I might be missing something, but the “soloware” project seems very likely to escalate misuse of AI by human beings and make things far less transparent (b/c now everything is custom-made and it’s harder to track the cause-effect chains of how anything was created, and imo we want a drastic increase in accountability for every action taken by human and machine).
It’s going to take all the bad parts of individualist capitalist patterns / competition-seeking and turn the dial up, which is basically what happens anytime anyone applies collective-optimization pressure to a technology.
I see a lot of risk and am unclear on the benefits.
But let me try to articulate why one might see soloware as a good idea…
If I believed that the issue with machines was this emphasis on ‘replacing humans’ by automating stuff, rather than empowering humans, then I might get what was better about this approach. <-- This is VERY succinct. But hopefully enough.
However: There’s nothing inherently ‘robust’ about human beings’ alignment, at any scale. The alignment issue isn’t just about artificial intelligences; it extends to human intelligence also, and in fact extends to just “intelligence” generally. Intelligence doesn’t have any way to align itself, at any scale. (It needs ‘wisdom’ which is in some sense the “opposite” of intelligence.)
You seem to point at this yourself with point #2 RE: humans and indifference risks. The problem isn’t rooted in the AI’s agency; it’s also in the human’s own agency. You won’t solve one without solving both.
So I guess it raises a Question:
Does one believe human intelligence or empowerment is somehow anchored to an alignment solution in some special way?
FTR, I agree with the points made about the chatbot interface being problematic, for the reasons expressed. But my solutional direction would not be soloware, at least not as I understand or am seeing it implemented (i.e. by inviting some ppl to start trying it out more).
My solutional direction might be something I’d call “Sacredware.”
In all likelihood, Sacredware would be designed by a small group of “only-corrigible-to-the-inconceivable” (Bodhi-corrigible) for “Everyone”. This might end up looking like soloware, or something similarly flexible. And yet, it would also need to, at scale, rigorously avoid red queen dynamics, tragedy of the commons, and unnecessary killing of life. It would not be designed to cater to human preferences or comfort, but rather their collective spiritual development and ability to steward life on Earth. (In other words, it constantly affirms or upholds a set of virtues that transcends human societies or particular time&place, without major deviation, while being extremely adaptable for specific time&place.)
It would not be possible to get to Sacredware by starting with the soloware project, as far as I understand it. But that is difficult for me to explain here.
I might be missing something, but the “soloware” project seems very likely to escalate misuse of AI by human beings and make things far less transparent (b/c now everything is custom-made and it’s harder to track the cause-effect chains of how anything was created, and imo we want a drastic increase in accountability for every action taken by human and machine).
It’s going to take all the bad parts of individualist capitalist patterns / competition-seeking and turn the dial up, which is basically what happens anytime anyone applies collective-optimization pressure to a technology.
I see a lot of risk and am unclear on the benefits.
But let me try to articulate why one might see soloware as a good idea…
If I believed that the issue with machines was this emphasis on ‘replacing humans’ by automating stuff, rather than empowering humans, then I might get what was better about this approach. <-- This is VERY succinct. But hopefully enough.
However:
There’s nothing inherently ‘robust’ about human beings’ alignment, at any scale. The alignment issue isn’t just about artificial intelligences; it extends to human intelligence also, and in fact extends to just “intelligence” generally. Intelligence doesn’t have any way to align itself, at any scale. (It needs ‘wisdom’ which is in some sense the “opposite” of intelligence.)
You seem to point at this yourself with point #2 RE: humans and indifference risks. The problem isn’t rooted in the AI’s agency; it’s also in the human’s own agency. You won’t solve one without solving both.
So I guess it raises a Question:
Does one believe human intelligence or empowerment is somehow anchored to an alignment solution in some special way?
FTR, I agree with the points made about the chatbot interface being problematic, for the reasons expressed. But my solutional direction would not be soloware, at least not as I understand or am seeing it implemented (i.e. by inviting some ppl to start trying it out more).
My solutional direction might be something I’d call “Sacredware.”
In all likelihood, Sacredware would be designed by a small group of “only-corrigible-to-the-inconceivable” (Bodhi-corrigible) for “Everyone”. This might end up looking like soloware, or something similarly flexible. And yet, it would also need to, at scale, rigorously avoid red queen dynamics, tragedy of the commons, and unnecessary killing of life. It would not be designed to cater to human preferences or comfort, but rather their collective spiritual development and ability to steward life on Earth. (In other words, it constantly affirms or upholds a set of virtues that transcends human societies or particular time&place, without major deviation, while being extremely adaptable for specific time&place.)
It would not be possible to get to Sacredware by starting with the soloware project, as far as I understand it. But that is difficult for me to explain here.