Yeah, sounds interesting!
You don’t mention this, and maybe there is no research on it, but do we expect there to be much opportunity for resistance effects, similar to what we see with antibiotics and the evolution of resistant strains?
For example, would the deployment of large amounts of far-ultraviolet lamps result in selection pressures on microbes to become resistant to them? I think it’s not clear, since for example we don’t seem to see lots of heat resistant microbes evolving (outside of places like thermal vents) even though we regularly use high heat to kill them.
And even if it did would it be worth the tradeoff? For example, I think even if we knew about the possibility of antibiotic resistance bacteria when penicillin was created, we would still have used penicillin extensively because it was able to cure so many diseases and increase human welfare, although we might have done it with greater care about protocols and their enforcement, so with hindsight maybe we would do something similar here with far-ultraviolet light if we used it.
I think no.
History is full of plagues and other global threats of similar or worse scale. For example, I thought it could be argued that bubonic plague or smallpox were much bigger threats to humanity and individual humans than COVID 19. Yes from the inside COVID 19 feels particularly threatening, but I think that has more to do with the context in which it is happening, i.e. a world where it felt to many people like something like this couldn’t really happen. Smallpox, on the other hand, just kept killing people all the time for hundreds of years and everyone just accepted it as part of life. So on that measure COVID 19 doesn’t seem special to simulate vs. other similar types of threats humanity has faced.
Further, it’s hard to see why COVID 19 would be of interest to simulators. Presumably they would be technologically advanced enough that something like COVID 19 would not likely be interesting to learn from for some specific situation they are likely to deal with, so it would only be for historical purposes, hence I think the only relevant question is if COVID 19 is interesting enough that it would be more likely to be simulated than other past events, and I think no, so I think it offer no update to the likelihood that we are in a simulation.
There are at least two things that go by the term “nanotechnology” but are really different things: atomically precise manufacturing (e.g. Drexler, grey goo, and other stuff that is what originally went by the term “nanotech”) and nanoscale materials science (e.g. advanced modern materials science that uses various techniques, but not APM, to create materials with properties based on controlling nanoscale features of the material). Which did you have in mind? I think that will affect the kinds of answers people will give.
My impression is that many of these beggars are earning enough to survive, albeit in poverty, so your marginal dollar is probably more effective elsewhere given most people are not making the choice to give to them or not based on EA principles and others will continue to support them. If you consider local homelessness a top priority, my guess is that other interventions than small direct giving would be more effective, though I have not looked into it.
Thanks for this. I didn’t know that option value is a thing in the literature as opposed to just a common pattern in reasoning. Having handles for things is often useful, and I really appreciate it when people help bring those things in explicitly to EA, since like the rationality community I find it has a tendency to reinvent terms for existing things because of unfamiliarity with wider literature (which is not a complaint, since humans have figured out so much stuff, it’s sometimes hard to know that someone else already worked out the same ideas, especially if they did so in a different domain from the one where you are working).
Sure, this was just me taking a guess because I needed a figure to work out the numbers. I expect better analysis, if this is of interest to someone, might produce a different figure and different conclusion about cost effectiveness.
A quick scan of the article makes me want to say “more evidence needed before we can conclude much”: they ran two studies, one on 50 Stanford students, one on 400 Mechanical Turkers. Neither seems to provide very strong evidence to me about how people might make giving decisions in the real world since the study conditions feel pretty far to me to what actual giving decision feel like. Here’s the setup of the two studies from the paper:
Study 1 involves data from 50 Stanford University undergraduate students in April 2014 who made a series of binary decisions between money for charities and/or money for themselves. In addition to receiving a $20 completion fee, participants knew that one of their decisions would be randomly selected to count for payment.14 The design and results for Study 1 are detailed below (and see Online Appendix B.1 for instructions and screenshots).
Three types of charities are involved in Study 1. The first charity type involves three Make-A-Wish Foundation state chapters that vary according to their program expense rates, or percentages of their budgets spent directly on their programs and services (i.e., not spent on overhead costs): the New Hampshire chapter (90%), the Rhode Island chapter (80%), and the Maine chapter (71%).15 The second charity type involves three Knowledge Is Power Program (KIPP) charter schools that vary according to college matriculation rates among their students who completed the eighth grade: Chicago (92%), Philadelphia (74%), and Denver (61%).16 The third charity type involves three Bay Area animal shelters that vary according to their live release rates: the San Francisco SPCA (97%), the Humane Society of Silicon Valley (82%), and the San Jose Animal Care and Services (66%).
And the second one:
Study 2 involves data from 400 Amazon Mechanical Turk workers in January 2018 who made five decisions about how much money to keep for themselves or to instead donate to the Make-A-Wish Foundation.32 In addition to receiving a $1 completion fee, participants knew that one of their decisions would be randomly selected to count for payment.33 Relative to Study 1, Study 2 allows for a test of excuse-driven responses to charity performance metrics on a larger sample and via an identification strategy that does not require a normalization procedure. The design and results for Study 2 are detailed below (and see Online Appendix B.4 for instructions and screenshots).
I’ve noticed something similar around “security mindset”: Eliezer and MIRI have used the phrase to talk about a specific version of it in relation to AI safety, but the term, as far as I know, originates with Bruce Schneier and computer security, although I can’t recall MIRI publications mentioning that much, possibly because they didn’t even realize that’s where the term came from. Hard to know, a probably not very relevant other than to weirdos like us. ;-)
In the US, especially for federal elections and especially especially for election of the president, I expect voting reform to have low tractability because I believe it requires constitutional reform at the national and possibly the state level. Given how hard it is to pass amendments to the federal constitution and given that there are a lot of incentives to maintain the status quo, this seems like an uphill battle that can suck up money and generate no results.
Local election reform is probably much more tractable, especially at the municipal level, since the voting procedures are managed in ways that are more easily changed.
This makes we think of a useful perspective on this post: we still have a long way to go to spread EA within the cultures/regions where it has already taken root such that there is still a lot to be gained from doing that without dealing with the added complications of taking EA to new cultures.
I don’t have a source for previous discussions, but it’s been my impression that expansion of EA to new regions/cultures is currently intentionally conservative due to a belief that success hinges on getting it right the first time and the difficulty of crafting the EA message to resonate with a particular culture.
Ugh, I’d have to dig things up, but some things that come to mind that could be confirmed by looking that I count as evidence of this:
lag to figuring out the thing about the DES recommended magic numbers vs. when they were given out
NSA lead on public key crypto and sending agents to discourage mathematicians from publishing (this one was likely shorter because it was earlier)
lag on figuring out the problems with elliptic curve during which the NSA encouraged its use
Regarding the 14% estimate, I’m actually surprised it’s this high. I have the opposite intuition, that there is so much uncertainty, especially about whether or not any particular thing someone does will have impact, that I place the likelihood of anything any particular person working on AI safety does producing positive outcomes at <1%. The only reason it seems worth working on to me despite all of this is that when you multiply it against the size of the payoff it ends up being worthwhile anyway.
I see you mention the NSA in a footnote. One thing worth keeping in mind is that the NSA is both highly secretive and is generally believed based on past leaks and cases of “catching up” by public researchers that they are roughly 30 years ahead of publicly disclosed cryptography research. It’s possible this situation is not stable, but my best guess as an outsider is that they are a proof by example that secrecy as a strategy for maintaining a technological lead against adversaries can work, but there are likely a lot of specifics to making that work that you should probably expect any random attempt at secrecy of this sort not to be as successful as the NSA’s, i.e. the NSA is a massive outlier in this regard.
unless it’s an exceptionally good opportunity
Echoing some of the discussion in your post, I think it’s very hard for us to determine in what cases political giving impact is “an exceptionally good opportunity” due to strong biases on what we think is good and, I think importantly, given how much most people value signaling their values even if the person they vote for to send the signal fails to adequately deliver on their stated values. To me this is one of the great challenges of making political choices: many candidates stand for things you might like, but then after the fact they consistently take or approve of government action that goes against those things you stand for in the name of “compromise” to “get things done”.
I have no special beef with realpolitik—that’s just how people works—but it does make it very hard to know what the net impact of a voting choice is since it’s hard to find politicians without mixed records that sometimes contain surprises that, in the final evaluation, might swap them from net positive to net negative effect on the world.
A related notion from computer security, defense in depth.