If We Simulated Every Neuron, Would We Have a Brain?

Artificial neural networks encourage a particular intuition.

A biological neuron receives signals, transforms them and sends outputs.

An artificial neuron does something mathematically analogous.

Scale this up:

neuron → network → brain.

That analogy is useful.

It is also dangerously incomplete.

A modern artificial neural-network "neuron" is generally a simple numerical operation.

A biological neuron is an electrochemical cell.

Its behavior can depend on:

• branching dendritic geometry,
• voltage-sensitive channels,
• synapse location,
• calcium concentration,
• receptor composition,
• recent firing history,
• biochemical cascades,
• surrounding extracellular chemistry,
• and interactions with neighboring non-neuronal cells.

Modern brain-emulation research increasingly uses multi-compartment neuron models, multiple synapse types and biochemical processes precisely because a single scalar activation does not capture much of this biology.

Sources [2]

Another simplification is to imagine the human brain as a neuronal machine surrounded by biological support material.

That is no longer an adequate description.

Glial cells—including astrocytes, oligodendrocytes and microglia—participate in:

• metabolic support,
• synaptic regulation,
• neurotransmitter clearance,
• myelination,
• immune signaling,
• development,
• and modulation of neural circuits.

Modern experimental systems explicitly study human glial cells because their behavior can alter neural development and function.

Sources [3]

This does not prove that every glial biochemical event must be digitally simulated.

It does show why:

“Copy the neuron wiring diagram and you're done”

is scientifically unjustified.

Suppose we somehow scanned a brain perfectly.

Would the scan contain the person?

This depends on the physical storage of long-term memory.

The 2025 neuroscientist survey is especially revealing because it asked researchers directly what they thought memory depends on.

Most agreed that connectivity and synaptic strength are central.

But there was no agreement about exactly what additional neurophysiological features matter.

Sources [1]

When researchers were asked whether a static preserved brain could contain enough information to construct an emulation reproducing long-term-memory behavior, the median probability estimate was about 40%.

When detailed dynamic recordings of that same brain were hypothetically added, the median rose to 62%.

This shows that the scientific uncertainty is not merely computational.

It is informational.

We do not yet know exactly what would need to be measured.

There are at least four different claims that people often collapse together.

Claim 1: Same outputs

A simulation responds to stimuli approximately as the biological person would.

This is the weakest criterion.

Current AI can already imitate aspects of individual human behavior without remotely reproducing the person's brain.

Claim 2: Same internal functional organization

The simulation reproduces the causal relationships among brain processes.

This is much stronger.

Claim 3: Same memories and personality

The emulation remembers the person's childhood, relationships, beliefs and skills.

This requires preserving individual-specific information.

Claim 4: Same conscious person

The simulation is literally the continuation of the original person's subjective experience.

This is a philosophical claim that neuroscience currently cannot establish.

Even a perfect behavioral copy would not by itself resolve Claim 4.

The strongest argument for digital brain equivalence rests on substrate independence.

If cognition depends on causal organization rather than the particular atoms implementing that organization, then the same computational process could in principle exist in another physical medium.

The analogy is software.

A chess program remains the same algorithm whether it runs on silicon, an emulator, or another computer architecture.

If the brain is fundamentally an information-processing dynamical system, perhaps the same principle applies.

INFERENCE

Nothing in established physics currently demonstrates that carbon-based neurons possess a unique property that silicon or another substrate could never reproduce.

But this is not equivalent to proof that cognition is substrate-independent.

We have only one confirmed class of systems displaying human consciousness:

living human brains.

What if consciousness or memory depends on molecular details below the level we choose to simulate?

For example:

• protein phosphorylation,
• receptor trafficking,
• gene regulation,
• intracellular signaling,
• local protein synthesis,
• epigenetic state.

Then a neuron-and-synapse simulation could reproduce normal behavior for minutes or hours yet drift away over longer periods because it lacks the mechanisms responsible for learning and adaptation.

This is why whole-brain emulation cannot simply be treated as a scaling problem.

We first need to discover the minimum sufficient model.

Probably not.

Almost nobody thinks a useful brain model must explicitly simulate every proton and electron.

The 2025 neuroscientist survey found broad skepticism that atom-level molecular composition would be required, while many respondents believed subcellular structural information would matter.

Sources [1]

That suggests a hierarchy:

too crude: neuron identities and simple connections only

possibly sufficient: detailed cell morphology + synapses + receptor/cellular state + critical biochemical variables

probably unnecessary: explicit quantum simulation of every atom

But the scientifically important word is possibly.

The boundary has not been located.

Brains do not operate in vats.

They continuously interact with:

• eyes,
• ears,
• muscles,
• skin,
• vestibular organs,
• internal organs,
• endocrine systems,
• immune systems,
• and autonomic physiology.

The body changes the brain.

The brain changes the body.

Hunger, fatigue, inflammation, pain, hormones and cardiovascular state all alter cognition.

So if we copy a brain into a computer and give it no body, we have already changed its causal environment.

Does that matter?

Almost certainly for behavior.

Does it prevent consciousness?

Unknown.

A virtual body with sufficiently realistic sensory and physiological feedback might substitute for a biological one.

But again, this is a hypothesis—not an established result.

Large language models are sometimes discussed as if they represent an alternative route to human brains.

They do.

But they are doing something fundamentally different.

An LLM is trained to develop statistical representations useful for predicting and generating sequences.

It does not attempt to reconstruct:

• cortical anatomy,
• individual biological neurons,
• a human connectome,
• biochemical plasticity,
• or a particular person's brain.

Current AI therefore tells us something important:

intelligence does not require copying human neurobiology exactly.

But it tells us almost nothing about whether a digital copy of a specific human brain could preserve that person's identity.

Those are separate questions.

Suppose we scan you without destroying the original.

The computer copy wakes up.

Now there are two systems.

Both remember being you.

Both insist they are the continuation of your consciousness.

They cannot both be numerically identical to the single pre-scan person in the ordinary sense.

This reveals two different problems:

Engineering problem

Can we reproduce the cognitive organization?

Identity problem

Does reproducing it transfer the self?

Science may eventually answer the first without resolving the second.

No credible timetable exists.

The 2025 survey asked neuroscientists for their own estimates.

The median 50% estimates were approximately:

• C. elegans: 2045
• mouse: 2065
• human: 2125

Sources [1]

Those numbers should not be treated as forecasts.

They are surveys of expert intuition about a highly uncertain future technology.

What they demonstrate is that working neuroscientists generally regard human whole-brain emulation as a serious theoretical possibility but not a near-term engineering project.