I am interested in the technical challenge of whole brain emulation. My talk is to some degree about whether in an ideal world if we have the characters that ramble. It's going to be about optimization. This option .. we have to pick something and be straight in causality. Why are we interested in emulating brains when there are so many good reasons to do pure ai? This is a good exercise. We actually know quite a lot about the brain, possible scanning methods, and we can look at time scales. I am looking to show a rough time scale of what we can do and when we could do brain emulation. That puts a time limit on ai because if you don't get there before we do, then yeah. We know roughly what to do with them. We have a lot of experience dealing with intelligence. An intelligence explosion.. research ethics. IN general I believe that this could be a very big thing. The philosophical impact can bne demonstrated. The philosophy departments is taring their hair. Robin has written about the economic impact of copying. We might want to know how far away this is, because it could be dangerous. If you happen to subscribe to functional views, this could be a way of expending your life indefinitely. I am talking about emulations.

This is about engineering without understanding of what's going on. What does the brain do and what is memory, consciousness or intelligence? You can also study it on a more fundamental level. How do ion channels work. What if yhou get to adjust the drawings for this? You can make new use of this. If we have a low-level understanding, we might be abl eto get something tghat could wokr. But, a high level understanding might not allow us to make a copy. So, in general, I don't have time to go into the details. There are many philosophical questions. In particular there is the problem if this is mind uploading? This is about brain emulation. I don't know if a complete simulation will make a mind. I want to test it. I want something that will do what the brain does. There is a level of detail. I don't know where I would need. We could make an atomic-level copy of the brain. That would also be ab rain. The personal identity issues aside. What about the high-level? What if we replace some of the structures, some of the functions. Well, you can't just get intelligence by stringing together some vision processing system. Or just auditory and decision system. That would most likely require very deep understanding of a brain. That wouldn't correspond to a particular brain. What goes on chemically in the brain? This doesn't require us to figure out what the mind is. It's just a matter of chemicals and figuring out how to simulate them. There are levels. Hopefully we don't have to deal with quantum states. At some level we can do this. The interesting problem is that gradually we can acquire data. It is easy to acquire data on the low level. We don't need much understanding. Just use lots of scanners. It takes lots of computation to simulate the quantum level. Simulating abstract neurons like in a neural network is even less complex.

Scale separation in the brain. There is a lot going on in between atoms and the gas. You can ignore that with statistical mechanics because they average together to make pressure, volume, and we get properties that we can't understand without understanding what the molecules. Stuff going on at the small scale going on at the large scale. We also have a pretty large stuff breaking down into small steps. So, at scale separation, we might be able to separate it. If the brain is similar, we might be able to do this emulation because we always need the lower levels. I am rather optimistic about this. I haven't had any good arguments about why there is a ... this is a scientific question that we need to resolve.

Function from image. How much information about the function of the brain that we can get from a brain. There's brain tissue. Those pictures of very beautiful neurons floating in space. That's a false image. The brain is more like a 3D puzzle. It's not only where the neurons are. There are branches going through the image going through there. There's also inhibitory and excitatory stuff going on here. How much could you get from an electron microscope graph? There are interesting complications. There are fundamental issues. What about the glia? They actually turn out to be not too computationally expensive to simulate. The rough consensus that we arrived at is that we have to get this to .. we need to know the cells, we probably don't need too much of the really low level. That wsould require a resolution of five times times five. But it turns out that it can't go far beyond that. We need to figure out the relative information. This is what's limiting us currently. We can't scan a large volume. They can only scan a little voxel. A micron or smaller. And we want to scan a 1.4L brain and we want to scan it fast. There are many scanning machines. This would correspond to many methods.

This is my favorite. This is optical resolution. This is from Texas A&M. Knife edge scanning microscope. It has a knife slicing some brain tissue as it goes along. The result is , for example, this is a piece of mouse spinal cord. You can 3D reconstruct it. Thanks to Todd Huffman, I found this. They actually scanned an entire mouse brain at this resolution. These are golgia neurons in the cerebellum of the mouse. They have a nice structure. There's a lot of fibers running through these branches. Not quite as messy as the cerebral cortex. This is real data and real projects that we have right now. We need better ways to ahndle this data. There is a lot of ways to get the data out of there.

Array tomography is a good way to get the chemical state of the data. You can use electron microscope and AFM to get finer resolution. The problem is interpreting the scans. That will take a lot of development. :People are working on it, but more people are needed. YOu can do this by hand. You can have grad students that can do this by hand, and grad students are cheap. It's great to have them around. The intelligence explosion would allow us to have even cheaper grad students. There's some work on turning this messy construct into this nice 3D stuff. So the next step is figuring out softwares that takes this and figures out a computational model. We have a lot of good models based on compartment modeling.