you could generate some artifacts like approximate HCH with this method.
Is there a strong statement you would assert with substantial confidence (IDK, like >20% or something), like “Some 30-second humans in some feasible HCH structure would be able to save the world in 5 years.”? I think this might be a crux for me as to whether connectomics-forward WBE research is net harmful. (I also wonder how a 30-second human improves over a modern AI (allowed to run for a range of times >30 seconds) in this regard.)
connectomics-based WBE methods, on the other hand, might not? (here I’m using “connectomics” as a stand-in for any data-collection mechanism that has e.g. single-neuron resolution). Or at least they’re more hopeworthy than the vibe I get from reading this post
I think connectomics-based WBE research is a somewhat safer approach compared to the approach you outlined, yeah. And also more likely to work to create a 30-second human. One issue is something like:
Either we’re talking about a full human WBE, including neural learning algorithms, or not. If we are, then this research doesn’t pay off until you get the neural learning algorithms, producing PBEs along the way.
If we’re not, we’re talking about some electrical-only WBE. This is going to be a 10-second human or something like that, probably not a 5-minute human. Can such a t-human make us safe? Far from clear.
If no, then there wasn’t much payoff.
If yes, then you’ve made a powerful artifact. This seems to fairly strongly suggest that you also make a dangerous artifact previously. For example, you made the ability to scan a human brain. So you also made the ability to scan a mouse brain. While you were scaling to a human brain, someone else could take the mouse brain, or reconstruct it using similar methods. Then they could multiply the mouse cortex by and use machine learning (since, unlike you, they are crazy and reckless) to train their weird hypermouse to be highly performant at many tasks; they can try whatever interconnectivity hyperparameters they want, biorealistic or not, and they can try any neural learning algorithms they want, because they’re crazy and don’t care about AI capabilities risk. They seem to have a lot of tools for expropriating this stuff, which you did help to make available.
mechanistic understandings of e.g. Steering Subsystem seems potentially much more differentially useful for alignment
I’m pretty skeptical of this being a significant effect, but I’d guess it’s not a crux (unless it’s a huge effect) for whether [connectomics research aiming to quickly produce a human WBE] is net harmful. WBE research would be trying to get a whole brain connectome, including cortex. (I guess you’re somewhat agreeing with that?)
Is there a strong statement you would assert with substantial confidence (IDK, like >20% or something), like “Some 30-second humans in some feasible HCH structure would be able to save the world in 5 years.”? I think this might be a crux for me as to whether connectomics-forward WBE research is net harmful. (I also wonder how a 30-second human improves over a modern AI (allowed to run for a range of times >30 seconds) in this regard.)
I think connectomics-based WBE research is a somewhat safer approach compared to the approach you outlined, yeah. And also more likely to work to create a 30-second human. One issue is something like:
Either we’re talking about a full human WBE, including neural learning algorithms, or not. If we are, then this research doesn’t pay off until you get the neural learning algorithms, producing PBEs along the way.
If we’re not, we’re talking about some electrical-only WBE. This is going to be a 10-second human or something like that, probably not a 5-minute human. Can such a t-human make us safe? Far from clear.
If no, then there wasn’t much payoff.
If yes, then you’ve made a powerful artifact. This seems to fairly strongly suggest that you also make a dangerous artifact previously. For example, you made the ability to scan a human brain. So you also made the ability to scan a mouse brain. While you were scaling to a human brain, someone else could take the mouse brain, or reconstruct it using similar methods. Then they could multiply the mouse cortex by
I’m pretty skeptical of this being a significant effect, but I’d guess it’s not a crux (unless it’s a huge effect) for whether [connectomics research aiming to quickly produce a human WBE] is net harmful. WBE research would be trying to get a whole brain connectome, including cortex. (I guess you’re somewhat agreeing with that?)