This post is far less cogent and complete than the dozens of well-footnoted, (nearly-) GiveWell quality analyses. After some initial research and writing a few weeks back, I stalled, unsure about the next appropriate step, and eventually lost motivation.
What I’m left with is a haphazard collection of notes that I’m cleaning up only slightly in the final hours of the competition Some parts are decent, some are practically scribbles, and most fall in-between.
Epistemic status: ⅓ real cause area investigation, ⅔ haphazard collection of notes (but notes collected in earnest), still stand by everything.
Why I think it’s worth sharing now
This “map” is pretty meager as a report and research product, not the kind of thing I’d like handing in as a final project in an EA college course, but the map is not the territory, and the territory is worth looking at!
Although my failure to investigate, document, and cite more than I have renders the cause area’s true promisingness high variance, my current sense is that wild fish slaughter:
Deserves more total resources through some channel than it currently receives or has received over the last few years (90%)
Deserves more resources from OpenPhil than it is currently getting (75%)
Is more deserving of marginal dollars than any GiveWell top charity, conditional on accepting the in-principle validity of such a comparison (95%)
Is more deserving of marginal resources than any of the big existing “EA approved” farmed animal welfare interventions like chicken cage-free campaigns, cultured meat development, and improving farmed fish aquaculture conditions (65%)
Is the single, non-longtermist,[1] object level large cause area most deserving of “EA resources” (20%)
It’s more likely that someone from OpenPhil will read and consider the points below if I post it as a competition entry rather than a miscellaneous forum post in a few weeks, and further that this is morally good on net (even accounting for various opportunity costs)
As a final bit of preamble, my intention here isn’t to make suffering porn, but I also don’t want to dance around the issue by using terms that are misleadingly benign. Not sure I strike this balance well, but I urge you not to stop reading (if you’ve made it this far) just because what I’m describing is bad.
Importance
Section summary
I estimate that wild fish slaughter accounts for 68 million years of extreme fish suffering (given that fish are sentient) per year. Further, using neuron count as a proxy for capacity to suffer, I estimate this is the moral equivalent of 680,000 years of human torture, which in turn corresponds to (with very large uncertainty) the moral value of 1.58 billion human deaths per year.
BOTEC: Moral value of wild fish slaughter (annual)
Note: Each figure is supposed to represent a sort of conservative-but-realistic estimate, so maybe something around the 25th percentile of my distribution.
From FishCount: “gutting alive (gibbing in the case of herring): 25-65 minutes; asphyxiation without gutting: 55-250 minutes”
So using 90 mins as conservative estimate
[Update in response to Forum comment; see just below this list for more] adjusting down to 25 minutes, to account for small fish size
Also FC: median estimate of 1.55T wild fish killed annually circa 2010
Lower bound .79T
Will use 1.5T as conservative estimate
So conservative estimate = 135T minutes of intense suffering annually
Lower bound = 25*1.5T~=36T minutes
Upper bound = 140*2.5T=350T minutes
1T minutes~=1.9M years
So main conservative estimate = 256,849,315~=250 million years per year
[Updated thanks to this comment] now using lower bound on time of 36T minutes
This gives 68 million years/year
Moral weight
Point estimate: mean fish has hedonic capacity 1⁄10 of human
Upper (reasonable) bound: divide by log(neuron ratio)=log(1600)~=7.38
1⁄7.38 = .136
So upper bound is 13.6% human
Point estimate = geometric mean of bounds = .82% (very uncertain)
[Update] I now think that straight neuron count is the best we have
More reliable, conservative estimate of moral weight: using neuron counts as linear proxy:
Using zebrafish and human neuron counts, we get .1% moral weight relative to humans
So this is equivalent to 68 million*.001 ~= 680,000 years of human torture
How much normal life would I give up to avoid one minute of this torture?
Lower bound: 14 days = 20,160 minutes
Point estimate: 120 days ~= 170,000 mins
Upper bound: infinite time
So using point estimate, we get 115B QALYs or 1.58 billion typical (73 yr) human lives
I estimate that this is, as a lower bound, the moral equivalent of 2 million years of human torture (accounting for both moral weight given sentience and uncertainty about sentience) per year
Comment by EA Forum user Saulius a few weeks back, in response to me
Hey, thanks for writing this, there are some interesting ideas here. A bit of a nitpick, but I’m not sure that your “estimate 250 million fish years are spent in agony each year as wild fish are killed by asphyxiation or being gutted alive” is quite accurate . You are extrapolating from the length of time it takes for herring, cod, whiting, sole, dab and plaice to suffocate to all wild-caught fish. But I think that all of these are rather big fish and they likely were studied and mentioned by FishCount because it takes so long for them to suffocate. For example, 17%–65% of all wild-caught fishes are anchovies (295–908 billion fishes per year), and this video claims that “anchovies die immediately when they are out of water.” (though I don’t know how reliable that video is). I tried to estimate the same things (after reading the same text) here. I estimated that 0.7–49 million herring, cod, whiting, sole, dab, and plaice are suffocating in the air after being landed at any time (and didn’t make an estimate for other fishes). Also, there’s already some research on humane slaughter of fish, someof it is funded by Open Philanthropy, I don’t know if it is neglected or not.
Brief response
All points are well-taken, and I don’t doubt any of them. Nonetheless, if we introduce fish size into the model, we should consider not only time to death and numbers killed but also moral weight/capacity to suffer, which plausibly correlates with total size insofar as any measure of fish size correlates with e.g., brain size, brain mass, or neuron count.
Pending further research, I don’t know whether this update favors or disfavors the cause area because (using the simple big/small dichotomy to illustrate) its sign depends on whether more total minutes of suffering are endured by big fish or small fish, which in turn depends on the relative magnitudes of time of death and numbers caught by size.
Animal welfare sub-area comparisons
Chickens (total life): 2020 chicken population of 33 billion
In 2016, OpenPhil estimated 60 million chickens/year shifted from cage to cage-free
So species as cause area = 33 billion years per year of moderate(?) suffering
Using Wild Turkey as a comparison with ~500M neurons, this would correspond to about 1.5T fish-years (because about 50 times more neurons per animal)
But most of existence isn’t undergoing excruciating pain
Intuitively, to me it seems much better to live in (very poor) farmed conditions for 25 minutes than be slaughtered alive for one minute
So ~1/300th the amount of time spent living in farmed conditions vs being slaughtered alive
And, of course, this is all just for wild fish slaughter—excluding farmed fish suffering entirely. I have not had time to make a comparison with insect welfare or other wild animal suffering interventions, and may update this post with one or both going forward the future.
Neglectedness
The quoted phrase “wild fish slaughter” returns just eight results on Google, of which two are both (i) about wild fish slaughter as I mean the phrase and (ii) not written by me).
There seems to be plenty of awareness that it might be an issue, but little actual action taken, such as pilot projects
I’ll speculate that part of this might be due to the reluctance of welfare-minded individuals to take part in the gruesome process of killing fish, even if done with the purpose of reducing their suffering.
Some of the most forward-thinking, well-formed documents only address farmed fish slaughter
As I am discovering shortly before publication, in 2019, OpenPhil did make a grant (~$570k at the time of conversion) to the Humane Slaughter Foundation to work on this very issue!
If that grant (three years ago) is indeed the lone significant “EA action” on the issue, the area would still seem neglected in the relevant respects
It’s not immediately obvious to me what the relevant update(s) are to importance and tractability
Tractability: a few intuition-based notes
At present, most wild-caught fish killed by live gutting or asphyxiation
Likely, the most tractable ideas are not going to be great or even humane, but simply significantly less inhumane than existing practices, which are about as bad as one can imagine
In fact, doing just about anything to reduce the time between catch and death would probably be good for welfare,
The point then is that intensive, rigorous empirical inquiry into the subjective effects of various slaughter methods is good but not necessary to begin
Maybe catching bigger fish makes it easier to kill humanely
In the extreme case, imagine euthanizing a whale vs. trying to kill 10,000 minnows. For fish that are to be processed into fishmeal, one can imagine a physical grinder of sorts that kills the animals very quickly, albeit without preserving their bodily integrity.
A whole additional set of interventions are those that reduce the number of wild fish killed without causing enough other suffering to offset this effect
For instance, I can imagine subsidizing humane farmed fishing (with humane slaughter, which I believe is easier in aquaculture settings), which would then (I assume?) reduce the number of wild fish killed.
Cause exploration: improving wild fish slaughter [half-baked, post-deadline]
Intro
This post is far less cogent and complete than the dozens of well-footnoted, (nearly-) GiveWell quality analyses. After some initial research and writing a few weeks back, I stalled, unsure about the next appropriate step, and eventually lost motivation.
What I’m left with is a haphazard collection of notes that I’m cleaning up only slightly in the final hours of the competition Some parts are decent, some are practically scribbles, and most fall in-between.
Epistemic status: ⅓ real cause area investigation, ⅔ haphazard collection of notes (but notes collected in earnest), still stand by everything.
Why I think it’s worth sharing now
This “map” is pretty meager as a report and research product, not the kind of thing I’d like handing in as a final project in an EA college course, but the map is not the territory, and the territory is worth looking at!
Although my failure to investigate, document, and cite more than I have renders the cause area’s true promisingness high variance, my current sense is that wild fish slaughter:
Deserves more total resources through some channel than it currently receives or has received over the last few years (90%)
Deserves more resources from OpenPhil than it is currently getting (75%)
Is more deserving of marginal dollars than any GiveWell top charity, conditional on accepting the in-principle validity of such a comparison (95%)
Is more deserving of marginal resources than any of the big existing “EA approved” farmed animal welfare interventions like chicken cage-free campaigns, cultured meat development, and improving farmed fish aquaculture conditions (65%)
Is the single, non-longtermist,[1] object level large cause area most deserving of “EA resources” (20%)
It’s more likely that someone from OpenPhil will read and consider the points below if I post it as a competition entry rather than a miscellaneous forum post in a few weeks, and further that this is morally good on net (even accounting for various opportunity costs)
As a final bit of preamble, my intention here isn’t to make suffering porn, but I also don’t want to dance around the issue by using terms that are misleadingly benign. Not sure I strike this balance well, but I urge you not to stop reading (if you’ve made it this far) just because what I’m describing is bad.
Importance
Section summary
I estimate that wild fish slaughter accounts for 68 million years of extreme fish suffering (given that fish are sentient) per year. Further, using neuron count as a proxy for capacity to suffer, I estimate this is the moral equivalent of 680,000 years of human torture, which in turn corresponds to (with very large uncertainty) the moral value of 1.58 billion human deaths per year.
BOTEC: Moral value of wild fish slaughter (annual)
Note: Each figure is supposed to represent a sort of conservative-but-realistic estimate, so maybe something around the 25th percentile of my distribution.
From FishCount: “gutting alive (gibbing in the case of herring): 25-65 minutes; asphyxiation without gutting: 55-250 minutes”
So using 90 mins as conservative estimate
[Update in response to Forum comment; see just below this list for more] adjusting down to 25 minutes, to account for small fish size
Also FC: median estimate of 1.55T wild fish killed annually circa 2010
Lower bound .79T
Will use 1.5T as conservative estimate
So conservative estimate = 135T minutes of intense suffering annually
Lower bound = 25*1.5T~=36T minutes
Upper bound = 140*2.5T=350T minutes
1T minutes~=1.9M years
So main conservative estimate = 256,849,315~=250 million years per year
[Updated thanks to this comment] now using lower bound on time of 36T minutes
This gives 68 million years/year
Moral weight
Point estimate: mean fish has hedonic capacity 1⁄10 of human
Lower bound moral weight: .05% human (see SSC)
Tiny Zebrafish has 10m neurons
Seems like a reasonable estimate
Humans have 16b
Upper (reasonable) bound: divide by log(neuron ratio)=log(1600)~=7.38
1⁄7.38 = .136
So upper bound is 13.6% human
Point estimate = geometric mean of bounds = .82% (very uncertain)
[Update] I now think that straight neuron count is the best we have
More reliable, conservative estimate of moral weight: using neuron counts as linear proxy:
Using zebrafish and human neuron counts, we get .1% moral weight relative to humans
So this is equivalent to 68 million*.001 ~= 680,000 years of human torture
How much normal life would I give up to avoid one minute of this torture?
Lower bound: 14 days = 20,160 minutes
Point estimate: 120 days ~= 170,000 mins
Upper bound: infinite time
So using point estimate, we get 115B QALYs or 1.58 billion typical (73 yr) human lives
I estimate that this is, as a lower bound, the moral equivalent of 2 million years of human torture (accounting for both moral weight given sentience and uncertainty about sentience) per year
Comment by EA Forum user Saulius a few weeks back, in response to me
Brief response
All points are well-taken, and I don’t doubt any of them. Nonetheless, if we introduce fish size into the model, we should consider not only time to death and numbers killed but also moral weight/capacity to suffer, which plausibly correlates with total size insofar as any measure of fish size correlates with e.g., brain size, brain mass, or neuron count.
Pending further research, I don’t know whether this update favors or disfavors the cause area because (using the simple big/small dichotomy to illustrate) its sign depends on whether more total minutes of suffering are endured by big fish or small fish, which in turn depends on the relative magnitudes of time of death and numbers caught by size.
Animal welfare sub-area comparisons
Chickens (total life): 2020 chicken population of 33 billion
In 2016, OpenPhil estimated 60 million chickens/year shifted from cage to cage-free
So species as cause area = 33 billion years per year of moderate(?) suffering
Using Wild Turkey as a comparison with ~500M neurons, this would correspond to about 1.5T fish-years (because about 50 times more neurons per animal)
But most of existence isn’t undergoing excruciating pain
Intuitively, to me it seems much better to live in (very poor) farmed conditions for 25 minutes than be slaughtered alive for one minute
Farmed fish: ~80B/year
So ~1/300th the amount of time spent living in farmed conditions vs being slaughtered alive
And, of course, this is all just for wild fish slaughter—excluding farmed fish suffering entirely. I have not had time to make a comparison with insect welfare or other wild animal suffering interventions, and may update this post with one or both going forward the future.
Neglectedness
The quoted phrase “wild fish slaughter” returns just eight results on Google, of which two are both (i) about wild fish slaughter as I mean the phrase and (ii) not written by me).
“Slaughter of wild fish” returns a few thousand. Going through them…
Long 2010 report by FishCount
A short list of bulleted notes
Faunalytics report
One researcher has gotten a grant to study the issue
There seems to be plenty of awareness that it might be an issue, but little actual action taken, such as pilot projects
I’ll speculate that part of this might be due to the reluctance of welfare-minded individuals to take part in the gruesome process of killing fish, even if done with the purpose of reducing their suffering.
Some of the most forward-thinking, well-formed documents only address farmed fish slaughter
As I am discovering shortly before publication, in 2019, OpenPhil did make a grant (~$570k at the time of conversion) to the Humane Slaughter Foundation to work on this very issue!
If that grant (three years ago) is indeed the lone significant “EA action” on the issue, the area would still seem neglected in the relevant respects
It’s not immediately obvious to me what the relevant update(s) are to importance and tractability
Tractability: a few intuition-based notes
At present, most wild-caught fish killed by live gutting or asphyxiation
Likely, the most tractable ideas are not going to be great or even humane, but simply significantly less inhumane than existing practices, which are about as bad as one can imagine
In fact, doing just about anything to reduce the time between catch and death would probably be good for welfare,
The point then is that intensive, rigorous empirical inquiry into the subjective effects of various slaughter methods is good but not necessary to begin
A few amateur solution ideas and notes
Fish pipelines/cannons exist; could this be integrated with slaughter?
Maybe catching bigger fish makes it easier to kill humanely
In the extreme case, imagine euthanizing a whale vs. trying to kill 10,000 minnows. For fish that are to be processed into fishmeal, one can imagine a physical grinder of sorts that kills the animals very quickly, albeit without preserving their bodily integrity.
A whole additional set of interventions are those that reduce the number of wild fish killed without causing enough other suffering to offset this effect
For instance, I can imagine subsidizing humane farmed fishing (with humane slaughter, which I believe is easier in aquaculture settings), which would then (I assume?) reduce the number of wild fish killed.
By “non-longtermist,” I mean not motivated by longtermism. I’m agnostic about its long-run importance.