Nice, thanks for sharing, I’ll actually give you a different answer than last time after thinking about this a bit more (and maybe understanding your questions better). :)
> Would you still be clueless if the vast majority of the posteriorcounterfactual effect of our actions (e.g. in terms of increasing expected total hedonistic utility) was realised in at most a few decades to a century? Maybe this is the case based on the quickly decaying effect size of interventions whose effects can be more easily measured, likes ones in global health and development?
Not sure that’s what you meant, but I don’t think the effects of these decay in the sense that they have big short-term impact and negligible longterm impact (this is known as the “ripple in a pond” objection to cluelessness [1]). I think their longterm impact is substantial but that we just have no clue if it’s good or bad because that depends on so many longterm factors the people carrying out these short-term interventions ignore and/or can’t possibly estimate in an informative non-arbitrary way.
So I don’t know how to respond to your first question because it seems it implictly assumes something I find impossible and goes against how causality works in our complex World (?)
> Do you think global human wellbeing has been increasing in the last few decades? If so, would you agree past actions have generally been good considering just a time horizon of a few decades after such actions? One could still argue past actions had positive effects over a few decades (i.e. welfare a few decades after the actions would be lower without such actions), but negative and significant longterm effects, such that it is unclear whether they were good overall.
Answering the second question: 1. Yes, one could argue that. 2. One could also argue we’re wrong to assume human wellbeing has been improving to begin with. Maybe we have a very flawed definition of what wellbeing is, which seems likely given how much people disagree on what kinds of wellbeing matter. Maybe we’re neglecting a crucial consideration such as “there have been more people with cluster headaches with the population increasing and these are so bad that they outweigh all the good stuff”. Maybe we’re totally missing a similar kind of crucial consideration I can’t think of. 3. Maybe most importantly, in the real World outside of this thought experiment, I don’t care only about humans. If I cared only about them, I’d be less clueless because I could ignore humans’ impact on aliens and other non-humans.
And to develop on 1:
> Do we have examples where the posterior counterfactual effects was positive at 1st, but then became negative instead of decaying to 0?
- Some AI behaved very well at first and did great things and then there’s some distributional shift and it does bad things. - Technological development arguably improved everyone’s life at first and then it caused things like the confection of torture instruments and widespread animal farming. - Humans were incidentally reducing wild animal suffering by deforesting but then they started becoming environmentalists and rewilding. - Alice’s life seemed wonderful at first but she eventually came down with severe chronic mental illness. - Some pill helped people like Alice at first but then made their lives worse. - The Smokey Bear campaign reduced wildfires at first and then it turned out it increased them.
[1] See e.g. James Lenman’s and Hilary Greaves’ work on cluelessness for rejections of this argument.
big short-term impact and negligible longterm impact
If these were not so for global health and development interventions, I would expect to see interventions whose posterior effect size increases as time goes by, whereas this is not observed as far as I know.
2. One could also argue we’re wrong to assume human wellbeing has been improving to begin with. Maybe we have a very flawed definition of what wellbeing is, which seems likely given how much people disagree on what kinds of wellbeing matter. Maybe we’re neglecting a crucial consideration such as “there have been more people with cluster headaches with the population increasing and these are so bad that they outweigh all the good stuff”. Maybe we’re totally missing a similar kind of crucial consideration I can’t think of.
I think welfare per human-year has increased in the last few hundred years. However, even if one is clueless about that, one could still conclude human welfare has increased due to population growth, as long as one agrees humans have positive lives?
3. Maybe most importantly, in the real World outside of this thought experiment, I don’t care only about humans. If I cared only about them, I’d be less clueless because I could ignore humans’ impact on aliens and other non-humans.
I agree there is lots of uncertainty about whether wild and farmed animals have positive or negative lives, and about the impact of humans on animal and alien welfare. However, I think there are still robustly positive interventions, like Shrimp Welfare Project’s Humane Slaughter Initiative, which I estimate is way more cost-effective than GiveWell’s top charities, and arguably barely changes the number of farmed and wild animals. I understand improved slaughter will tend to increase the cost of shrimp, and therefore decrease the consumption of shrimp, which could be bad if shrimp have positive lives, but I think the increase in welfare from the less painful slaughter is the driver of the overall effect.
- Some AI behaved very well at first and did great things and then there’s some distributional shift and it does bad things.
Not all AI development is good, but I would say it has generally been good at least so far and for humans.
- Technological development arguably improved everyone’s life at first and then it caused things like the confection of torture instruments and widespread animal farming.
Fair. However, cluelessness about whether technological development has been good/bad does not imply cluelessness about what to do, which is what matters. For example, one could abstain from supporting technological development more closely linked to wars and factory-farming if one does not think it has generally been beneficial in those areas.
- Humans were incidentally reducing wild animal suffering by deforesting but then they started becoming environmentalists and rewilding.
I think it is very unclear whether wild animals have positive/negative lives, so I would focus on efforts trying to improve their lives instead of increasing/decreasing the number of lives.
- Alice’s life seemed wonderful at first but she eventually came down with severe chronic mental illness.
I agree there are many examples where the welfare of a human decreases. However, we are far from clueless about improving human welfare. Even if welfare per human-year has not been increasing, welfare per human life has been increasing due to increases in life expectancy.
- Some pill helped people like Alice at first but then made their lives worse.
There are always counterexamples, but I suppose taking pills recommended by doctors still improves welfare in expectation (although I guess less than people imagine).
- The Smokey Bear campaign reduced wildfires at first and then it turned out it increased them.
It is unclear to me whether this interventions was positive at 1st, becaues I do not know whether wild animals have positive or negative lives, and I expect the effects on these are the major driver of the overall effect.
Nice, thanks for sharing, I’ll actually give you a different answer than last time after thinking about this a bit more (and maybe understanding your questions better). :)
> Would you still be clueless if the vast majority of the posterior counterfactual effect of our actions (e.g. in terms of increasing expected total hedonistic utility) was realised in at most a few decades to a century? Maybe this is the case based on the quickly decaying effect size of interventions whose effects can be more easily measured, likes ones in global health and development?
Not sure that’s what you meant, but I don’t think the effects of these decay in the sense that they have big short-term impact and negligible longterm impact (this is known as the “ripple in a pond” objection to cluelessness [1]). I think their longterm impact is substantial but that we just have no clue if it’s good or bad because that depends on so many longterm factors the people carrying out these short-term interventions ignore and/or can’t possibly estimate in an informative non-arbitrary way.
So I don’t know how to respond to your first question because it seems it implictly assumes something I find impossible and goes against how causality works in our complex World (?)
> Do you think global human wellbeing has been increasing in the last few decades? If so, would you agree past actions have generally been good considering just a time horizon of a few decades after such actions? One could still argue past actions had positive effects over a few decades (i.e. welfare a few decades after the actions would be lower without such actions), but negative and significant longterm effects, such that it is unclear whether they were good overall.
Answering the second question:
1. Yes, one could argue that.
2. One could also argue we’re wrong to assume human wellbeing has been improving to begin with. Maybe we have a very flawed definition of what wellbeing is, which seems likely given how much people disagree on what kinds of wellbeing matter. Maybe we’re neglecting a crucial consideration such as “there have been more people with cluster headaches with the population increasing and these are so bad that they outweigh all the good stuff”. Maybe we’re totally missing a similar kind of crucial consideration I can’t think of.
3. Maybe most importantly, in the real World outside of this thought experiment, I don’t care only about humans. If I cared only about them, I’d be less clueless because I could ignore humans’ impact on aliens and other non-humans.
And to develop on 1:
> Do we have examples where the posterior counterfactual effects was positive at 1st, but then became negative instead of decaying to 0?
- Some AI behaved very well at first and did great things and then there’s some distributional shift and it does bad things.
- Technological development arguably improved everyone’s life at first and then it caused things like the confection of torture instruments and widespread animal farming.
- Humans were incidentally reducing wild animal suffering by deforesting but then they started becoming environmentalists and rewilding.
- Alice’s life seemed wonderful at first but she eventually came down with severe chronic mental illness.
- Some pill helped people like Alice at first but then made their lives worse.
- The Smokey Bear campaign reduced wildfires at first and then it turned out it increased them.
[1] See e.g. James Lenman’s and Hilary Greaves’ work on cluelessness for rejections of this argument.
Thanks for following up, Jim.
If these were not so for global health and development interventions, I would expect to see interventions whose posterior effect size increases as time goes by, whereas this is not observed as far as I know.
I think welfare per human-year has increased in the last few hundred years. However, even if one is clueless about that, one could still conclude human welfare has increased due to population growth, as long as one agrees humans have positive lives?
I agree there is lots of uncertainty about whether wild and farmed animals have positive or negative lives, and about the impact of humans on animal and alien welfare. However, I think there are still robustly positive interventions, like Shrimp Welfare Project’s Humane Slaughter Initiative, which I estimate is way more cost-effective than GiveWell’s top charities, and arguably barely changes the number of farmed and wild animals. I understand improved slaughter will tend to increase the cost of shrimp, and therefore decrease the consumption of shrimp, which could be bad if shrimp have positive lives, but I think the increase in welfare from the less painful slaughter is the driver of the overall effect.
Not all AI development is good, but I would say it has generally been good at least so far and for humans.
Fair. However, cluelessness about whether technological development has been good/bad does not imply cluelessness about what to do, which is what matters. For example, one could abstain from supporting technological development more closely linked to wars and factory-farming if one does not think it has generally been beneficial in those areas.
I think it is very unclear whether wild animals have positive/negative lives, so I would focus on efforts trying to improve their lives instead of increasing/decreasing the number of lives.
I agree there are many examples where the welfare of a human decreases. However, we are far from clueless about improving human welfare. Even if welfare per human-year has not been increasing, welfare per human life has been increasing due to increases in life expectancy.
There are always counterexamples, but I suppose taking pills recommended by doctors still improves welfare in expectation (although I guess less than people imagine).
It is unclear to me whether this interventions was positive at 1st, becaues I do not know whether wild animals have positive or negative lives, and I expect the effects on these are the major driver of the overall effect.