We Can’t Upload You, Sorry

Why we can’t put your mind in the machine

Immortality awaits. As you draw your dying breath, we will inject a preservative into your brain that will fix in place every one of the trillion or so connections between your 86 billion neurons. We will then trace those wires, building the complete map of your brain’s connections, your “connectome”. Upload that complete wiring diagram to a computer, simulate the brain’s dynamics upon it — and you live again.

Such is the promise of mind uploading. It is predicated on a common idea: that the wiring between neurons is what stores your memories, is what makes you, well, you. Companies and nonprofits are working on making mind uploading a reality by exploring how we can best preserve the complete wiring diagram, every little synapse made between one neuron and another. Enthused media articles, journal papers, and controversial figures proselytise the coming age of restoring you from your brain wiring. As Sebastian Seung claimed: you are your connectome.

Sadly, it’s not true. Mind uploading from a preserved brain cannot ever be a thing. Not just because we don’t have the technology (which we really do not). But because it is literally impossible to rebuild you from wiring alone.

The reasons are not hard to grasp. They all come down to the same thing: a connectome is a static thing, a snapshot of the brain’s wiring at the instant it was captured. The brain is not a static thing. It is dynamic, from its billions of neurons talking to each other, through the constant rewiring of the connections between them, to the churning molecular machinery inside each neuron, controlling and changing every aspect of that neuron’s form and function.

There are many ways in which your brain is dynamic, but focussing on one of them is enough to illustrate my point. My favourite is how neurons talk to each other by sending pulses of electricity, “spikes’’ of voltage that sweep along their axons towards their targets (indeed, I wrote a book all about them and how they reveal your brain’s workings). Spikes being passed between neurons are how we do anything. Most prosaically, spikes from the motorneurons in your spine are what make your muscles contract: your fingers flex, your arm stretch; the swing of your leg and the tension in your shoulders. Beyond this, your brain is full of spikes for seeing, spikes for hearing, spikes for deciding, spikes for suddenly, uselessly remembering the name of that actor who was in the thing you watched yesterday.

What does the wiring between neurons tell us about the spikes they send to each other? Nothing at all. It only tells us where spikes would be sent, and only in that snapshot of the brain’s wiring at the time it was preserved.(Which also raises the question of why anyone would want to be reconstituted as they were the moment before their death, which in all likelihood was not a good time to be alive in the first place). Wiring does not tell us when and how a spike would be sent.

When and how often a neuron sends spikes depends on how it responds to its own inputs. To be fair, knowing the type of neuron can give us a rough idea of how it might respond. We know that in the cortex, a pyramidal neuron responds to its inputs differently than a fast-spiking interneuron; and that interneuron, in turn, responds differently to its inputs than does a projection neuron of the striatum.

But that is all knowing the type of a neuron tells us, just these qualitative differences in how it responds compared to other types of neuron. It tells us nothing about the specific neuron at each point in your wiring diagram. The dynamics of a specific single neuron that we pluck at random will depend on its specific shape and size, on where precisely fall the inputs to its dendrites, and on the precise combination of the different types of ion channels it expresses.

Point to point wiring, the connectome, gives us none of this, none of these properties of the neuron, and never can. We cannot tell from the connectome how a neuron will respond to its inputs. We cannot simulate the response of a specific neuron in your brain from knowing all of its wiring, and not even from knowing what type of neuron it is. Worse still, when a spike arrives at the connection, the synapse, from one neuron to the next, the effect of that spike is also invisible to the connectome.

There are two crucial questions the connectome cannot answer.

How strong is the connection? The stronger the connection from neuron A to neuron B, the bigger the influence a spike from neuron A has on when neuron B will make its own spike. This we cannot tell from wiring alone. Granted there are correlations between features of a synapse and its strength, like the size of the spine on which the synapse is made, or the number of neurotransmitter receptors found in the synapse. But these are not things knowable from point-to-point wiring — finding these requires reconstructing not just the connections but the nano-scale details of each synapse’s structure.

Doing such a reconstruction of 1 trillion synapses is, to put it mildly, unfeasible. And even if it were possible, it is only a correlation, and quite a weak one at that. The only way to know the specific strength of a specific synapse in a brain is to subject it to a barrage of recording experiments. The specific strength of a specific synapse is invisible to the connectome. Yet knowing the strength of each synapse is essential to accurately simulating a given piece of your brain.

The other question about synapses is harder still: How reliable are they? Some connections between neurons always work, every spike arriving at a synapse eliciting a response, however small, from the receiving neuron. Many connections fail at astonishing rates. Connections between neurons in the hippocampus can fail up to 90% of the time. That’s 90% of arriving spikes that create no response in the receiving neuron — that fail to do anything. The failure rate of a specific synapse is invisible to the connectome. Yet knowing how often a synapse fails is essential to accurately simulating a given piece of your brain.

And, of course, none of these things are static: neurons shrink and grow their dendrites; inputs to those dendrites arrive, go, and rewire to different points; the expression of ion channels changes; the strength and failure of a synapse fluctuate. This ever-changing brain is you — is you thinking and remembering and planning and doing all you do.

Trying to replicate your brain from wiring alone is like trying to replicate your entire social life by only looking at who you are in contact with, in this instant. Nothing about how you interact, about how often or how intensely you interact with someone, about whether that relationship is good, bad, indifferent, toxic, or merely someone at work you tolerate for their homemade cakes. And nothing about the history and dynamics of each relationship, how they started, developed, and fluctuated, changing the strength of your ties to every person you have ever had any connection with.

Even if we somehow knew all of the above — knew the exact shape, size, and ion channel make-up of every one of the 86 billion neurons in your brain, and the strengths and rates of failure in every one of your roughly 1 trillion synapses, it wouldn’t help. There’s a simple fact I missed off about spikes. When and how often a neuron sends its spikes depends most strongly on when it receives spikes from other neurons. And when they send spikes depends on when they receive spikes from their inputs. And so on, ad infinitum. To know you, we need to be able to simulate this infinite recursion of spikes; and to do that we’d need to recapitulate your entire development, from the moment your first spikes appeared.

Don’t get me wrong, there are good, solid scientific reasons to make connectomes: to learn about the principles of wiring between neurons, to learn about how widely the brains of the same species differ, to find out if there are changes in wiring that predict brain disorders from schizophrenia to Alzheimer’s disease. And so there are good, solid scientific reasons to work on ways to perfectly preserve brains, so these connectomes can be recovered. But connectomes are a static thing, a fixed thing, which you are not. So I’m sorry to break it to you, but it turns out we can’t upload your mind to a computer because you are not, in the slightest, defined by the wiring between your neurons. You are not your connectome.

Mark Humphries researches computational neuroscience at the University of Nottingham, UK, and is the author of “The Spike: An Epic Journey Through the Brain in 2.1 Seconds” (Princeton University Press) out now.

Twitter: @markdhumphries

Theorist and neuroscientist. Writing at the intersection of neurons, data science, and AI. Unceasingly British.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store