What's the GiveDirectly of longtermism & existential risk?

This is probably dumb, but I wonder if there are precious metal bottlenecks (like U-235 for nukes and platinum for cars) for specialized deep learning hardware/GPUs/computers that you can buy mines etc of (and then shut down) to slow down AI progress slightly. 

I highly doubt this is the most leveraged thing you can do, but this seems like it can scale to arbitrary amounts of dollars invested, certainly much more money than EA currently has access to.

It also makes more sense in a worldview where AI is "the only game in town" and less for a worldview where you think e.g. climate or bio or nukes or broad longtermism is >0.1x as important as AI.

Patient Philanthropy Fund 

Edited* 

Like GD it gives agency to the people who are suffering to make their own decisions. It has small overheads and could take far larger sums than most other spaces.

Unlike GiveDirectly, which is solving the problem in itself, the PPF kicks the can down the road. 

 *This was originally a placeholder - I was going to bed so said that someone should else write this as a proper answer. But everyone upvoted my placeholder anyway. [Thanos voice] "Fine, I'll do it myself"

Genomic mass screening of wastewater for unknown pathogens, as described here:

[2108.02678] A Global Nucleic Acid Observatory for Biodefense and Planetary Health (arxiv.org)

A few test sites can already help to detect a new (natural or manmade) pandemic at an early stage. Nevertheless, there is room for a few billion dollars if you want to build a global screening network.

Unfortunately, I do not know if there is any organisation with need for funding working on this. 

Here's Will MacAskill at EAG 2020:

I’ve started to view [working on climate change] as the GiveDirectly of longtermist interventions. It's a fairly safe option.

Command-f for the full context on this.

Vaccines

 (Originally suggested by Ben Todd)

@CEPIvaccines a $3.5bn program to develop vaccines to stop the next pandemic

Cons

• a cause area in itself

Some relatively measurable, concrete megaprojects that might help the long-term future (super-high cost effectiveness not necessarily guaranteed):

• Backing up civilization by constructing giant underground bunkers and the like.

• Backing up civilization by colonizing Mars or other space efforts. (So far, SpaceX has been very successful as a for-profit business, but maybe there are aspects of space colonization where philanthropy could help. I also think a serious bunker-building operation, with higher standards than most bunker construction today, could potentially be run as a profitable business selling to governments, prepper billionaires, etc.)

To the extent that long-run economic growth is a longtermist goal and isn't totally overshadowed by X-risk, there are lots of ways we might encourage a faster pace of economic development:

• Lobbying for generally neoliberal pro-growth policies, including things like free trade, immigration, lighter regulation especially in cost-diseased sectors like healthcare and housing, etc. (Although like all politics, this is controversial and does not have the "everyone agrees it is helping" property of Givewell charities.)

• Experimenting with promising new economic systems like quadratic funding of public goods, Georgist land taxes, etc. Trying to improve governance, social technology, and institutional decision-making by supporting things like prediction markets and charter cities. (Although experimentation is necessarily a "hits-based" business, thus doesn't have the guaranteed-impact aspect of Givewell charities).

• Trying to advance scientific fields either by directly funding promising areas, or by trying to improve the functioning of academia and science grantmakers as institutions. (Although the opportunities here are probably less scalable than elsewhere.)

It would be great if we had some way of putting money towards reducing "existential risk factors" -- if improving US-China relations was as straightforward as buying carbon offset credits, I think that would attract a hell of a lot of funding.

Paying AI researchers to do slightly useful alignment research (or even nothing) rather than advancing capabilities.

This easily scales up to the low tens of billions per year, at which point it turns into acquiring startups. In the low hundreds of billions per year, one could even acquire significant stakes in Nvidia/Google/etc., but this seems horrendously inefficient.

GiveDirectly is about giving resources to other present people, even though they will not use it in a very targeted manner. The obvious analogy for the future is to save/invest money, which very slightly accelerates economic growth and transfers resources to future people, even if not in a very targeted manner. 

It's not a great option; we should be able to do a lot better. But it does seem roughly equivalent to GiveDirectly, which is also not a great option.

Downvote me to deprive Nathan of karma.