thousands of very smart people put in lots of effort to apply, and the org says "Oh no, not any of you guys"
(I'll note that I'm qualified to be a grantmaker for the area of human intelligence amplification, which many leaders in AI safety view as a crucial second or third priority behind "stop AI" and "at least try to solve alignment". But it seems like a waste of my time to apply to be a grantmaker at CG without some indication that they'd be open to this. I did message a couple people hoping for a quick "nah" or "worth applying" but didn't hear back.)
As a first draft, roughly, I think the speculation is worthwhile IF AND ONLY IF it's in a context where it will then be followed up by maker/breaker investigation, on the question of "whether / how this can actually lead to SHIA in the real world". This includes going back and forth between skeptically searching for flaws, and optimistically searching for workarounds/alternatives/reasons for hope. It also includes thinking about the whole process of getting to the working tech, including
would this even increase intelligence meaningfully, and how would we know
getting researchers and funding for the research at various stages
having intermediate feedback on success
questions about how society will receive it--researchers, regulation, funding, and deployment are all related to this, so if you're so dismissive of these questions that you don't consider them at all, there's a significant chance you're just barking up the wrong tree in terms of actually getting this done
You're talking about revascularization? It's interesting, but would need a lot of fleshing out.
To step back a bit, I appreciate you thinking about these things and proposing ideas, but in order to make something actually work, I think there has to be a lot more in depth exploration. In particular, there'd have to be iterative maker/breaker investigation of the idea. In other words, I think you should argue against your own idea, then improve the idea and counterargue in favor, then critique the new version again, and repeat. Then for some ideas, you might actually convince yourself that the idea isn't that workable or promising; for other ideas, you might be able to make a more convincing case and/or put together a promising version of the project.
Possibly. I'm not so clear what's going on with the experiments. The one you cite that has a healthy animal getting smarter also states "Mice allografted with murine GPCs showed no enhancement of either LTP or learning.", which suggests the same wouldn't work with humans. Possibly you could do it with gene-edited human neurons / stem cells. But it feels super speculative whether that would improve much. But maybe.
Like what? Which ones plausibly significantly increase intelligence?
Note that this has significant risks and huge resulting difficulties. People generally don't want to offer crazy elective risky surgeries.
(I general I don't much buy "try it on animals and see what works" because of the issue where human brains are exceptional, and where we wouldn't have a great way of testing intelligence in animals IIUC.)
This is possibly interesting, but it would need more argument / detail. At the moment, mainly I'd view it as an interesting vector for gene editing a bunch of neurons. I'm skeptical of things like "add a bunch of BDNF" increasing intelligence much, but I could maybe be convinced otherwise.
Probably for me to want to suggest that someone fund a project on this, you'd need an expert on board, who can explain well what's been tried, what the bottlenecks are, what you're going to try that's different, why it would plausibly work, why it would be able to support the cargo, what the cargo is supposed to do, etc.
(I'll note that I'm qualified to be a grantmaker for the area of human intelligence amplification, which many leaders in AI safety view as a crucial second or third priority behind "stop AI" and "at least try to solve alignment". But it seems like a waste of my time to apply to be a grantmaker at CG without some indication that they'd be open to this. I did message a couple people hoping for a quick "nah" or "worth applying" but didn't hear back.)
Cf. https://tsvibt.blogspot.com/2022/09/dangers-of-deferrence.html https://www.lesswrong.com/posts/jzy5qqRuqA9iY7Jxu/the-problem-of-graceful-deference-1 https://www.lesswrong.com/posts/zLG2DnJw6oEZqAgaE/tools-for-deferring-gracefully
I'm talking both about politics, but also and mainly about the technical plan.
As a first draft, roughly, I think the speculation is worthwhile IF AND ONLY IF it's in a context where it will then be followed up by maker/breaker investigation, on the question of "whether / how this can actually lead to SHIA in the real world". This includes going back and forth between skeptically searching for flaws, and optimistically searching for workarounds/alternatives/reasons for hope. It also includes thinking about the whole process of getting to the working tech, including
What I mean is, what's been tried regarding using bacteria as persistent delivery mechanisms in the brain.
You're talking about revascularization? It's interesting, but would need a lot of fleshing out.
To step back a bit, I appreciate you thinking about these things and proposing ideas, but in order to make something actually work, I think there has to be a lot more in depth exploration. In particular, there'd have to be iterative maker/breaker investigation of the idea. In other words, I think you should argue against your own idea, then improve the idea and counterargue in favor, then critique the new version again, and repeat. Then for some ideas, you might actually convince yourself that the idea isn't that workable or promising; for other ideas, you might be able to make a more convincing case and/or put together a promising version of the project.
It's not the higher intelligence that's bad, it's the forced breeding or other dangerous experiments on much smarter animals.
Like what?
Possibly. I'm not so clear what's going on with the experiments. The one you cite that has a healthy animal getting smarter also states "Mice allografted with murine GPCs showed no enhancement of either LTP or learning.", which suggests the same wouldn't work with humans. Possibly you could do it with gene-edited human neurons / stem cells. But it feels super speculative whether that would improve much. But maybe.
Like what? Which ones plausibly significantly increase intelligence?
Note that this has significant risks and huge resulting difficulties. People generally don't want to offer crazy elective risky surgeries.
(I general I don't much buy "try it on animals and see what works" because of the issue where human brains are exceptional, and where we wouldn't have a great way of testing intelligence in animals IIUC.)
This is possibly interesting, but it would need more argument / detail. At the moment, mainly I'd view it as an interesting vector for gene editing a bunch of neurons. I'm skeptical of things like "add a bunch of BDNF" increasing intelligence much, but I could maybe be convinced otherwise.
Probably for me to want to suggest that someone fund a project on this, you'd need an expert on board, who can explain well what's been tried, what the bottlenecks are, what you're going to try that's different, why it would plausibly work, why it would be able to support the cargo, what the cargo is supposed to do, etc.