Imagine: a mad scientist with a ray gun shoots at a neuron somewhere in cortical layer IV of your visual area MT, burning it up in a matter of microseconds (just for fun, imagine also that the ray gun leaves everything else intact).
With one neuron missing, you probably won’t notice any perceptual change. But what if, one by one, all neurons in are MT went AWOL? You’d be stuck with an annoying inability to visually detect motion.
Now imagine that for every cell that our fancy ray gun hits, it replaces it with a magical transistor equivalent. These magical transistors have wires in place of each and every dendrite, a processing core, and some wires in place of axon(s). Naturally, the computational core analyzes the sum of all inputs and instructs the axon to “fire” accordingly. Given any set of inputs to the dendrite wires, the output of the axon wires is indistinguishable from that of the deceased neuron.
We can still imagine that with one neuron replaced with one magical transistor, there wouldn’t be any perceptual change. But what happens when more and more cells are replaced with transistors? Does perception change? Will our subject become blind to motion, as if area MT weren’t there? Or will motion detection be just as good as with the real neurons? I am tempted to vote in favor of “No change [we can believe in],” but have to remain skeptical: there is simply no direct evidence for either stance.
Ray guns aside, it is not hard to see that a computational model of a brain circuit may be a candidate replacement of real brain parts (this is especially true considering the computational success of the Blue Brain Project’s cortical column, which comprises 10,000 neurons and many more connections among them). For example, we can imagine thousands of electrodes in place of inputs to area MT that connect to a computer model (instead of to MT neurons); the model’s outputs are then connected, via other electrodes, to the real MT’s outputs, and ta-da! Not so fast. This version of the upgrade doesn’t shed any more light on the problem than the first, but it does raise some questions: do the neurons in a circuit have to be connected in one specific way in order for the circuit to support perception? Or is it sufficient simply for the outputs of the substitute to match those of the real circuit, given any set of inputs? And, what if the whole brain were replaced with something that produced the same outputs (i.e. behavior) given a set of sensory inputs – would that “brain” still produce perception?
August 21, 2010
You said that in such a brain transplant, “Given any set of inputs to the dendrite wires, the output of the axon wires is indistinguishable from that of the deceased neuron.”
This is a bold statement, and assumes quite a bit. Firstly, in order to guarantee that the output of both the natural and artificial axons is identical, you would have to write a program that *exactly* mimics the behavior of real neurons. This may seem like a simple task, and some scientists might even say that it’s already been done. But think about it for a second: what program does a neuron follow? The answer is that it doesn’t follow any; the neuron is a physical and chemical unit, and is bound only by the rules that govern such things. A neuron doesn’t follow code in order to complete its task. One could potentially program a virtual environment in which the same “physical” laws governed all entities. But for that virtual environment to be complete, you’d have to build a virtual neuron, and everything else, from the atom up. Then, and only then, would you even come close to being able to simulate the input/output relationship that determines whether or not a neuron fires, and how much it fires.
Secondly, why is there this assumption that the physical nature of a neuron is the not the very thing which makes perception possible? True, neurons operate in a network through the exchange of electrical impulses, but the ways in which the impulses are transmitted and received are dependent on the physicality of the neurons. If you wanted to build an actual network of neuron-replacing transistors, it’s likely that anything short of artificially grown biological neurons would fail to do the job. I mean, what’s the alternative? Copper wire? Fiber optics?
And lastly, how exactly did the Blue Brain project experience “computational success”? It’s a good thing that the project intends to construct a virtual brain from the molecule up, (as mentioned above as necessary), and I have faith that they will ultimately achieve success, but they’ve only produced a simulation of 10,000 neurons out of a whopping 10^11 neurons in the entire human brain (that’s one ten-millionth). And the number of processors they’ve used to create this network? 8,000. I’m pretty sure that the volume of 8,000 processors far exceeds that of a real cortical column. Not to mention, if they were to use the same method to build the rest of the brain, they’d have to use 80 BILLION processors of the same strength. Success? Maybe, but we have to wait for some vast improvements in hardware and software technology before it will be possible to tell.
August 23, 2010
To answer your concerns:
1. a) No program has to be written to model a neuron, as a neuron may be represented with discrete electronics (resistor, capacitor and transistors). The Hodgkin-Huxley model of the neuronal action potential actual assumes that the cell is equivalent to a circuit where the membrane is a capacitor, the ion channels are resistors and the electrochemical gradient of ions is a battery. (see: http://en.wikipedia.org/wiki/Hodgkin–Huxley_model). Also of interest is BEAM robotics, which uses simple electronics, mimicking neurons, as the robotic “brain” (see: http://www.beam-wiki.org/wiki/Nu_Neurons). Transistors, like neurons, do not follow any program but operate on physical principles (flow of electrons).
b) It is correct that the neuron does not follow any program. But why does that mean that a model of a neuron (that does follow a program) can’t be a suitable replacement for a real neuron? It’s unlikely that satisfactory models would have to describe from the atom-up because details at such a fine level may be irrelevant to what the neuron does electrochemically.
2. There is no assumption that the physical nature of a neuron is not the thing that makes perception possible: we don’t know if it is or isn’t. But going with the hypothesis that the physicality is not essential gives us an experimental route. While assuming that a neuron’s physiology is essential to producing perception still gives us an experimental option to reproduce consciousness (that is, to make a synthetic neuron and brain out of the same stuff that real neurons and brains are made of), the former option would be more practical since more hypotheses could be tested in shorter periods of time than with a ‘wet’ version. Of bigger concern is the fact that models only describe what happens without actually doing it (I will venture to guess that this is your main concern); that concern may be alleviated if the model is hooked up to actuators an actual brain.
3. We certainly have to wait for improvements in computation to have anything close 100 billion neurons with 1,000 connections each (don’t forget the glia; they are even more numerous). But the point is that those 80 processors simulate the cortical column faithfully (the limit of which is computational power). And while it seems outrageous that so much power and space should be dedicated to something as small as a cortical column, just remember that early computers faced the same problems; your PC is many times more powerful and smaller than the first computers were. We can expect the same increase in computational power for the future.
The purpose of replacing neurons is clinical in nature. Say a lesion to a particular part of the brain can be replaced with a prosthesis. Neural prostheses are, of course, around already. Retinal implants and speech prosthetics (as by Frank Guenther) are good examples (www.bostonretinalimplant.org/ and http://speechlab.bu.edu/prosthetics.php.) But these are sensory prostheses and show only that the brain can interact with foreign inputs (limb prostheses show that the brain can interact with foreign effectors, or outputs). For prostheses of brain parts in general, one would have to prove that the brain can work with foreign integrators (i.e. something that takes a neural input, processes it, and sends a faithful output to appropriate neurons). I think this is possible in theory.
The question is whether we can expect one to be aware of what goes on in their neural prosthesis. For example, if one’s primary somatosensory cortex is replaced, would they be conscious of incoming tactile stimulation? (This is where the functionalists separate from those who think that biological matter is key to consciousness.) I would guess that one would have a normal percept of tactile stimulation, but only if all electrical connections (synapses etc) that the prosthesis is substituting are accounted for.
August 23, 2010
What about the fact that neurons are plastic? If one designs a fixed computational device, it would no longer be plastic. I see using stem cells as a more potentially beneficial clinical treatment. Also, I say that constructing a model without omniscence is likely to produce error out of ignorance. In other words, trying to understand how neurons work by building models of them seems to be counter-intuitive, since the study of them would be likely to exclude knowledge about the original absent from the imitation. From the lab work alone I did over the summer on LTP, we seem far from understanding completely things such as integration of signals on neurons, plastic functioing, and the electrochemical (with special empahsis on the chemical and organic) nature of neuronal interactions. Additonally, the interaction of neurons across specific regions in giving rise to functions like memory (regions of hippocampus interacting with one another and the cortical areas, etc.) goes beyond the already complex and obscure process of individual neuronal function. It would seem to me that a claim to perfect neuronal replication given the current imperfect state of knowledge would be presumptive.
August 23, 2010
But there’s no reason why the model should be static. Neuronal plasticity (and metaplasticity) are poorly understood but are nonetheless physical processes (same goes for signal integration and everything else the neuron does) and so should be featured in our model.
As for the presumptive nature of modeling neurons, I think that will be sure to make mistakes but will thus give more clues about the proper function of the system.
And stem cells are certainly more realistic and practical now…
March 1, 2011
Okay, there is something else to this that no-one seems to be considering.
The components cells of organs are designed to pull in resources in order to produce an output function. Each individual cell type has a different method of performing its’ function. Neurons are NO different.
Neurons, if severed, regrow to re-attach to severed nerve endings. There was a little girl named Cameron Mott who was suffering from violent seizures who underwent surgery to have the right side of her brain removed. After the surgery, she was able to use her left leg and had slight control of her left arm. With the right brain removed, she should have not had control of her right leg at all, but the neurons that were severed regrew and re-attached to their previous nerve endings. Slowly, they are healing and Cameron is getting more and more in control.
After her surgery, she is doing well in school, and can run and play. Cameron is still Cameron, only now, the problem was fixed. Imagine if we had attached a device that facilitated the regrowth and replacement of those severed neurons. What if we “tricked” the neurons into attaching to other nerve endings, say to an electronic device? What if that electronic device were something like an electronic arm?
If you have the neural function that enables an “arm” to be utilized, do you need the bones, blood vessels, skin, and other contents of the arm? What about both arms? What about the legs? Both legs? The torso? Can you isolate a human being into a brain in a jar with control of a completely prosthetic body?
I think all of these questions have answers that are not only possible, but much more simple than they had been thought to be. Even if you were to take the human eye, the vision component that must be preserved would be the retina itself. The retina is attached to the choroid layer of the eye, which is essentially a muscle that causes various contractions in the eye that manipulate the iris and the lense. Those functions can be replicated EASILY with current technology.
I think the “individual” within us all is collective data that could possibly be copied or transferred to another “host”. Humans could become parasites that create their own hosts, and I think the humor in that sentiment is worth the pursuit! And if you think about it, our brain essentially created our bodies to supply itself with energy. The brain is basically a DNA Mega-Hub where all collected data goes. If the brain is utilizing data, that data could be replicated.
Anyway, my two cents and another perspective on a great topic.
October 12, 2011
Thank you for this terrific and though-provoking post about the possibility of artificially replacing neurons or computationally recreating the brain.
I agree that if we can computationally emulate all functional aspects of a neuron, and the important supplementary role of the glia, then we should be able to reconstruct a brain in silico. However, if we assume that the only way a neuron sends information is through the firing of action potentials and the only way it receives inputs is from depolarization via dendrites, then it’s unlikely that we’ll be able to do this successfully. We’d be neglecting the role of the LFP on neurons, the role of retrograde neurotransmitters, glial interactions, any so many other factors which are not directly modeled by the H&H model of neural activity. Despite this, I think it would be technically possible to eventually model all aspects of a neuron (and any other cells which contribute to neuronal activity) and recreate basic neural circuits.