I think the emphasis even on preservation is misguided at this point. I think it’s time now to shift emphasis to uploading & emulation.
Chemical fixation is good, sure. But we need to assemble a full map of the neurons and their connections in the brain. The current best way to do this is to image the brain in the relatively thick sections that are allowed when you have chemically fixed and optically clarified the tissue. This greatly facilitates axon tracing since there are many fewer cuts needed. You can have like 4 cm or a bit more thick slices instead of needing slices thinner than tissue paper.
Also, the risk of damage from cryoslicing brain tissue is significant. I’ve accidentally destroyed quite a few cryopreserved brain slices while trying to transfer them to slides. If you are trying to slice an entire non-clarified human brain thin enough for light microscopy, you’ll have quite a lot of risk of damage. The other alternative is hard plasticization and even thinner slicing for electron microscopy. Also a damage risk, and also something you can optionally do to the clarified brain once you are done light-imaging it. The advantage of doing thick-slice light imaging first is that it gives you a map, the full set of 3D positions of all the neurons and their axons & dendrites, which you can then add the additional detail (e.g. synapse strength) from the later electron microscopy. If you don’t have the map, then you have to correctly assemble all those super-thin slices into 3D structure without mismatching axons and dendrites. That’s another huge source of error.
Also, when doing electron microscopy, you must choose one single set of things to label. You don’t get to change your label. There are tens or hundreds of importantly relevant proteins it would be useful to know about when trying to upload a brain. With clarified brain tissue, you can label these proteins a few at a time, image them, wash the labels out and put in new labels. You can safely repeat this process hundreds of times. Thus, you get a much more complete picture of all the proteins. And you can label with multiple label colors at once, which means you can have a ‘reference label’ to which all the others get registered in 3D space on your model. This greatly reduces the localization error in your model.
Humanity is at the hinge of history, and one of the possible paths to safety from the AGI transition is having successful Whole Brain Emulations that give us digital entities we can trust more than alien-minded ML models. Thus, getting this tech working in the next 5-10 years could be critical.
This seems far more important to me than the work of trying to help people preserve their brains so they can be uploaded once humanity has survived the singularity. That’s a thing that impacts just a few people, rather than all of humanity and the future of our descendants throughout our lightcone.
I think the emphasis even on preservation is misguided at this point. I think it’s time now to shift emphasis to uploading & emulation.
Chemical fixation is good, sure. But we need to assemble a full map of the neurons and their connections in the brain. The current best way to do this is to image the brain in the relatively thick sections that are allowed when you have chemically fixed and optically clarified the tissue. This greatly facilitates axon tracing since there are many fewer cuts needed. You can have like 4 cm or a bit more thick slices instead of needing slices thinner than tissue paper.
Also, the risk of damage from cryoslicing brain tissue is significant. I’ve accidentally destroyed quite a few cryopreserved brain slices while trying to transfer them to slides. If you are trying to slice an entire non-clarified human brain thin enough for light microscopy, you’ll have quite a lot of risk of damage. The other alternative is hard plasticization and even thinner slicing for electron microscopy. Also a damage risk, and also something you can optionally do to the clarified brain once you are done light-imaging it. The advantage of doing thick-slice light imaging first is that it gives you a map, the full set of 3D positions of all the neurons and their axons & dendrites, which you can then add the additional detail (e.g. synapse strength) from the later electron microscopy. If you don’t have the map, then you have to correctly assemble all those super-thin slices into 3D structure without mismatching axons and dendrites. That’s another huge source of error.
Also, when doing electron microscopy, you must choose one single set of things to label. You don’t get to change your label. There are tens or hundreds of importantly relevant proteins it would be useful to know about when trying to upload a brain. With clarified brain tissue, you can label these proteins a few at a time, image them, wash the labels out and put in new labels. You can safely repeat this process hundreds of times. Thus, you get a much more complete picture of all the proteins. And you can label with multiple label colors at once, which means you can have a ‘reference label’ to which all the others get registered in 3D space on your model. This greatly reduces the localization error in your model.
Humanity is at the hinge of history, and one of the possible paths to safety from the AGI transition is having successful Whole Brain Emulations that give us digital entities we can trust more than alien-minded ML models. Thus, getting this tech working in the next 5-10 years could be critical.
This seems far more important to me than the work of trying to help people preserve their brains so they can be uploaded once humanity has survived the singularity. That’s a thing that impacts just a few people, rather than all of humanity and the future of our descendants throughout our lightcone.