"Out of Distribution" essay
I would like brainstorming help on an essay about why it is hard for the world to think about and react sensibly to AI x-risk.
Intro in
Intro in
CI confess that one of the things I loved about Leopold's “Situational Awareness” was the delightful title and metaphor.
C----
That isn't the main thing that provokes me to plan an essay called “Out of Distribution” but I confess it was the initial spark.
That isn't the main thing that provokes me to plan an essay called “Out of Distribution” but I confess it was the initial spark.
C----
I perceive that many failure modes when folk (especially smarter folk) engage with AIXR topics can themselves be viewed as the misapplication of usually helpful trained policies and heuristics to the “out of distribution” realm of responding to x-risk, with disastrous consequences.
Quick examples...
I perceive that many failure modes when folk (especially smarter folk) engage with AIXR topics can themselves be viewed as the misapplication of usually helpful trained policies and heuristics to the “out of distribution” realm of responding to x-risk, with disastrous consequences.
Quick examples...
C----
Normalcy biases, hopium, and other results of evolutionary psychology. Readers here are mostly familiar, the public are not. We have Joep's psychology page, but if you agree there's insufficiently leveraged opportunity here, I want your insight.
Normalcy biases, hopium, and other results of evolutionary psychology. Readers here are mostly familiar, the public are not. We have Joep's psychology page, but if you agree there's insufficiently leveraged opportunity here, I want your insight.
C----
Taking an empirical try / fail / improve approach to technology and its impacts. This is basically the classic “visiting the moon safely on your first try is hard” Eliezer argument, fitting this context.
Taking an empirical try / fail / improve approach to technology and its impacts. This is basically the classic “visiting the moon safely on your first try is hard” Eliezer argument, fitting this context.
C----
This one hasn’t been discussed much, I think: that very serious people usually demand proofs and precision in dialectic and require to be convinced as an individual rather than aggregating collective wisdom across the crowd. That fine heuristic has historically assisted scientific progress, but when spotting and mitigating x-risk it is very bad.
This one hasn’t been discussed much, I think: that very serious people usually demand proofs and precision in dialectic and require to be convinced as an individual rather than aggregating collective wisdom across the crowd. That fine heuristic has historically assisted scientific progress, but when spotting and mitigating x-risk it is very bad.
C----
If some subset of that makes sense to you then bring me your ideas and your asks for clarification: I want to brainstorm and scope more before even writing the first draft of this.
Thanks!
If some subset of that makes sense to you then bring me your ideas and your asks for clarification: I want to brainstorm and scope more before even writing the first draft of this.
Thanks!
FI really like this idea, but I've struggled to precisely understand what I should do to help, or whether I'm equipped to do so.
FI guess I can ask a lot of questions:
- Who is the intended audience? (The people who tend to have these biases? The general public, to make them aware that many of their thought-leaders have these biases?)
- What is the intended format of the essay? (A simple list and explanation of various ways that intellectual people fail to acknowledge or grapple with x-risk? A narrative that smoothly ties in these ideas? A series of vignettes on each failure with a surrounding narrative structure?)
- If the goal is to change the audience's intuitions, what kind of structure does that best? What kinds of intuition pumps are most compelling?
FAs I'm trying to imagine what the end product looks like, I'm picturing an essay that is neutral in tone, but very narratively rich, and interesting to read for people that don't already agree and aren't actively looking to have their mind changed.
PIn some ways this problem is similar to the problem with climate change. It's devestating effects are nearing, there are lots of signs but we as a collective don't feel the pain yet to make it real. In the Netherlands we needed a big disaster with a raging storm and failing sea walls that resulted in thousands of deaths before we took matters truely in our own hands and redesigned basically our hole coastal defense network.
There might be a chance that AI developments provides us with the opportunity to fight back as well after a disaster and improve our defense/protocols but there is a chance that we don't get that opportunity.
There might be a chance that AI developments provides us with the opportunity to fight back as well after a disaster and improve our defense/protocols but there is a chance that we don't get that opportunity.
CNathan the questions are great. My intuitions align but keen to bash them around a bit more.
Any "the way you are thinking is probably encumbered" message is difficult to communicate with love and invites rejection.
Truly, I do want to run something of an intervention because I think that's where we are. Copium has proved very addictive. But the dynamics don't support it when it's such a minority view.
So a "this is what other people do" framing appeals. As does a personal "this is what I found/find myself doing" testimony, so long as it avoids recent acolyte evangelism overtones and self-indulgence.
Any "the way you are thinking is probably encumbered" message is difficult to communicate with love and invites rejection.
Truly, I do want to run something of an intervention because I think that's where we are. Copium has proved very addictive. But the dynamics don't support it when it's such a minority view.
So a "this is what other people do" framing appeals. As does a personal "this is what I found/find myself doing" testimony, so long as it avoids recent acolyte evangelism overtones and self-indulgence.
CYep, projects-đź’ŞPlans for warning shots / wake-up calls are definitely a thing. This essay is meant for now, though.
CArgh, it's been so long.
I did finally register the domain for the essay the other day as a forcing function on me.
I did finally register the domain for the essay the other day as a forcing function on me.
C----
And that's perhaps timely even so, since https://discord.com/channels/1100491867675709580/1362483765825503343
And that's perhaps timely even so, since https://discord.com/channels/1100491867675709580/1362483765825503343
