"The way the prophets of the twentieth century went to work was this. They took something or other that was certainly going on in their time and then said that it would go on more and more until something extraordinary happend." G. K. Chesterton
What if you could emulate the brain the way we emulate computer worms?
C elegans is a 1mm long worm with 302 neurons, 3 Nobel prizes and has survived a space shuttle crash. It is one of the simplest animals and has been studied in massive detail. Like the fruitfly or ecoli anything these lab animals do that you cant explain you wont be able to explain in people either.
Whole brain emulation is a prediction that we will be able to simulate the brain in enough detail to create artificial intelligence very like us.
Robin Hanson on econtalk talked about the possible results of whole brain emulation.
"This scenario, which we've called whole brain emulation--taking a whole brain and emulating it on a computer--requires three technologies. One is scanning--you have to be able to scan something in sufficient detail; have to see exactly which parts are where and what they are made of. Two, you have to have models of these cells, a model of the cell input signature and then what comes out of it as a mapping--doesn't have to be exactly right, just has to be close enough. Three, you need a really big computer. A lot of cells, a lot of interactions."
Hanson blogs about brain emulation here. There are interesting fights about whether whole brain emulation is a reasonable prediction or just "the rapture for nerds".
Many biologists seem to think computer people are completely misunderstanding how complicated biological systems are and computer sciencey whole brain Emulation types say that biologists do not understand abstractions because they deal with this complexity all the time.
'[Robin] Hanson’s fundamental mistake is to treat the brain like a human-designed system we could conceivably reverse-engineer rather than a natural system we can only simulate. We may have relatively good models for the operation of nerves, but these models are simplifications, and therefore they will differ in subtle ways from the operation of actual nerves. And these subtle micro-level inaccuracies will snowball into large-scale errors when we try to simulate an entire brain, in precisely the same way that small micro-level imperfections in weather models accumulate to make accurate long-range forecasting inaccurate.' is an example of the biologists argument against brain emulation.
'We should expect brain emulation to be feasible because brains function to process signals, and the decoupling of signal dimensions from other system dimensions is central to achieving the function of a signal processor.
"We can do trend extrapolation and say: Where are we now; if trends continue how long would it take? The computing technology has a nice solid trend; we can project that pretty confidently into the future. The problem is we don't really know how detailed we're going to need to go into these cells. The scanning technology, we have decent trends. This is a vastly smaller industry; small demand. That technology actually looks likely to be ready first. We've actually done a scanning of a whole mouse brain at a decent resolution. A thousandth smaller than a human brain. What does that mean--scanning of a brain? They slice a layer, do a two-dimensional scan of that layer at a fine resolution, go across each cell, and then they slice another layer and do the same thing again. Let me ask again, sort of naive question: If you could take a person's brain out of their head while they were still alive, are you going to be able to get access to my memories in this process? my creativity? All these things we think of as more than a physical process, but of course as you say, it's just chemicals interacting. Is it imaginable that we would be able to reconstruct my memories? To the extent we are confident that your memories and personality are encoded in these cells and where they are and how they talk to each other, so we get that right, we get it all right. That's all you are. Let me say it differently. Looking at it isn't enough. Scanning means noticing the chemical densities. There's thousands of kinds of cells in your brain, and each cell sort of behaves a bit differently. What we need is to know when a cell gets a signal from the outside, electrical or chemical signal, how does that change a cell and what kind of signal does it send out. So, we need to have a model of each of those cell types. We have, actually, models of a wide range of cell types. Doesn't seem that hard to model these cells. We just have a lot of cells to go through and not that much motivation to do it all in a rush. We have actually pretty good models of some particular cells. We have a cell on a dish, we send a signal in, model on the computer, do the same things."
Both sides here. The brain is really complicated squishy stuff and the simcity looks like a real city if you squint sides here could be right. I want a comparison of the predictions of these two theories now and not in 2040 though.
If we had Hanson's 1,2,3 met for an organism and we were not emulating it that would seem to be a problem for the theory.
One is scanning--you have to be able to scan something in sufficient detail;
The c elegans connectome was mapped in 1986
Two, you have to have models of these cells, a model of the cell input signature and then what comes out of it as a mapping--doesn't have to be exactly right, just has to be close enough. There are not many types of neurons in c elegans so we should have a fairly good model of when they will fire.
Three, you need a really big computer. A lot of cells, a lot of interactions."
How big a computer would you need to model all these cells and interactions?
In When will computer hardware match the human brain?
Hans Moravec (1997) gives some nice graphs of how much processing you get for $1000
Kurzweil gives similar figures here
This puts the amount of processing available to a C. elegans at about 1990 levels for $1000. So in 1986 that processing power would have easily been available to university researchers. Maybe that graph is optimistic but if it is out by 25 years for something as simple as c elegans that means predictions of whole brain emulation by 2050 also on the graph will be out as well.
For the last 25 years we have had the power to emulate the whole brain of C. Elegans. Why haven't we?
1. We have not actually because our neuron firing models has not been accurate enough
2. No one cares about emulation fo a worm. A lot of people care about this worm the numbers of neuroscience papers on it confirm this.
3. They are just a bit delayed. There is a thread here on less wrong about this
an open source project openworm*
4. It is hard to get output from a worm 'Our first goal is to combine the neuronal model with this physical model in order to go beyond the biophysical realism that has already been done in previous studies. The physical model will then serve as the "read out" to make sure that the neurons are doing appropriate things.' Pixar, special effects companies and computer game programmers must have fairly good worm emulation programs. If there is a big problem making an animal bodies simulation surely one of them could easily enough make a good model of a tiny bag of gunk?
'Whole Brain Emulation A Roadmap' acknowledges the gap that exists in our emulation of the animal and suggests alternatives ' While the C. elegans nervous system has been completely mapped (White, Southgate et al., 1986), we still lack detailed
electrophysiology, likely because of the difficulty of investigating the small neurons. Animals
with larger neurons may prove less restrictive for functional and scanning investigation but
may lack sizable research communities'
Why 25 years after having a good map and enough computation to run the calculation have we not emulated C Elegans? If it is the modelling of the cells
'I have talked several times to one of the chief scientists who collected the original connectome data and has been continuing to collect more electron micrographs (David Hall, in charge of www.wormatlas.org). He has said that the physiological data on neuron and synapse function in C. elegans is really limited and suggests that no one spend time simulating the worm using the existing datasets because of this. I.e. we may know the connectivity but we don't know even the sign of many synapses.'
The openworm project is really cool. and it might be a good way to get some evidence into the whole brain emulation debate now.
'The problem is we don't really know how detailed we're going to need to go into these cells. '
If Ken Hayworth is right and it is just that 'He has said that the physiological data on neuron and synapse function in C. elegans is really limited' is this because the biologists are right and step 2 the cell models will not be as easy to build as supporters of whole brain emulation claim?
These is a project here to emulate C Elegans. And a good paper here on the problem involved Dynamics of the model of the C Elegans neural network. Just in time to make me look more stupid is A Worm's Mind In A Lego Body. It is not full emulation yet but it is at lest on the path there. * An article from Popular Science on Whole Brain Emulation here that made me resurrect this post I drafted two years ago. Since then the openworm project has moved on massively.
No comments:
Post a Comment