There’s an interesting conversation in the New York Times: a neuroscientist, Kenneth D. Miller, argues that brain uploading ain’t gonna happen. I agree with him, only in part because of the argument from complexity he gives.
Much of the current hope of reconstructing a functioning brain rests on connectomics: the ambition to construct a complete wiring diagram, or “connectome,” of all the synaptic connections between neurons in the mammalian brain. Unfortunately connectomics, while an important part of basic research, falls far short of the goal of reconstructing a mind, in two ways. First, we are far from constructing a connectome. The current best achievement was determining the connections in a tiny piece of brain tissue containing 1,700 synapses; the human brain has more than a hundred billion times that number of synapses. While progress is swift, no one has any realistic estimate of how long it will take to arrive at brain-size connectomes. (My wild guess: centuries.)
Second, even if this goal were achieved, it would be only a first step toward the goal of describing the brain sufficiently to capture a mind, which would mean understanding the brain’s detailed electrical activity. If neuron A makes a synaptic connection onto neuron B, we would need to know the strength of the electrical signal in neuron B that would be caused by each electrical event from neuron A. The connectome might give an average strength for each connection, but the actual strength varies over time. Over short times (thousandths of a second to tens of seconds), the strength is changed, often sharply, by each signal that A sends. Over longer times (minutes to years), both the overall strength and the patterns of short-term changes can alter more permanently as part of learning. The details of these variations differ from synapse to synapse. To describe this complex transmission of information by a single fixed strength would be like describing air traffic using only the average number of flights between each pair of airports.
Underlying this complex behavior is a complex structure: Each synapse is an enormously complicated molecular machine, one of the most complicated known in biology, made up of over 1,000 different proteins with multiple copies of each. Why does a synapse need to be so complex? We don’t know all of the things that synapses do, but beyond dynamically changing their signal strengths, synapses may also need to control how changeable they are: Our best current theories of how we store new memories without overwriting old ones suggest that each synapse needs to continually reintegrate its past experience (the patterns of activity in neuron A and neuron B) to determine how fixed or changeable it will be in response to the next new experience. Take away this synapse-by-synapse malleability, current theory suggests, and either our memories would quickly disappear or we would have great difficulty forming new ones. Without being able to characterize how each synapse would respond in real time to new inputs and modify itself in response to them, we cannot reconstruct the dynamic, learning, changing entity that is the mind.
That’s part of the problem: the brain is really, really complicated. That tiny scrap of brain tissue where they mapped out all the synapses? I wrote about that here; it was a tiny slice, 1500µm3, or a little dot about 12µm on a side…1/80th of a millimeter. It contained all of those synapses, took a huge effort (an effort that destroyed the tissue), and it recorded only a snapshot of cellular and subcellular structure. There was no information about those thousands of proteins, or the concentration of ions, or any of the stuff we’d need to know to reconstruct activity at a single synapse — all that was also destroyed by the chemical processing required to preserve the structure of the cell.
We aren’t even close to being able to take apart a brain at the level necessary. Miller is exactly right. And as he points out, one additional problem is that the brain isn’t static — it’s not going to hold still long enough for us to get a full snapshot.
But as I said, complexity is only part of the problem, and if you focus on just that issue, it opens you up to this kind of rebuttal from Gary Marcus.
Two hundred years ago, people thought that flying machines were more or less impossible; cellphones were inconceivable as real-world artifacts. Nobody knew what a genome was, and nobody could have imagined sequencing one for a thousand dollars.
Mr. Miller’s articulation of the complexity of the brain is reasonable, but his extrapolation proceeds without any regard whatsoever to the pace of technological progress. It is the ratio between the two — complexity and progress — that matters.
Brain uploads won’t be here tomorrow, but there is a very good chance that they will be here within a century or two at most. And there is no real argument to the contrary.
We’ve got a problem with lots and lots of parts, and it’s too complicated for us to even count the parts. But technology marches on, and we can expect that someday we’ll have widgets that can track and count far more parts than we can even imagine now. It doesn’t matter how many parts you postulate, that is a merely quantitative problem, and we’ve been really good at solving quantitative problems. Why, any day now we’ll figure out how to squeeze enough energy into a teeny-tiny box so that we can build jet-packs.
As for that genome argument, that is correct: we’re really good and getting better at sequencing a few billion nucleotides at a time. With a sufficiently simple definition of the constitution of the cell, you could claim that it’s a solved problem: we can figure out the arrangement of the letters A, T, C, and G in a linear polymer just fine. Now telling me how that gets translated into a cell…well, that’s a little more difficult. That’s a whole ‘nother problem we aren’t even close to solving in detail. It’s also not going to be solved by enumerating the bits.
Another problem here, beyond complexity, is specificity. My brain and your brain are equally complex, have about the same number of parts, and are arranged in roughly equivalent ways, but they differ in all the specifics, and it’s those specifics that matter. If you were to disintegrate my brain molecule by molecule so you could attempt to reconstruct it in a computer, it does me no good if you build your brain in the machine, or Jeffrey Dahmer’s brain, or a profoundly malfunctioning artifact with randomized cognitive connections, or a blank blob with a potential to learn. All the transhumanists want personal immortality by placing their personal, unique awareness in a box less likely to rot than our standard organic hardware. So not only do you have to build something immensely complicated, it’s got to be a nearly exact copy of something just as complicated.
And the bits in this copy are specified right down to the arrangement of individual molecules and even the concentration of ions in tiny compartments…all of which are changing constantly to generate the mind. You would have to freeze my brain in place instantaneously, capture the position and state of every molecule in it, and then build a copy with astonishing accuracy at the molecular level — all while the copy is locked down and not reacting in any way with it’s components — and then restart it all instantaneously as well. There are physical limits to how precisely individual molecules can be manipulated. This problem goes beyond building better mapping and inventory of a multitude of parts. It’s bumping up against limitations of the physical universe.
I agree with Marcus that someday we might be able to build something as complicated as a brain — we already do it, every time we make a baby. But making an exact copy of 1.5kg of insanely intricate meat, in a medium that isn’t meat, somehow? Nah, that’s not a realistic proposal.