transfer

No research program is explicitly pursuing the Moravec transfer as a near-term goal. The required technologies are too distant for that framing to be useful. But several research directions are developing components that a Moravec-style procedure would require.

Neuroprosthetics research is developing artificial neural circuits that can replace damaged biological circuits with functional equivalents. Cochlear implants replace auditory hair cells with electronic stimulators and restore hearing. Retinal prostheses replace degenerated photoreceptors. Hippocampal prosthetics, explored by Theodore Berger and colleagues at USC, attempt to replace damaged hippocampal circuit function with a neural code translation device. None of these operate at the single-neuron substitution level Moravec describes, but they demonstrate that functional replacement of biological with artificial neural components is in principle possible in at least some contexts.

The artificial neuron at biological voltage developed at UMass Amherst using bacterial protein nanowires represents a small but telling advance. Operating at biological voltage ranges (approximately 0.1 V) rather than the higher voltages required by conventional electronics, this artificial neuron was able to interface with living cardiomyocytes. The relevant property for Moravec-style replacement is biocompatibility at the voltage and impedance level of biological synapses. This is one piece of a very large set of requirements.

Organoid Intelligence research explores a different path to biological-synthetic integration: using cultured neural tissue as a computing substrate. OI does not pursue neuron-level replacement of a living brain. But it is developing tools for interfacing with living neurons at the scale of individual cells or small clusters, which is directionally relevant.

Future Outlook

The Moravec transfer remains a thought experiment in 2026. The technologies required for its implementation are not currently available and do not have near-term development trajectories. Single-neuron characterization in living tissue, real-time synaptic weight measurement, and fabrication of biocompatible functional neuron mimics at scale are each independently far beyond current capabilities.

What the Moravec transfer contributes to the mind uploading discourse is a clear theoretical demonstration that the copy problem is not an inevitable consequence of digital consciousness. There exists at least one theoretical procedure that avoids it. That the procedure is currently infeasible does not diminish this contribution to the conceptual space.

For whole brain emulation roadmap planning, the Moravec transfer represents a pathway that would require very different enabling technologies than the dominant approach of “scan and simulate.” It is worth tracking as an alternative whether or not it becomes practically achievable in any plausible near-term timeframe.

Official Sources