Oh, this is nice to read as I agree that we might be able to get some reasonable enough answers about Shrimp Welfare Project vs AMF (e.g. RP’s moral weights project).
Some rough thoughts: It’s when we get to comparing Shrimp Welfare Project to AI safety PACs in the US that I think the task goes from crazy hard but worth it to maybe too gargantuan a task (although some have tried). I also think here the uncertainty is so large that it’s harder to defer to experts in the way that one can defer to GiveWell if they care about helping the world’s poorest people alive today.
But I do agree that people need a way to decide, and Anthropic staff are incredibly time poor and some of these interventions are very time sensitive if you have short timelines, so that just begs the question: if I’m recommending worldview diversification, which cause areas get attention and how do we split among them?
I am legitimately very interested in thoughtful quantitative ways of going about this (my job involves a non-zero amount of advising Anthropic folks).
Right now, it seems like Rethink Priorities is the only group doing this in public (e.g. here). To be honest, I find their work has gone over my heard, and while I don’t want to speak for them my understanding is they might be doing more in this space soon.
I [Nathan] think that shrimp QALYs and human QALYs have some exchange rate, we just don’t have a good handle on it yet.
I think being able to compare the welfare of shrimps and humans is far enough. I do not know about any interventions which robustly increase welfare in expectation due to dominant uncertain effects on soil animals. I would be curious to know your thoughts on these.
Oh, this [the point from Nathan quoted above] is nice to read as I agree that we might be able to get some reasonable enough answers about Shrimp Welfare Project vs AMF (e.g. RP’s moral weights project).
I believe there is a very long way to robust results from Rethink Priorities’ (RP’s) moral weight project, and Bob Fischer’s book about comparing welfare across species, which contains what RP stands behind now. For example, the estimate in Bob’s book for the welfare range of shrimps is 8.0 % that of humans, but I would say it would be quite reasonable for someone to have a best guess of 10^-6, the ratio between the number of neurons of shrimps and humans.
Oh, this is nice to read as I agree that we might be able to get some reasonable enough answers about Shrimp Welfare Project vs AMF (e.g. RP’s moral weights project).
Some rough thoughts: It’s when we get to comparing Shrimp Welfare Project to AI safety PACs in the US that I think the task goes from crazy hard but worth it to maybe too gargantuan a task (although some have tried). I also think here the uncertainty is so large that it’s harder to defer to experts in the way that one can defer to GiveWell if they care about helping the world’s poorest people alive today.
But I do agree that people need a way to decide, and Anthropic staff are incredibly time poor and some of these interventions are very time sensitive if you have short timelines, so that just begs the question: if I’m recommending worldview diversification, which cause areas get attention and how do we split among them?
I am legitimately very interested in thoughtful quantitative ways of going about this (my job involves a non-zero amount of advising Anthropic folks). Right now, it seems like Rethink Priorities is the only group doing this in public (e.g. here). To be honest, I find their work has gone over my heard, and while I don’t want to speak for them my understanding is they might be doing more in this space soon.
Hi Elliot and Nathan.
I think being able to compare the welfare of shrimps and humans is far enough. I do not know about any interventions which robustly increase welfare in expectation due to dominant uncertain effects on soil animals. I would be curious to know your thoughts on these.
I believe there is a very long way to robust results from Rethink Priorities’ (RP’s) moral weight project, and Bob Fischer’s book about comparing welfare across species, which contains what RP stands behind now. For example, the estimate in Bob’s book for the welfare range of shrimps is 8.0 % that of humans, but I would say it would be quite reasonable for someone to have a best guess of 10^-6, the ratio between the number of neurons of shrimps and humans.