Just briefly, I want to note I'm trying to be clear about some things that we generally aren't so careful about. As I've explored mind uploading, I've limited the concept to preserving personal identity. That would be the subjective experience of losing consciousness in a body and waking up as a machine mind, probably embodied in a robot or perhaps a virtual world.
I'll never say never, but my conclusion so far is that this would be like uploading a toaster to a computer or a hurricane to a weather model. You'll never have toast and you won't get wet. This is where I keep coming back to the fundamental issue that these computer models are maps, not the territory. You end up with an emulation, not a copy of the thing itself.
When it comes to emulating functions of mind, the last few years have been revelatory. I think current LLMs easily pass the classic Turing Test and I think far exceed what a human brain can do in some aspects. Not all, but a subset. And our computer minds can easily beat us at non-verbal tasks like playing chess or Go.
But next week’s post will be about the failure we've had in even emulating C. Elegans and other simple neuronal networks even though they are all specified. Biology is something very special.
Hey, nice piece! I must say though, as I understand you, that I disagree quite strongly.
Whether it's feasible at all within a century or two is interesting in itself, but doesn't determine if it is a dead end. Also, I think much of the interest is about the hypothetical feasibility.
Uploading a mind would have to be a fine-grained enough model of a brain AND body to be able to reproduce all the same behaviours in all kinds of situations. This requires simulating an environment too.
Training is irrelevant. You don't need to train anything at all. The mind is already trained, all the parameters are uploaded. Or else it's not a mind upload.
Thanks for the comment.
Just briefly, I want to note I'm trying to be clear about some things that we generally aren't so careful about. As I've explored mind uploading, I've limited the concept to preserving personal identity. That would be the subjective experience of losing consciousness in a body and waking up as a machine mind, probably embodied in a robot or perhaps a virtual world.
I'll never say never, but my conclusion so far is that this would be like uploading a toaster to a computer or a hurricane to a weather model. You'll never have toast and you won't get wet. This is where I keep coming back to the fundamental issue that these computer models are maps, not the territory. You end up with an emulation, not a copy of the thing itself.
When it comes to emulating functions of mind, the last few years have been revelatory. I think current LLMs easily pass the classic Turing Test and I think far exceed what a human brain can do in some aspects. Not all, but a subset. And our computer minds can easily beat us at non-verbal tasks like playing chess or Go.
But next week’s post will be about the failure we've had in even emulating C. Elegans and other simple neuronal networks even though they are all specified. Biology is something very special.
Hey, nice piece! I must say though, as I understand you, that I disagree quite strongly.
Whether it's feasible at all within a century or two is interesting in itself, but doesn't determine if it is a dead end. Also, I think much of the interest is about the hypothetical feasibility.
Uploading a mind would have to be a fine-grained enough model of a brain AND body to be able to reproduce all the same behaviours in all kinds of situations. This requires simulating an environment too.
Training is irrelevant. You don't need to train anything at all. The mind is already trained, all the parameters are uploaded. Or else it's not a mind upload.
Perhaps I'm misunderstanding you!