I really appreciate this post and the conclusion here seems very reasonable. As someone who is personally guilty of using neuron counts as a sole proxy for moral weight, I would love to include additional metrics that more closely proxy something like capacity for suffering and pleasure. However, my problem is that while the metrics mentioned (mirror-test, trace conditioning, unlimited associative learning, reversal learning) might be more accurate proxies, they are (as far as I can tell) not available for a wide variety of species. For me, the main goal of employing these moral weights is to get a framework that decisionmakers can use for evaluating the impact of any project. I am particularly interested in government cost-benefit analyses, where the ideal use case would be to have a spreadsheet where government economists could just plug in available proxies for moral weight and get an estimated valuation for suffering reduction for an individual of a particular species. Neuron counts are nice for this because you can pretty easily find an estimated neuron count for almost any species. With this issue in mind,
Are you aware of any papers/databases that have a list of species for which any of the four recommended factors have been tested and the results? It seems, for example, that scientists make headlines when they find a species that passes the mirror test but I can’t tell which species have “failed” it versus which have not been tested.
Other factors that are widely available for many species include brain mass, body mass, brain-to-body mass ratio, cortical neurons, whether the animal has any particular brain/anatomical structure, class/order, etc. It sounds like maybe the ratio of cortical neurons to brain size might be a reasonable proxy based on the section on processing speed—would you agree that would be an improvement over just neurons? Do any of these other characteristics stand out as plausible proxies?
We tend to think that if the goal is to find a single proxy, something like encephalization quotient might be the best bet. It’s imperfect in various ways, but at least it corrects for differences in body size, which means that it doesn’t discount many animals nearly as aggressively as neuron counts do. (While we don’t have EQs for every species of interest, they’re calculable in principle.)
Finally, we’ve also developed some models to generate values that can be plugged into cost-benefit analyses. We’ll post those in January. Hope they’re useful!
Thank you, this is very helpful and I definitely agree that EQs are available/practical enough to use in most cases. Really looking forward to seeing the new models in January!
I really appreciate this post and the conclusion here seems very reasonable. As someone who is personally guilty of using neuron counts as a sole proxy for moral weight, I would love to include additional metrics that more closely proxy something like capacity for suffering and pleasure. However, my problem is that while the metrics mentioned (mirror-test, trace conditioning, unlimited associative learning, reversal learning) might be more accurate proxies, they are (as far as I can tell) not available for a wide variety of species. For me, the main goal of employing these moral weights is to get a framework that decisionmakers can use for evaluating the impact of any project. I am particularly interested in government cost-benefit analyses, where the ideal use case would be to have a spreadsheet where government economists could just plug in available proxies for moral weight and get an estimated valuation for suffering reduction for an individual of a particular species. Neuron counts are nice for this because you can pretty easily find an estimated neuron count for almost any species. With this issue in mind,
Are you aware of any papers/databases that have a list of species for which any of the four recommended factors have been tested and the results? It seems, for example, that scientists make headlines when they find a species that passes the mirror test but I can’t tell which species have “failed” it versus which have not been tested.
Other factors that are widely available for many species include brain mass, body mass, brain-to-body mass ratio, cortical neurons, whether the animal has any particular brain/anatomical structure, class/order, etc. It sounds like maybe the ratio of cortical neurons to brain size might be a reasonable proxy based on the section on processing speed—would you agree that would be an improvement over just neurons? Do any of these other characteristics stand out as plausible proxies?
Hi Monica! We hear you about wanting a table with those results. We’ve tried to provide one here for 11 farmed species: https://forum.effectivealtruism.org/posts/tnSg6o7crcHFLc395/the-welfare-range-table
We tend to think that if the goal is to find a single proxy, something like encephalization quotient might be the best bet. It’s imperfect in various ways, but at least it corrects for differences in body size, which means that it doesn’t discount many animals nearly as aggressively as neuron counts do. (While we don’t have EQs for every species of interest, they’re calculable in principle.)
Finally, we’ve also developed some models to generate values that can be plugged into cost-benefit analyses. We’ll post those in January. Hope they’re useful!
Thank you, this is very helpful and I definitely agree that EQs are available/practical enough to use in most cases. Really looking forward to seeing the new models in January!