Jesse Mazer writes:

[quoting Stathis Papaioannou]
Recent theory based on the work of Eric Kandel is that long term memory is mediated by new protein synthesis in synapses, which modulates the responsiveness of the synapse to neurotransmitter release; that is, it isn't just the "wiring diagram" that characterises a memory, but also the unique properties of each individual "connection".

But these unique properties would be included in the simulation if you could record the levels of different neurotransmitters and proteins at each synapse, no? Wouldn't they still be present in the brain of a person that had recently died, at least if the individual neurons had not yet died at the moment the brain was frozen?


In theory, it should be possible to scan a brain in vivo using some near-future MRI analogue and determine the state of each of the 10^11 neurons, and store the information as a binary srtring on a hard disk. Once we had this data, what would we do with it? The details of ionic gradients, type, number and conformation of cellular proteins, anatomy and type of synaptic connections, etc. etc. etc., would be needed for each neuron

I don't think you'd need the details of ionic gradients at the moment of death, as long as you know how ionic gradients work in a generic neuron you'd probably just be able to "restart" them.

The ionic gradients set the membrane resting potential, and it is the movement of ions through voltage-dependent ion channels that creates the action potential when a neuron "fires". The gradients are actively maintained by various energy-requiring transmembrane proteins, so it might be possible to reconstruct what they would have been at the moment of death if you could simulate the activity of these proteins.

along with an accurate model of how they all
worked and interacted, in order to calculate the next state, and the state after that, and so on. This would be difficult enough to do if each neuron were considered in isolation, but in fact, there may be hundreds of synaptic connections between neurons, and the activity of each connected neuron needs to be taken into account, along with the activity of each of the hundreds of neurons connected to each of *those* neurons, and so on.

Yes, but the hope is that neurons are not overly idiosyncratic in how they behave, that once you've learned to simulate a single neuron, then you could apply the same rules to other neurons as long as you knew their shape (branching of synapses, length of axon and so forth), the levels of various significant molecules at each synapse (and maybe elsewhere in the neuron), and maybe some other parameters I haven't thought of. And if you have a general algorithm that can simulate any neuron once you've measured the relevant parameters, then simulating an entire brain would mainly be a matter of being able to record these parameters in every neuron of a frozen brain, and of having a big enough computer (although you'd also have to know how to simulate the way that sensory neurons are stimulated by sense organs, how motor neurons affect simulated muscles, and how neurons interact with organs to affect things like hormone levels).

I agree that if you knew enough about the brain, it should be possible to emulate it. My criticism of "mind uploading" was based on web pages like this one:

in which it is suggested that anatomical knowledge of the neurons and their connections in a frozen brain would be enough to create the upload (as an afterthought, it is stated that knowing all the other details about each neuron would also be helpful). I should probably apologise to Eugen Leitl for the contemptuous tone of my earlier posts, because it is clear that he *does* appreciate the complexity of the neuron. Nevertheless, I still think it would be *extremely* difficult to emulate a whole brain. Just about every physical parameter for each neuron would be relevant, down to the atomic level. If any of these parameters are slightly off, or if the mathematical model is slightly off, the behaviour of a single neuron may seem to be unaffected, but the error will be amplified enormously by the cascade as one neuron triggers another.

--Stathis Papaioannou

SEEK: Over 80,000 jobs across all industries at Australia's #1 job site.

Reply via email to