I liked this post and would like to see more of people thinking for themselves about cause prioritization and doing BOTECs.
Some scattered thoughts below, also in the spirit of draft amnesty.
I had a little trouble understanding your calculations/logic, so I’m going to write them out in sentence form: GiveWell’s current giving recommendations correspond to spending about $0.50 to save an additional person an additional year of life. A 10% chance of extinction from misaligned AI means that postponing misaligned AI by a year gets us 10%*current population number of person days, or about 10 million. If we take GiveWell’s willingness to spend and extrapolate it to the scenario of postponing misaligned AI, we get that GiveWell might be willing to spend $500 million to postpone misaligned AI by a day.
I think it’s important that these are different domains, and the number of people who would be just as happy to see their donation buy a bednet as lobby for tech regulation (assuming similar EV) is unfortunately small. Many donors care about much more than some cause-neutral how-much-good their donation does. e.g., for instance I see to care (I’m confiused) that some of my donations help extremely poor people.
You point out that maybe regulation doesn’t work, but there’s the broader problem which is that we don’t have shovel ready projects that can turn $500 million into postponing misaligned AI by a day; I suspect there are many interventions which can do this for much cheaper, but they are not interventions which can just absorb money and save lived like many global health charities can (perhaps they need projects to be founded and take years to develop).
The above problems point to another important idea: The GiveWell bar is able to be where it is because of what projects to improve the world can actually be funded by GiveWell dollars — not because of some fact about the value of postponing a life by a day. You might think about the GiveWell bar as, the cheapest scalable ways to save lives in the global health and development space can provide an additional day of life for $0.50. If you ask individuals in the US how much they would pay to extend their own life or that of a loved one by a day, you will get numbers much higher than this; if you look at spending on healthcare in the developed world my guess is that it is very normal to spend thousands of dollars to postpone a life by a day. GiveWell’s bar for funding would be higher if there were other great opportunities for saving lives salably for cheap (at least in global health and development).
An abstraction I notice I’m using is thinking about $0.50/person/day as the current market price. However, this is not an efficient market, for a number of reasons. This post draws the parallel of “hey look, at that price we should be willing to spend $0.5b on postponing misaligned AI by a day”. However, if we actually had many opportunities to spend $0.5b on postponing misaligned AI by a day, the funding bar would increase, because there isn’t enough money in the cause-neutral altruism bucket.
Some implications: cause-neutral donors who put above negligible probability on existential risks from AI will probably get much more bang for their buck trying to reduce existential risks or buy time, at least contingent on there being projects that can absorb money in that space. More importantly, those working on reducing AI x-risk have a lot of work to do in terms of closing the cost-effectiveness gap between themselves and global health. By closing the gap I mean getting enough projects to exist in the space such that we can consistently take in more money and turn it into x-risk reduction or buying time.
I liked this post and would like to see more of people thinking for themselves about cause prioritization and doing BOTECs.
Some scattered thoughts below, also in the spirit of draft amnesty.
I had a little trouble understanding your calculations/logic, so I’m going to write them out in sentence form: GiveWell’s current giving recommendations correspond to spending about $0.50 to save an additional person an additional year of life. A 10% chance of extinction from misaligned AI means that postponing misaligned AI by a year gets us 10%*current population number of person days, or about 10 million. If we take GiveWell’s willingness to spend and extrapolate it to the scenario of postponing misaligned AI, we get that GiveWell might be willing to spend $500 million to postpone misaligned AI by a day.
I think it’s important that these are different domains, and the number of people who would be just as happy to see their donation buy a bednet as lobby for tech regulation (assuming similar EV) is unfortunately small. Many donors care about much more than some cause-neutral how-much-good their donation does. e.g., for instance I see to care (I’m confiused) that some of my donations help extremely poor people.
You point out that maybe regulation doesn’t work, but there’s the broader problem which is that we don’t have shovel ready projects that can turn $500 million into postponing misaligned AI by a day; I suspect there are many interventions which can do this for much cheaper, but they are not interventions which can just absorb money and save lived like many global health charities can (perhaps they need projects to be founded and take years to develop).
The above problems point to another important idea: The GiveWell bar is able to be where it is because of what projects to improve the world can actually be funded by GiveWell dollars — not because of some fact about the value of postponing a life by a day. You might think about the GiveWell bar as, the cheapest scalable ways to save lives in the global health and development space can provide an additional day of life for $0.50. If you ask individuals in the US how much they would pay to extend their own life or that of a loved one by a day, you will get numbers much higher than this; if you look at spending on healthcare in the developed world my guess is that it is very normal to spend thousands of dollars to postpone a life by a day. GiveWell’s bar for funding would be higher if there were other great opportunities for saving lives salably for cheap (at least in global health and development).
An abstraction I notice I’m using is thinking about $0.50/person/day as the current market price. However, this is not an efficient market, for a number of reasons. This post draws the parallel of “hey look, at that price we should be willing to spend $0.5b on postponing misaligned AI by a day”. However, if we actually had many opportunities to spend $0.5b on postponing misaligned AI by a day, the funding bar would increase, because there isn’t enough money in the cause-neutral altruism bucket.
Some implications: cause-neutral donors who put above negligible probability on existential risks from AI will probably get much more bang for their buck trying to reduce existential risks or buy time, at least contingent on there being projects that can absorb money in that space. More importantly, those working on reducing AI x-risk have a lot of work to do in terms of closing the cost-effectiveness gap between themselves and global health. By closing the gap I mean getting enough projects to exist in the space such that we can consistently take in more money and turn it into x-risk reduction or buying time.
If you haven’t read Astronomical Waste, you might like it.