13964 karmaJoined Nov 2022Working (15+ years)


I am an attorney in a public-sector position not associated with EA, although I cannot provide legal advice to anyone. My involvement with EA so far has been mostly limited so far to writing checks to GiveWell and other effective charities in the Global Health space, as well as some independent reading. I have occasionally read the forum and was looking for ideas for year-end giving when the whole FTX business exploded . . . 

How I can help others

As someone who isn't deep in EA culture (at least at the time of writing), I may be able to offer a perspective on how the broader group of people with sympathies toward EA ideas might react to certain things. I'll probably make some errors that would be obvious to other people, but sometimes a fresh set of eyes can help bring a different perspective.


Sorted by New
· 1y ago · 1m read


Topic contributions

I wonder if anyone else will getting a thinly veiled counterpart -- given that the lead character of the show seems somewhat based on MacKenzie Scott, this seems to be maybe a thing for the show.

At least often. I don't know if would be safe to assume that there will ~always be convergence -- for instance, switching to higher-welfare breeds that grow less efficiently and less rapidly might plausibly not be net positive from a climate perspective. Cultured meat might not be if its production were extremely energy intensive.

Yeah, I didn't vote either way, which is fine. I'm just confused about how to interpret the votes of those who did! Did they agree/disagree on both parts, or vote based on which part they thought was primary?

That's fair.

Broader concerns about sovereignty and the integrity of the extradition system likely played a role. Although it's understandable why US prosecutors asked for provisional arrest and extradition despite not having obtained an indictment for campaign-finance violations yet, it's also understandable to me why the Bahamas wants to signal its expectation that the US comply with the treaty as written -- especially where the argument that the first extradition was good enough for these charges is legally weak.

Note that the US could have asked for permission to proceed on other charges -- see Article 14(b) of the treaty -- but I think that may have required going back through the Bahamas legal system again, and probably delaying the trial. Hence, the prosecution severed off those charges, and then decided not to proceed after getting its conviction on others. I don't know Bahamas campaign or extradition law, but there would also be difficulties if the offense was deemed one of a political character (Article 3(1)(a)) or was one that was not a criminal offense punishable by more than a year in prison under Bahamas law (Article 2(1)).

Also, I don't think "completely uninvestigated" is a correct characterization -- they were investigated enough to be presented to a grand jury, which indicted SBF for campaign-finance violations. Federal prosecutors do not generally indict without a pretty good investigation first, especially in high-profile cases. I think we have a pretty decent idea of what he did (see pp. 18-22 of the prosecution's sentencing memo). Moreover, Salame and Singh -- who don't have extradition-related issues -- pled guilty to campaign-finance violations.

One problem here is that a solution that operates continuously and provides "value way before an extinction level event plays out" is likely to be much more complex -- and have considerably more points of potential failure -- than a simpler solution.

Imagine I have a requirement that I be able to receive radio transmissions during an emergency. I could build a fairly simple emergency radio and pressure-test it. It doesn't provide me any value in ordinary times, so it just sits there on the shelf between pressure tests. It's cheap, so I could build several for redundancy. I could also build a smartphone with radio capacity -- this would be operated continuously and provide value in normal times. It is also much more complex and has many more failure modes than a simple radio. Many of those failure modes are common to both the ordinary-times use and the emergency-times use. But some are less so -- I might not be using the radio feature on my cellphone much at all in normal times because that particular function doesn't provide much if any pre-emergency value.

I think you're likely accepting a significantly higher level of failure modes for a better opportunity to detect and fix them. It's not clear to me if this is a good tradeoff or not. If you go back to the cellphone analogy, most things have to evolve and grow to continue providing enough value to justify their maintenance and operation. Maybe we could have a super-high-reliability smartphone if we locked in 2010 capabilities and worked on iterating the flaws out of it. But a 2010-level smartphone isn't going to get much use. And adding new features and capabilities adds new points of failure, so manufacturers are stuck playing a never-ending game of whack-a-mole.

I'm reminded of the troubles that the US government has had with getting cell phones up to the point that they can handle classified information. The poor user experience of cell phones approved for even secret information famously contributed to getting a former Secretary of State into some trouble. As of 2016, phones appeared to have only been approved to the secret level (not top secret, or ~"above top secret" like certain SCIs). In contrast, we've had landline devices cleared for all security levels since at least 1987 (I didn't check older devices). The security reliability risks are apparently a lot easier to manage with the simpler devices.

For similar reasons, the solution with pre-emergency value will likely be much more expensive to develop and operate. All critical components need to be hardened / highly reliable. The more you expect those components to do, the bigger / more complex they are going to be. If the power facility needs to power a small town, it will be much more expensive to harden than a solution for a bunker. If you rely on a different power solution for emergency situations and only harden that solution, then a critical system isn't going to be regularly operated. It's probably going to be considerably more expensive to run and maintain a hardened system than a standard one even in normal times -- so normal-time operation is likely to also require some significant subsidies. As a result, you're going to have to put a lot of eggs in a single basket for cost reasons alone.

- so I'm saying it should not only be pressure tested but be in continuous operation in order to flush out failure modes before a catastrophic scenario plays out, it needs to be providing value way before an extinction level event plays out. 

This seems to rely on an assumption that the failure modes that would exist in "normal mode" are related or correlated, to a fairly high degree, to the failure modes that could show up in "catastrophic mode." That's not obvious to me.

Grassroots projects like this seem like a natural extension of this, where a community as a whole decides where they need resources in order to uplift everyone. 

This makes sense, but I don't think it gets us very far on the question of whether to fund this particular project. 

There are many grassroots projects in developing countries, and it is often difficult for a Westerner to evaluate the effectiveness of those projects (at least where their theory of impact is more complex than bednets --> less malaria). It's even difficult for us to assess the relative extent of informed public support for a grassroots project from afar. Those issues are less formidable with GiveDirectly; if the main theory of impact is to benefit Person X, we have both theoretical reasons and experiential evidence that giving Person X money is a good way to accomplish that. That doesn't necessarily scale well to supporting grassroots work.

That being said, I can see a decent theoretical argument for (as it were) GiveDirectly for Communities -- give a community a certain amount of money, and let the community decide what needs that money should go to. I can see a number of practical problems with that, though. I think your average Westerner is going to be considerably worse at evaluating projects valued at ~50 times GDP per capita than in making their own consumption/investment decisions, and I suspect that may be true in many places. The quality of many decisions made by various democratic political systems also gives me pause. So I think there would need to be evaluation and selection of proposals rather than the total deference of the GiveDirectly approach.

Ideally, we would have evaluation organizations that were more local to the populations that were being served, rather than having the big GH/D evaluator be in the United States. That should give us evaluators with more local knowledge, and (to be honest) those with a cost structure and business processes that would make evaluating five-to-six figure projects more feasible.

On the one hand, I am sympathetic to the differential scrutiny applied to different projects.

On the other hand, there are conceptual models in which this scrutiny makes sense. In one of them, money has been pre-committed by donors ex ante to cause areas. So money in the global health/development (GH/D) bucket (or maybe the neartermism bucket) is only competing for funding against projects in the same bucket (while the HPMOR and Wytham projects were in something like a "longtermism community growth" (LCG) bucket).

As a practical matter, this is approximately true -- in the context of an appeal on the Forum for funds, Anthony's project is very likely to receive funds that would counterfactually have gone to other GH/D work (and there's a good chance they would have counterfactually gone to GiveWell-style work). In contrast, the odds of someone like me giving money to HP fanfic distribution or Wytham are ~0.

Relatedly, one could see the GH/D and LCG fields as too methodologically different for this kind of comparison to be fitting. There's a certain appeal to that, since the numbers in LCG cost-effectiveness analyses tend to be much less grounded in data than GH/D numbers. And if you make the "number of future lives potentially saved" high enough, the LCG project will always win even if the "chance of preventing catastrophe" is miniscule indeed. Of course, one extension to this approach would be to argue that (e.g.) EA-style GH/D projects and non-EA-style GH/D projects are also too methodologically different for this kind of analysis -- and that would leave us with no common yardstick to evaluate projects in GH/D itself. 

This comment shows the challenge of the agreevote/disagreevote system:

  •  I agree with the direction toward showing more cost-effectiveness analyses in other fields versus reducing their importance in global health/development. 
  • I do not think the fanfic CEA is plausible, even at the 2-minute level, for some of the reasons identified by @titotal. That being said, the people funding fanfic distribution were probably interested in "people drawn to AI safety / x-risk mitigation" as their outcome variable / theory of impact, not donations made or individual lives saved. So the BOTEC is simultaneous too kind to and too demanding of that project.

Strong upvote for a community member taking the time to evaluate an intervention presented by an "outsider," act on that evaluation, and share it with others. This adds a lot of value!

Load more