Maybe, but I didn’t say that I’d expect to see lots of projects trying to fix these issues, just that I’d expect to see more research into them, which is obviously the first step to determine correct interventions.
But you were talking about supposed deficiencies in EA modeling. Now you’re talking about the decision of which things to research and model in the first place. You’re shifting goalposts.
Voting mechanisms can be systemic if they’re approached that way. For instance, working backwards from a two party system in the US, figuring out what causes this to happen, and recommending mechanisms that fix that.
That’s no more systemic than any other way to decide how to improve how to improve voting. Changing voting mechanisms is basically working backwards from the problem of suboptimal politicians in the US, figuring out what system causes this to happen, and recommending mechanisms that fix that. Whether “figuring out” is more guided by empirical observations or by social choice theory doesn’t change the matter.
What would count as useful speculation if you think that EAs cause prioritization mechanisms are biased?
Well you can point out arguments that people are ignoring or rejecting for bad reasons, but that requires more concrete ideas instead of speculation. Maybe the lesson here is to dabble less in “speculation” and spend more time trying to make concrete progress. Show us! What’s a good cause we’ve missed?
This is another great example of EA bucking the trend, but I don’t see it as a mainstream EA cause.
Yes, because right now the only good way to approach it is to pretty much “get better at biology”—there is not enough fundamental knowledge on cognition to make dedicated progress on this specific topic. So EAs’ decisions are rational.
By the way, no other groups of “systems thinkers” are picking up on paradise engineering either.
These are certainly examples of root cause thinking, but to be truly systems thinking they have to take the next step to ask how can we shift the current system to these new foundations.
Like, uh, building institutions and advocacy for responsible AI design, and keeping them closely networked with the EA community, and spreading the idea of functional decision theory as a component of desirable AI design, with papers about FDT cooperation being published by multiple EA groups that focus on AI (MIRI and FRI)?
Consider for instance how hard it is to incorporate a feedback loop into a guesstimate model, not to mention flowthrough effects
Lol. I included “feedback loops” in arithmetic in a Word document. I had governance listed as 5% equal to the sum of other long-run policy issues, but due to the feedback loop of better governance begetting better governance, I decided to increase it to 10%. Done.
Non-systemic solution: Seeing that people are irrational, then creating an organization that teaches people to be rational.
Systemic solution: Seeing that people are irrational, asking what about the system creates irrational people, and then creating an organization that looks to change that.
Right. Let’s build kibbutzim where children are conditioned to make rational decisions. Sounds super tractable to me! Those silly EAs have been missing this low-hanging fruit the entire time.
Also, it’s not even clear how this definition of systems fits with your earlier claims that systems solutions are incorrectly less amenable to EA methodology than non-systems solutions. The concrete thing you’ve said is that EA models are worse at flow-through effects and feedback loops, which even if true (dubious) seems to apply equally well to non-systemic solutions.
I’m including systems thinking as part of my definition. This often leads to “big” interventions, but oftentimes the interventions can be small, but targeted to cause large feedback loops and flowthrough effects.
Except apparently you aren’t including poverty relief, which has large feedback loops and flowthrough effects; and apparently you aren’t including for animal advocacy, which has the same; and apparently you aren’t including EA movement growth, which has the same; and apparently you aren’t including promoting the construction of safe AGI, which has the same; and so on for everything else that EA does.
This looks very no-true-Scotsman-like.
They “have to” do that? Why?
Because they only have a hundred million dollars or so, and uh they don’t have the ability to coerce the general population? Come on.
“Hopefully” getting it to catch on elsewhere also seems silly. Perhaps they could try to look into ways to model the network effects, influence and power structures, etc, and use systems thinking to maximize their chances of getting it to catch on elsewhere
This is pedantry. Saying “hopefully” doesn’t imply that they’re not going to select the option with the highest cause for hopes. It merely implies that they don’t have control over how these things actually play out.
But you were talking about supposed deficiencies in EA modeling. Now you’re talking about the decision of which things to research and model in the first place. You’re shifting goalposts.
That’s no more systemic than any other way to decide how to improve how to improve voting. Changing voting mechanisms is basically working backwards from the problem of suboptimal politicians in the US, figuring out what system causes this to happen, and recommending mechanisms that fix that. Whether “figuring out” is more guided by empirical observations or by social choice theory doesn’t change the matter.
Well you can point out arguments that people are ignoring or rejecting for bad reasons, but that requires more concrete ideas instead of speculation. Maybe the lesson here is to dabble less in “speculation” and spend more time trying to make concrete progress. Show us! What’s a good cause we’ve missed?
Yes, because right now the only good way to approach it is to pretty much “get better at biology”—there is not enough fundamental knowledge on cognition to make dedicated progress on this specific topic. So EAs’ decisions are rational.
By the way, no other groups of “systems thinkers” are picking up on paradise engineering either.
Like, uh, building institutions and advocacy for responsible AI design, and keeping them closely networked with the EA community, and spreading the idea of functional decision theory as a component of desirable AI design, with papers about FDT cooperation being published by multiple EA groups that focus on AI (MIRI and FRI)?
Lol. I included “feedback loops” in arithmetic in a Word document. I had governance listed as 5% equal to the sum of other long-run policy issues, but due to the feedback loop of better governance begetting better governance, I decided to increase it to 10%. Done.
Right. Let’s build kibbutzim where children are conditioned to make rational decisions. Sounds super tractable to me! Those silly EAs have been missing this low-hanging fruit the entire time.
Also, it’s not even clear how this definition of systems fits with your earlier claims that systems solutions are incorrectly less amenable to EA methodology than non-systems solutions. The concrete thing you’ve said is that EA models are worse at flow-through effects and feedback loops, which even if true (dubious) seems to apply equally well to non-systemic solutions.
Except apparently you aren’t including poverty relief, which has large feedback loops and flowthrough effects; and apparently you aren’t including for animal advocacy, which has the same; and apparently you aren’t including EA movement growth, which has the same; and apparently you aren’t including promoting the construction of safe AGI, which has the same; and so on for everything else that EA does.
This looks very no-true-Scotsman-like.
Because they only have a hundred million dollars or so, and uh they don’t have the ability to coerce the general population? Come on.
This is pedantry. Saying “hopefully” doesn’t imply that they’re not going to select the option with the highest cause for hopes. It merely implies that they don’t have control over how these things actually play out.