Jump to content

Could you copy the brain to a computer?


gmpd2000

Recommended Posts

I see all of you are ignoring my earlier post. I've said that the brain is not a bunch of neurons communicating by ionic pulses. It's a wet, squishy pile of gooey tissue which also secretes hormons which are then dispersed by its vessels to the whole body, including the other parts of the brain. Those hormons are chemical signalization and they are included in lots of feedback loops. Other organs interact with the brain, chemically and electrically. How are you going to simulate something that's driven by chaos?

I wouldn't assume that people have ignored you. I thought it was an excellent point. Personally I didn't reply because I didn't have an answer. What you took as silence might just have been the sound of chins being stroked.

Link to comment
Share on other sites

It's a wet, squishy pile of gooey tissue which also secretes hormons which are then dispersed by its vessels to the whole body, including the other parts of the brain. Those hormons are chemical signalization and they are included in lots of feedback loops. Other organs interact with the brain, chemically and electrically. How are you going to simulate something that's driven by chaos?

From what you said it does not follow that the brain is driven by chaos.

However it does mean the workings of the brain is even more complicated than just a "neural network". Otoh that in turn does not mean it can not be simulated.

Link to comment
Share on other sites

Of course, to do this you need to both map the entire brain, and measure, very precisely, concentrations of various chemicals at all of the synapses. We don't have the tech to do this yet, but the basic approach of freeze-and-slice is definitely workable here. So it's a matter of time before we'll be able to get all of this information from an actual brain. If you want to do this in-viva, it's a different question entirely.

Naruhodo. Figuring out the strengths of the connections and then simulating it may be possible using the freeze and slice technique. But why is it necessary to also measure the concentrations of various chemicals in the synapses? I thought the connections strengths was just a matter of the topology of the dendrites? Doesn't the chemical makeup of the synapse depend mostly on what phase of the spike a neuron is in?

Of course, once you have a simulation running, working out the actual way the information is stored sounds much more plausible. But "decoding" the brain would require way more processing power than just simulating it. So again, this is not something we're going to do for a long, long time.

Link to comment
Share on other sites

Naruhodo. Figuring out the strengths of the connections and then simulating it may be possible using the freeze and slice technique. But why is it necessary to also measure the concentrations of various chemicals in the synapses? I thought the connections strengths was just a matter of the topology of the dendrites? Doesn't the chemical makeup of the synapse depend mostly on what phase of the spike a neuron is in?

How do you imagine memory works? Certainly, things you learn to do or very long term memory is topological. And very short term memory can be dynamic. But you can memorize something short term, not think about it at all, and then recall it. And that's only possible with altering the weights of existing connections. And in fact, we do know that there are some chemical concentrations that can linger at a particular synapse and affect neurotransmission. I don't know if there is any research that demonstrates conclusively that this is the very thing that does short term memory, but I don't see any other way. And at very least, we should assume that it's relevant.

Of course, it might not matter too much. We might be happy enough doing a reset on all of these. At worst, we'll have effect of a person who awoke from a short comma, having lost some of the recent memories. Since procedure discussed here is terminal, I'd guess that it'd only be performed on someone who just kicked the bucket, and given the choice between awaking like I've had a head-trauma-induced comma or just being dead, I'd chose the former.

So I would say that it's very likely that we'll need to simulate these, and again, to within an effect of a mind-altering substance, we certainly can, but you might be right that it's not critical that we get these from the scans. Which would make things considerably easier. We could probably build an electron microscope that takes sufficiently high resolution images of the slices. (We have good pictures of dendrites due to electron microscopes, the challenge is taking the picture of an entire slice and process it.) As for going from slice to slice, we'd probably need to find a way to evaporate a thin layer of freeze-dried brain, take a picture, repeat.

Link to comment
Share on other sites

Nope. I didn't said the whole being would be an hallucination, but just that you can't know if anyone else has a consciousness like yours.

Without induction we can infer solipsism, and even if we could not, I could still doubt whether you are a consciousness.

It doesn't deny empiricism, it merely states that all empirical knowledge is necessarily inductive, since it follows from an induction.

You have denied empiricism by denying induction, an empirical cornerstone.

A cornerstone conveniently shattered by quantum phenomena linking subject and object. We already reject monism in many fields.

Your rebuttal contains three fallacies:

-It begs the question by presuming that souls exist in order to prove that they do.

-It presumes the false premise that some known quantum phenomenon links subject to object: you may be thinking of the "observer effect" whereby photons by hitting quantum-sized things bump them.

-It presumes the false premise that monism is rejected in many (presumably scientific) fields: monism is a cornerstone of scientific metaphysics.

Right. Inducted. An induction might be enough for most practical purposes, but it isn't enough to solve the binding problem, and that's the point here.

The "binding problem" begs the question by assuming the existence dualistic qualitative phenomena in order to prove that monism cannot explain psychology and that the universe must be at least dualistic. Consistently assuming that fundamental principles can from data be inducted--rather than only for "most practical purposes"--would allow us to from neuropsychological data induct other consciousness' existence.

-Duxwing

Link to comment
Share on other sites

I'm joining the discussion late, but here's my answer to the original question:

I don't feel that we understand the human brain well enough to definitively answer this question.

But,

My gut feeling is that yes, it would be possible. We would need way better computers and mastery of neuroscience, but the human brain is a real world phenomenon. I see no reason it couldn't be modeled given sufficient computing power.

How much computing power? Lots. How much is lots? Enough to model the human brain. Could be a while... or it could be this century. Or we might not be smart enough to ever make it happen, even though it is possible, in principle.

Link to comment
Share on other sites

We could probably build an electron microscope that takes sufficiently high resolution images of the slices. (We have good pictures of dendrites due to electron microscopes, the challenge is taking the picture of an entire slice and process it.)

The current tools would do, unless you were in a crashing hurry. Probe tips can raster back and forth to build up a picture of a large area. Resolution is already above what would suffice to map large structures like neurons and identify molecules, we can image individual atoms after all.

Link to comment
Share on other sites

The current tools would do, unless you were in a crashing hurry. Probe tips can raster back and forth to build up a picture of a large area. Resolution is already above what would suffice to map large structures like neurons and identify molecules, we can image individual atoms after all.

To image individual atoms, you need an AFM, which is going to scan the surface a few angstroms at a time. I don't think you comprehend just how long it would take to scan a volume. The sun will be big and red before you're remotely done.

The only way to get this done in half-reasonable time is taking full 2D snapshots at a time, and the only tech that can do that with sufficient resolution is electron microscope. I just don't think we have any that have sufficiently large matrix to take a snapshot of an entire slice in one go. Even as is, you'll need to get through 10cm at about 1nm a go, which means you'll have to capture 108 images. At 100 images per second, it will take you two weeks. That should put it into a bit of a perspective.

Link to comment
Share on other sites

To image individual atoms, you need an AFM, which is going to scan the surface a few angstroms at a time. I don't think you comprehend just how long it would take to scan a volume. The sun will be big and red before you're remotely done.

To image squishy biological samples using an AFM you're in tapping mode, with a step measured in several nm at least. Rapid AFM imaging is used in the biological sciences, they even use it for imaging dynamic processes. Reported tip speeds are above 0.5mm s-1, which does seems extremely fast to me, but I guess they know what they're doing. Now, I'm not an instrument engineer, so I don't pretend to understand all the constraints on AFM tip design, so I wouldn't like to guess the density of tips you could cram into a hypothetical multi-headed machine, but I imagine it would be a fair few if you needed a fast machine.

AFM probably wouldn't be the only tool you'd use. You'd want to be combining 2D imaging with 3D if possible. You'd be combining AFM with stuff like NMR too. The imaging technique might depend on what you were trying to identify and what section of the brain you're looking at. Some structures (vasculature, sinuses, etc) wouldn't need to be imaged at high resolution at all.

Bottom line though, imaging a whole brain with the kind of precision we're talking about isn't a job for a single machine in a single lab. You're talking about an international effort over several years, akin to the HGP. You'd have useful results after you'd managed to fully image and simulate the first brain, since beyond that you'd be wanting to manipulate it digitally. It may not be necessary (depending on your objective) to ever image more than one brain. Certainly there would be diminishing returns from doing so.

Link to comment
Share on other sites

I'll let you cram these heads every .5mm. That's already way smaller than any AFM/STM head I've ever used or seen. And we'll do that in the entire plane. We definitely want resolution in the nm scale, but if you insist that 1nm is too precise, lets do 5nm. Certainly we can't be doing worse than that. And that agrees well with your "every few nm" estimate.

So with heads every .5mm, according to your numbers, we'll be moving across the entire row in 1s. Then we need to shift heads across by 5nm and repeat. And again, we need to cover .5mm. That means to scan a layer with 5nm resolution we need 105 seconds with this incredible AFM array. That's over a day. For one layer. Just 2x107 layers to go. You'll be done in, oh, 60 thousand years, give or take.

Have fun with that. I'm going to stick with my electron microscope approach.

I keep trying to impress on you just how mind bogingly vast the number of points you need to sample is, and you keep giving me things that improve on it by factors of a few hundred. I mean yeah, this is a huge improvement over trying to do this with a single head, but we're still talking geological time scales. You can try and compress these heads to 100 per square mm. I mean, that's almost as dense as pixels on your screen. (Or denser, if it's a large screen.) And we'd still be talking thousands of years to scan the whole thing. This is far beyond plausible for any sort of a project, and we've pushed the tech far past what can actually be built.

I'm fine with it being a long project. But we should be talking years. Decades at the worst. And it's not like we don't have tech to do that. We can literally take a picture of an entire layer at once with an electron microscope. If you really need it to be 3D, there are techniques for getting relief using an electron microscope. That's not a problem. And unlike the AFM, we can get the matrix as large as we want, because matrix can be much larger than sample. If we have to break up the beam and image different portions of the plane with different matricies, that's fine too. We can do that. In fact, we can do almost anything we can do with ordinary microscope and imaging, but at much higher resolution.

Electron microscope is definitely the way to go here. It's just a matter of building one that's capable of imaging such large objects with such fine resolution in one go. Which would be ridiculously expensive, but it's purely an engineering problem. Not a conceptual one.

Link to comment
Share on other sites

You'll be done in, oh, 60 thousand years, give or take.

Or ten years using 6,000 machines. A big project, but not unrealistic.

Have fun with that. I'm going to stick with my electron microscope approach.

Well, your machine doesn't exist, but go for your life. Since we're both talking completely out of our arses about a totally hypothetical situation (which is frankly a little bit ridiculous) I'm not going to tell you that your crazy idea is off limits because I think it's slightly more crazy than mine.

I keep trying to impress on you just how mind bogingly vast the number of points you need to sample is, and you keep giving me things that improve on it by factors of a few hundred.

I know. There's a lot about the whole concept of copying a brain into a computer that I'm massively sceptical about. I'm not dishing out certainties here, and suspect anyone who is doesn't understand the problem or the actual state of the technologies we might use.

it's purely an engineering problem. Not a conceptual one.

This is phrase I hear a lot on these forums, and it always makes me laugh. Declaring something to be "an engineering problem" helps very little. Virtually everything we can't do is "an engineering problem", so you're not including it in a very exclusive set. Spacecraft propulsion at up to light speed is "an engineering problem", for example, as are invisibility cloaks and living for 500 years. Declaring something to be "an engineering problem" is a lazy answer unless you actually know how to solve the problem. We might as well say teleportation is possible, it's "just a science problem".

Edited by Seret
Link to comment
Share on other sites

Or ten years using 6,000 machines. A big project, but not unrealistic.

Sure. As soon as you tell me how you plan to slice the brain into 6000 layers without disturbing anything at the interface and losing information on cruicial connections. Any known method of slicing is going to cause you a lost of fractions of the micron at the very best. We are talking physical limits here. If you wanted an example of something that's a conceptual problem, rather than an engineering one, here it is.

Well, your machine doesn't exist

Electron microscopes don't exist? Or are we talking about one with sufficiently large matrix? Because it's just matter of scaling. When I'm saying an engineering problem, it literally means we just need to put a bunch of engineers down, and they'll put together a project. Just want to be clear on that.

Link to comment
Share on other sites

it's just matter of scaling.

Pretty much everything about the problem is about scale. The whole problem requires us to operate at scales both greater and lesser than we're actually able to do. I would say that scale IS the problem.

Link to comment
Share on other sites

im convinced you are going to need an instantaneous snapshot (il let the engineers figure this one out, im stumped). you not only need to know where the neurons are and whats connected to what, but also the electrical potential and neurotransmitter levels at each. if you sample the brain over time you are going to lead to all kinds of information skew. we dont know how much of our conscious is stored in the layout of neurons and how much is stored as electrical impulses. but if you dont have a consistent picture of both in a sufficiently small enough slice of time, your chances of the simulated brain starting up where it left off is going to be nil.

this is a horrible analogy for many reasons, but if you try to copy the state of a running cpu the hard way (cpus actually have this functionality to facilitate multitasking oses, you can grab the state of all the registers and save them to ram), by looking at every transistor and recording its state, by the time you record the state of the last one, the cpu has gone through several cycles. you copy that state into a new cpu, and it will just spew out random garbage until the instruction cache runs dry. doing something similar to the human mind would be traumatic at best fatal at worst.

Edited by Nuke
Link to comment
Share on other sites

One of the problems is that the brain and nervous system runs not only on electrical impulses but also chemical, the reason for emotion. So you would have to figure out a way of stimulating that as well. Another problem to consider is stimulation and interfacing with the mind. As we all know when we speak we are using our voice box. Same with movement, when we want to move our arm our brain causes muscles to contract. So unless you just want the mind to sit in the computer with no stimulation you first need to figure out how to interface with it.

Link to comment
Share on other sites

but also the electrical potential and neurotransmitter levels at each.

We are taking apart a dead brain here. Electrical potentials are at ground level. There is no way to do any of this in-viva, so it's not something we can get around. Fortunately, this only resets current thought process. And as we've discussed earlier, even getting just the topological information would be sufficient for recovering all of the personality, skills, and most of the memories. It'd be equivalent from waking up from severe head trauma, but that's acceptable if the other alternative is plain death.

Pretty much everything about the problem is about scale. The whole problem requires us to operate at scales both greater and lesser than we're actually able to do. I would say that scale IS the problem.

Well, based on your numbers I need snapshots with 5nm resolution. An electron microscop can be built to be entirely digital, none of that archaic phosphorecent stuff. Effectively, we can use a modern digital camera matrix almost without modifications. These things have about 2k pixels packed per cm. So a single 1x1 cm cell can image 10x10 micron area of the brain. That isn't much, but since we can control the electron energy as precisely as needed, at least 10kHz is doable on all of this. So a single cell manages to give us 1mm² per second. To give us some room for errors, lets say we cover a 20x20 cm area. That's 400 camera sensors and steering/focusing magnet assemblies to image a single layer at 5nm resolution in one second.

2x107 layers, 2x107 seconds, and we're done in well under a year.

All based on existing technology, with a system that's effectively just 400 modern electron microscopes tied into one machine. And we can probably use the same electron beam to evaporae off layers.

So again, when I say "matter of scale," I do mean just a matter of reasonable scale. I do make these mental estimates any time I throw out an approach or a figure. I suggest you learn to do the same.

Link to comment
Share on other sites

From what you said it does not follow that the brain is driven by chaos.

Chaos theory states that phenomenon that are seemingly random are actually complex systems that are determined by pre-existing conditions. The conclusion is that nothing is really random and everything is actually predictable, as long as you can create a complex enough model.

In other words, the shape of a cloud in the sky might appear random, but it is actually the result of water molecules acting on each other, combined with atmospheric pressure and temperature conditions, wind, gravity, etc... all linked to the pre-existing configuration of those molecules. You could theoretically predict the shape of that cloud if you had a precise enough algorithmic model and enough processing power.

The same goes for a functional brain. If you had a precise enough algorithmic model of the neurons, synapses, and their electrical and chemical bonds, then you could theoretically simulate a fully functional brain. The rest is just a matter of computer technology being capable of running such algorithms, which is just a matter of brute force.

Link to comment
Share on other sites

A note on chaos. Even if human brain is subject to chaos, all it means is that we can't make a simulation that's when running alongside real brain and receiving all the same stimuli gives identical thought process. But that's not necessary. Look at it this way. If all of the aspects of your persona were subject to chaos, then everything you do or think would be completely random. Even if that was the case, which it obviously isn't, then there is no need to simulate a particular person, because all people are completely random and indistinguishable anyhow. Of course, we can note certain aspects of personality in each individual, which tells us that the core aspects of what makes a person that person are not subject to chaos. Sure, when you wake up in the morning and decide which pair of socks to put on, that might be subject to chaos, and simulation will make a different choice. But does choosing one pair or another make you a different person? Of course not.

So we are looking at two aspects here. The exact thoughts and choices, which might be subject to chaos, and broader personality traits that make critical decisions which are not. The later can certainly be simulated with finite resources. And that's good enough for us. The only pitfall I can imagine with an artificial brain is that it might end up not being subject to sufficient chaos, and be too deterministic and predictable. But this is very easy to fix by introducing a bit of numerical noise into the neural network simulation, until the simulation behaves the same way as the real thing, qualitatively speaking.

The final point is that while both classical chaos and quantum uncertainties might play a role in the thought process, it's not an obstacle for us in simulating human brain. These effects can be replicated, and the fact that it prevents us from making an exact replica just says that we are truly simulating a personality, and not just carbon-copying a working mind.

Link to comment
Share on other sites

And decoding the 'meaning ' of individual neurons, or at least the activation pattern corresponding to certain ideas is fundamental to the problem of communication. Before disembodied brains can ever use the powers that come with being simulated on a computer (high-bandwidth control of machines, advanced senses, and advanced communication) they must be decoded. I can't share a complete experience with you if the machines don't know how to translate the activation patterns from my brain's code to yours.

K^2 already briefly state that neurons don't really need to be decoded, since the brain is a black box that kinda does the decoding by principle when interacting with the outside world. But your view on advantages of a simulated brain is kinda limited. Most likely any function principle of our brain we understand will allow us to do awesome stuff, a lot of it without the need to "decode" neurons. Just imagine multiple instance of the same brain being active at the same time & working on the same neuronal network (the good old "clone yourself to get stuff done" dream). Maybe we will speed up frequency of task specific regions/neurons to improve reaction times. What about suppressing regions/neurons, so you won't get distracted? Whatever principles we'll learn about our brain, a digital version allows us to modify it, play around with it without waiting lots of generations for evolutionary processes. Even genetic engineering will doubly be able beat that.

Edited by Faark
Link to comment
Share on other sites

In regard to chaos theory: Complex and seemingly random systems being built out of predictable individual particles with starting conditions was the general consensus among physicists prior to the 20th century and determinism was king.

Then along came quantum mechanics... and eigenstates. At the quantum level, things truly are random. Some eigenstates have higher probability but there is no way to predict exactly which state a quantum particle will choose when observed.

I choose the double slit experiment for my example: Carefully study and precisely measure all the initial conditions you want, you'll still never be able to predict with certainty exactly where an individual electron will end up.

You can predict, however, how a large sample size of electrons will end up being distributed based on the probability distribution of eigenstates... well fairly closely, anyway.

Link to comment
Share on other sites

A note on chaos. Even if human brain is subject to chaos, all it means is that we can't make a simulation that's when running alongside real brain and receiving all the same stimuli gives identical thought process. But that's not necessary. Look at it this way. If all of the aspects of your persona were subject to chaos, then everything you do or think would be completely random. Even if that was the case, which it obviously isn't, then there is no need to simulate a particular person, because all people are completely random and indistinguishable anyhow. Of course, we can note certain aspects of personality in each individual, which tells us that the core aspects of what makes a person that person are not subject to chaos. Sure, when you wake up in the morning and decide which pair of socks to put on, that might be subject to chaos, and simulation will make a different choice. But does choosing one pair or another make you a different person? Of course not.

So we are looking at two aspects here. The exact thoughts and choices, which might be subject to chaos, and broader personality traits that make critical decisions which are not. The later can certainly be simulated with finite resources. And that's good enough for us. The only pitfall I can imagine with an artificial brain is that it might end up not being subject to sufficient chaos, and be too deterministic and predictable. But this is very easy to fix by introducing a bit of numerical noise into the neural network simulation, until the simulation behaves the same way as the real thing, qualitatively speaking.

The final point is that while both classical chaos and quantum uncertainties might play a role in the thought process, it's not an obstacle for us in simulating human brain. These effects can be replicated, and the fact that it prevents us from making an exact replica just says that we are truly simulating a personality, and not just carbon-copying a working mind.

this is one of the reasons why i think it would be optimal to use an analog neural net rather than a computer.

Link to comment
Share on other sites

From what you said it does not follow that the brain is driven by chaos.

However it does mean the workings of the brain is even more complicated than just a "neural network". Otoh that in turn does not mean it can not be simulated.

Brain or any other organ is very similar to the weather or any other natural phenomena where high number of particles interact in a fluid. It is exactly chaotic by the very definition of it.

You might simulate it poorly, but after some time has passed, two identical models will start behaving in a totally different way. Brains are unpredictable and that's exactly why its emergent phenomena are so complex and unique.

Link to comment
Share on other sites

You might simulate it poorly, but after some time has passed, two identical models will start behaving in a totally different way.

And that's bad how? If you were to copy human brain exactly, molecule for molecule, and run original and copy side by side, they wouldn't behave exactly the same way either. Because that's what chaos means. So why should you demand more from simulation than the original?

Link to comment
Share on other sites

And that's bad how? If you were to copy human brain exactly, molecule for molecule, and run original and copy side by side, they wouldn't behave exactly the same way either. Because that's what chaos means. So why should you demand more from simulation than the original?

Well this is a good point, but doesn't change the fact a simulation on a computer would be very poor compared to the real thing. Otherwise any level of simulation quality could be considered valid.

Link to comment
Share on other sites

This thread is quite old. Please consider starting a new thread rather than reviving this one.

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...