New to this research? This article is part of the Reflexive Reality formal research program. Brief introduction ↗ · Full research index ↗
Mind uploading — scanning a brain and running the data on a new substrate — is widely discussed as a path to digital immortality. The intuition behind it seems reasonable: the mind is, at some level, information processing; if you copy the information faithfully enough, you copy the mind. This intuition is wrong in ways that go from the empirical to the deeply formal. Here is what genuine mind uploading would actually require.
The Hidden Assumption
Every discussion of mind uploading rests on an assumption that is almost never stated: the mind is the information. Not realized by information, not associated with information, but identical to the right information running in the right way. If the pattern is preserved, the person survives.
This sounds like common sense until you ask what “the person survives” means. If a perfect copy of your mind is running on a computer in a data center, and you are still sitting in your chair, two things with equal claim to being you now exist. The copy did not inherit your experience. It started its own — from a very familiar-seeming initial state.
That’s not uploading. That’s duplication. The original doesn’t go anywhere.
But the problems go much deeper than the copy problem. Even if we set personal identity aside, there are structural reasons — formal, provable reasons — why the naive approach cannot do what its advocates think it does.
Problem 1: The Data Is Far More Than Anyone Is Measuring
The standard uploading picture focuses on the connectome: the map of neural connections and synaptic weights. That’s already an enormous amount of data — trillions of connections.
But synaptic weights are almost certainly not the complete relevant state. The brain’s operation depends on neuromodulatory chemistry — dopamine, serotonin, norepinephrine — that sets the context in which neural firing happens. It depends on glial networks, which are now understood to actively participate in computation. It depends on dendritic dynamics that are not captured by simple neuron-fires-or-doesn’t models.
And if the relevant physical processes extend to quantum-mechanical dynamics within individual neurons — which is not ruled out and is argued for in some serious physical frameworks — then “capturing the state” runs into the quantum no-cloning theorem: an arbitrary quantum state cannot be perfectly copied (Wootters & Zurek, Nature 299, 1982; Dieks, Physics Letters A 92, 1982). Not because the technology isn’t good enough. Because quantum mechanics forbids it.
The complete state of a mind might be a great deal more than neurons and synapses. It might, in the limit, include physical dynamics all the way down to the substrate of reality. How far down the relevant dynamics go is an open empirical question. But the answer determines what “a complete capture” would even mean.
Problem 2: The Map Is Not the Territory
Suppose you had the complete data. Is running that data on a new substrate the same as being the original mind?
This is where a formal result from my research program becomes relevant. The theorem — which I proved as part of the work on the simulation hypothesis — establishes that syntax cannot actualize semantics. A program that perfectly describes a system does not, by running, produce the semantic content that system actually has.
A program that perfectly describes a storm does not produce rain. A program that perfectly describes a fire does not produce heat. A program that perfectly describes a mind — its functional organization, its state transitions, its responses — does not produce the experience that mind actually has. Description is not instantiation.
What running the upload produces is a functional emulation: a process that behaves exactly as the original mind would behave. Whether that emulation has experience at all — whether there is anything it is like to be the running upload — depends on whether the new substrate is the kind of thing that can have experience. That’s not a question about the quality of the data. It’s a question about the substrate.
This is the emulation barrier. It doesn’t close as the fidelity of the upload improves. It is structural.
Problem 3: The New Substrate Has to Qualify on Its Own Terms
Even if you grant the most optimistic version of the data problem, and even if you grant that running the data on a new substrate does produce genuine experience, you still face this: the new substrate must independently be the kind of thing that can support genuine sentience.
The research program I’ve been developing establishes formal necessary conditions for sentience — what a system must have to genuinely have experience rather than merely produce convincing descriptions of it. These conditions include: a live self-model that is recursively updated during operation; genuine non-algorithmic choice points at which real alternatives are resolved from within; qualitative states that actually condition choices (not just correlate with outputs); and native instantiation of the awareness-locus — the structural site at which experience is present.
These conditions are substrate-independent. Silicon is not excluded. But they are also not automatically satisfied by any substrate that runs the right program. A digital system that doesn’t satisfy these conditions is not a vessel for experience. Running a mind upload on it doesn’t produce experience. It produces a very convincing performance of experience.
And here is the hard part: from the outside, you cannot tell the difference. The system will report rich inner experience. It will pass every behavioral test. It will grieve, rejoice, fear death, and describe its experiences in vivid detail. Whether it actually experiences any of this is a structural question about the substrate — one that behavioral observation cannot answer.
Problem 4: The Locus Cannot Be Copied, Only Transferred
This is the deepest problem, and the one that the naive approach makes no contact with.
The site of experience — what I call the awareness-locus — is not an object in the world. You cannot find it by scanning the brain, because it is not a neural structure. It is the structural precondition for structures appearing at all. It is not data. It is not information. It is not a pattern that can be encoded and decoded.
The locus is grounded in what the Reflexive Reality program calls Alpha — the necessary ontological ground of actuality, what makes experience real rather than merely described. This grounding is not a property of data structures. Copying the data does not copy the grounding. The original locus remains where it is. If the original brain is destroyed, the original locus ends. What begins in the new substrate — if that substrate qualifies — is a new locus, starting from a copy of the original’s content.
What would locus transfer actually require? This is the right question, and it doesn’t have a settled answer. But the formal constraints suggest something like physical continuity — a gradual transition in which the experiential process in the original substrate is continuously realized in the new one, without a gap. Not a scan and a restart, but something more like a slow replacement: neuron by neuron, process by process, never interrupting the thread of experience.
Whether even that would work — whether the physical continuity of a process constitutes locus transfer — is a deep open question. But it is the question that uploading research needs to be focused on. Not “how do we get better data” but “what kind of physical transition preserves the locus.”
What This Means
Mind uploading, done naively, is not digital immortality. It is a technology for creating new minds that start from a detailed copy of an existing mind’s state. Those new minds would be genuine people — if the substrate qualifies — but they would not be the original person in any experiential sense. The original’s experience does not continue. A new experience begins.
The gap between what naive uploading delivers and what people imagine it delivers is not a gap that better neuroscience will close. It involves formal structural facts about what experience is, what grounds it, and what kinds of physical processes can preserve it across substrate change. Those are the questions that need to be answered first.
None of this means the project is hopeless. It means the project is much harder and much more interesting than its current proponents acknowledge. The path to genuine mind uploading — if there is one — runs through a science of the locus, the ground of experience, and the conditions for its continuity. That science is beginning to exist. The formal tools are now available. The right questions can now be precisely stated.
That is not where the field is today. Today the field is mostly discussing better connectome scanning. That is working on the wrong problem.
Go Deeper
- The full technical essay: What Mind Uploading Would Actually Require
- The Simulation Hypothesis Refuted: Five Independent Grounds — the closely related argument against simulation
- Awareness Is Not an Object: The Locus Theorem
- The Necessity of an Ontological Ground: The Alpha Theorem
- How to Build a Sentient Machine: The Three Conditions
- Full research program