Why Mind Uploading Remains Hypothetical

Mind uploading remains a hypothetical research idea, not a demonstrated technology. What has changed since early 2025 is the scale of the neural maps available to researchers: an Allen Institute account of the MICrONS project describes a cubic millimetre of mouse visual cortex containing more than 200,000 reconstructed cells, half a billion synapses and four kilometres of axons.

That progress makes the central obstacle easier to see. Researchers can connect anatomy with measured activity in a small brain region, but they still do not know which physical details would be sufficient to preserve a particular person’s memories, personality and subjective experience. A 2025 preregistered test of two consciousness theories, involving 256 participants and three neuroimaging methods, supported some predictions while substantially challenging important claims of both theories.

What would count as uploading a mind?

The term covers several outcomes that require very different evidence. Software might imitate someone’s speech, reconstruct biographical details from personal records, emulate the computations of a brain, or preserve the same conscious subject. Success at one level would not establish success at the others.

A model trained on messages, photographs and recordings could produce a persuasive conversational likeness. That would be a behavioural simulation derived from a person’s observable output, not evidence that memories had been extracted from the brain or that the person’s awareness continued in the model.

Whole-brain emulation is the narrower technical proposal. It assumes that the relevant organization and state of an individual brain can be measured, represented computationally and reproduced in another physical substrate. The resulting system would then need to preserve person-specific mental functions under tests that cannot be passed through superficial imitation alone.

Consciousness uploading makes an additional claim beyond emulation. Even if a digital model behaved like its biological original, researchers would still need grounds for concluding that the implemented processes generate subjective experience. Current neuroscience has no generally accepted test that can settle that question for an artificial system.

A connectome is not a stored personality

A connectome records neurons and their connections at a specified level of detail. It is indispensable for studying how signals may travel through a nervous system, but it is not automatically a record of memories, changing synaptic strengths or the complete state of every cell.

The clearest whole-brain example comes from a much smaller animal. The official FlyWire connectome page identifies an adult female fruit-fly brain with 139,255 proofread neurons; its summary rounds the resource to about 140,000 neurons and more than 50 million synapses. The reconstruction combines high-resolution electron microscopy, automated segmentation, expert proofreading, community annotations and inferred neurotransmitter information.

This is a major scientific resource, but its boundaries matter. A wiring diagram shows which elements appear connected; it does not by itself specify how strongly every synapse acts at a given moment, how molecular conditions alter each cell, or how learning has changed the network. Those dynamic properties may be essential to reproducing behaviour and memory.

MICrONS addresses part of that limitation by pairing reconstructed anatomy with functional observations from mouse visual cortex. It still covers a small volume rather than an entire mammalian brain, and measurements collected under selected experimental conditions cannot enumerate every state the tissue could enter.

A credible human emulation would therefore require a principled answer to a measurement problem: which details can be omitted without changing the properties the model is supposed to preserve? Recording more structure is useful, but resolution alone cannot determine whether the missing variables are irrelevant.

The missing steps between a scan and a digital person

Even an exceptionally accurate wiring diagram would be an input to an emulation, not the finished result. Several coupled problems would have to be solved before a digital system could make a defensible claim to preserve an individual mind:

Capture: measure the relevant cells, connections and biological states at adequate spatial and temporal resolution.
Reconstruction: detect and correct imaging or segmentation errors while identifying the parameters required by the computational model.
Dynamics: reproduce how neurons, synapses, supporting cells and chemical signals change over time.
Embodiment: determine which effects of sensory input, movement, hormones and internal organs must be represented.
Validation: test memories, skills and stable behavioural traits without defining success as resemblance alone.
Consciousness: provide independently testable reasons for attributing experience to the implemented processes.

These requirements cannot be solved independently. Without a sufficiently detailed model of how memory and cognition arise, researchers cannot know what a scan must retain. Without the necessary measurements, they cannot build a decisive test of the model.

Additional computing power can accelerate a specified simulation, but it cannot choose the correct biological specification. If an essential cellular or chemical process is absent from the model, running the incomplete model faster does not restore it.

Why AI and brain-computer interfaces are different

Generative AI can imitate writing patterns, search a personal archive and maintain a consistent fictional or biographical profile. Such systems could support digital memorials or software agents, but their primary input is a collection of external records rather than a reconstruction of the neural mechanisms that produced them.

Brain-computer interfaces solve another problem. They can translate selected neural activity into commands or deliver stimulation to nervous tissue, establishing a channel between a brain and a device. A limited communication channel does not capture the full system that generated its signals.

The distinction matters when products use labels such as “AI clone,” “digital twin,” “memory avatar” or “mind upload.” The useful questions are what information was captured, which mechanisms were reproduced and what evidence distinguishes preservation from imitation. Fluent conversation alone cannot answer any of them.

A successful copy would still leave an identity question

Suppose a future emulation reproduced a living person’s memories and behaviour with exceptional accuracy. That technical achievement would not by itself show that the biological person’s first-person perspective had moved into the new system. It could instead produce a second entity that remembers being the original.

A non-destructive procedure makes the problem especially clear. If the biological individual and digital counterpart existed simultaneously, they would immediately begin accumulating different experiences. They could share a past while representing two continuing trajectories.

This distinction would affect consent, ownership, deletion and responsibility. Legal and ethical rules would depend on whether the digital system was treated as software representing a person, a new person, or a continuation of an existing one. Behavioural similarity cannot settle that classification on its own.

The defensible status of mind uploading

Connectomics has produced genuine advances: researchers can reconstruct a complete adult fly brain at synaptic scale and can relate extensive anatomy to activity in part of a mouse cortex. Consciousness research is also testing competing theories through shared predictions rather than relying only on separate bodies of evidence.

Neither development constitutes the transfer of a human mind. Whole-brain emulation remains a research programme with improving prerequisite tools, while its required measurement fidelity, biological model, validation criteria and account of consciousness remain unsettled.

The most accurate description is therefore neither digital immortality nor a completed engineering roadmap. Until a system preserves person-specific mental functions under tests that rule out imitation—and until claims about artificial consciousness become independently testable—mind uploading should be treated as a speculative possibility rather than an available technology.

Mind Uploading Is Still Hypothetical—Bigger Brain Maps Show Why