|
|
|
|
|
|
by api
1081 days ago
|
|
|
The best way of looking at it is that a single biological neuron is itself a complex machine full of genetic control circuits that sort of resemble neural networks and most importantly have memory/state that persists over both short and long periods of time. Each neuron is a full-ass living organism that itself is capable of learning and behavior, not a parameter is a model. A virtual "neuron" by contrast is a very simple mathematical abstraction. It's vastly less computational complexity than a biological neuron. A connectome is only a very coarse grained map of how neurons relate, not a complete "neural network" layout. Not even close. It might be possible to model a biological neuron using a sub-neural-network with state within a larger neural network, but assuming that can be computationally equivalent we don't really know how many equivalent computational "neurons" would be required to model the full breadth of computationally relevant biological neuron behavior. So a worm with 302 biological neurons could be computationally equivalent to billions of virtual neurons. We really don't know. Given that neurons have memory it may look a little like LSTM networks, and biological neural networks are not just feed forward so they're definitely closer to an RNN. The above is why I laugh at the mind uploading people and would only stop laughing if we could both understand and model the relevant behavior of biological neurons and somehow extract usable state from living neurons. That's all 100% science fiction at the moment. The people who think we are about to upload minds are ignorant of biology. |
|
|
The even bigger issue is when mind uploading people fully admit this issue and try to claim some philosophical reason why it doesn't matter, and we should all be excited about tech to make what amounts to an interactive epitaph.