Last nontrivial update: 2024-02-01.
Send me anonymous feedback: https://docs.google.com/forms/d/1qDWHI0ARJAJMGqhxc9FHgzHyEFp-1xneyl9hxSMzJP0/viewform
Any type of feedback is welcome, including arguments that a post/comment I wrote is net negative.
I'm interested in ways to increase the EV of the EA community by mitigating downside risks from EA related activities. Without claiming originality, I think that:
- Complex cluelessness is a common phenomenon in the domains of anthropogenic x-risks and meta-EA (due to an abundance of crucial considerations). It is often very hard to judge whether a given intervention is net-positive or net-negative.
- The EA community is made out of humans. Humans' judgement tends to be influenced by biases and self-deception. That is a serious source of risk, considering the previous point.
- Some potential mitigations involve improving some aspects of how EA funding works, e.g. with respect to conflicts of interest. Please don't interpret my interest in such mitigations as accusations of corruption etc.
Feel free to reach out by sending me a PM. I've turned off email notifications for private messages, so if you send me a time sensitive PM consider also pinging me about it via the anonymous feedback link above.
This comment was written quickly and can easily contain errors and inaccuracies.
I haven't read the post, but here's a model that may be useful:
Nationalism is not a naturally occurring phenomenon. It is a goal optimized for by NatSec elites (the people who C. Wright Mills called "warlords"). In "democracies" that have a powerful NatSec community, nationalism can help NatSec elites gain more power by legitimizing a conflict. (Conflicts can be extremely useful for NatSec elites in "democracies" for gaining more power.)
(Perhaps some researchers/leaders in AGI labs should be considered "NatSec elites" for the purpose of this comment.)
If you're happy to elaborate further, I'm curious whether you believe that is also true conditional on a single person ending up controlling the first ASI system.
My understanding is that the term "domestic terrorism" as defined in the linked page can only apply to activities that:
This does not apply to the activity in the hypothetical situation that I'm considering here.
(I am not a lawyer.)
How do totalitarian regimes compare to non-totalitarian regimes in this regard?
Notice that this definition may not apply to a hypothetical state that gives some freedoms to millions of people while mistreating 95% of humans on earth (e.g. enslaving and torturing people, using weapons of mass destruction against civilians, carrying out covert operations that cause horrible wars, enabling genocide, unjustly incarcerating people in for-profit prisons).
Who is running "EA for Jews"? I failed to find any info about that on EAforJews.org. Is it run by EA Israel? If so, I feel that that covert relationship is problematic. Israel ≠ Jews.
Disclosure/context: I'm Jewish, I live in Israel, and I've served for 3 years in IDF (mandatory service). I think there's a substantial probability that most of the ~2M people who lived in the Gaza Strip when the war started will not be alive at the end of the war, due to genocidal officers in IDF (and other factors).
(haven't read the entire post)
I think the "good people'' label is not useful here. The problem is that humans tend to act as power maximizers, and they often deceive themselves into thinking that they should do [something that will bring them more power] because of [pro-social reason].
I'm not concerned that Dario Amodei will consciously think to himself: "I'll go ahead and press this astronomically net-negative button over here because it will make me more powerful". But he can easily end up pressing such a button anyway.
[brainstorming]
It may be useful to consider the % of [worldwide net private wealth] that is lost if the US government commits to certain extremely strict AI regulation. We can call that % the "wealth impact factor of potential AI regulation" (WIFPAIR). We can expect that, other things being equal, in worlds where WIFPAIR is higher more resources are being used for anti-AI-regulation lobbying efforts (and thus EA-aligned people probably have less influence over what the US government does w.r.t. AI regulation).
The WIFPAIR can become much higher in the future, and therefore convincing the US government to establish effective AI regulation can become much harder (if it's not already virtually impossible today).
If at some future point WIFPAIR gets sufficiently high, the anti-AI-regulation efforts may become at least as intense as the anti-communist efforts in the US during the 1950s.
Thanks!
Follow up questions to anyone who may know:
Is METR (formerly ARC Evals) meant to be the "independent, external organization" that is allowed to evaluate the capabilities and safety of Anthropic's models? As of 2023-12-04 METR was spinning off from the Alignment Research Center (ARC) into their own standalone nonprofit 501(c)(3) organization, according to their website. Who is on METR's board of directors?
Note: OpenPhil seemingly recommended a total of $1,515,000 to ARC in 2022. Holden Karnofsky (co-founder and co-CEO of OpenPhil at the time, and currently a board member) is married to Daniela Amodei (co-founder of Anthropic and sibling of the CEO of Anthropic Dario Amodei) according to Wikipedia.
There's also the unilateralist's curse: suppose someone publishes an essay about a dangerous, viral idea that they misjudge to be net-positive; after 20 other people also thought about it but judged it to be net-negative.
Quoting form the linked page (from the website of The Center for Global Development):
I suppose that the claim in the parent comment that the WFP "was ranked worst of 40 largest aid agencies" is based on that "data" spreadsheet. But notice that 27 of those 40 "aid agencies" are not aid agencies but rather countries (e.g. Australia, United States). So this is already a big red flag. For each agency/country, the spreadsheet provides 7 indicators that are grouped under the title "Maximising Efficiency". One of those indicators is called "ME4" according to which the WFP performed 3rd worst among all 40 agencies/countries. Quoting from the "detailed methodology" PDF (removing footnote references):
So if I understand correctly, the the QuODA scale seems to "punish" agencies that spend money on food assistance directly (rather than giving the money to the host state), and therefore does not seem like a good scale for evaluating the World Food Programme. (To be clear, I'm not overall familiar with QuODA scale; I'm just reporting what seems to me like a very big red flag).
Maybe the WFP employs many locals in low-GPD-per-capita states as part of their efforts to distribute food and the salaries are not a large % of WFP's budget? (I don't know whether that is the case; I'm just pointing out that that metric does not seem useful here.)