Thursday, November 05, 2015

Various functional forms for brain-weighting wild insects and farmed land animals favor the former

Summary: Some people have offered guesses or intuitions that when comparing animals with very different nervous system scales, they should be weighted by the logarithm or square root of neural capacity, while also taking the view that the impacts of animal agriculture on wild insects are not much greater than the impacts on farmed land animals. Considering these functional forms, and linear weighting with number of neurons, it appears they assign a much greater total weight to wild insect populations affected by agricultural land use than to the land animals being farmed. This suggests a revision of some combination of the weighting schemes and evaluations of agricultural impacts.

William MacAskill, in his Reddit AmA on his new book "Doing Good Better," provides me with a convenient foil in answer to a question from "cbr":
Would you rather save one horse-sized duck or a hundred duck-sized horses?
MacAskill replied:
I'd certainly rather save a hundred duck-sized horses. 
It's hard to know how to compare the moral importance of different creatures' experiences. How many happy chicken-days is as good as a happy chimp-day? 
The best guess I currently have is to use the logarithm of neural mass. And I think that the total log(neural mass) of a hundred duck-sized horses is much greater than that of one horse-sized duck. There's just a lot more experiencing entities, and even if the horse-sized duck's experiences are a bit more valuable in light of greater computational resources powering them, it's not that much greater. 
Moreover, horses live a little longer than ducks (25-30 years compared to about 20 years, according to a quick google). Insofar as I think we should care not about number of lives saved, but number of quality-adjusted life-years saved, then saving the duck-sized horses is clearly going to have the bigger impact.
MacAskill shortly after realized that the logarithm of neural capacity was a poor fit for his (highly provisional) intuitions when considering more creatures with small neural capacity, such as insects, but I have now heard several people propose the logarithm or square root of neural capacity as a rough summary of their intuitions, and I can use it as an existence proof for this post.

Wild insects vs farmed animals by functional form for neuron-weighting

Wikipedia's list of animals by number of neurons gives a figure for ants of 250,000 neurons (noting great variety by species), 960,000 for honeybees, and 1,000,000 for cockroaches. The larger numbers are for larger insects, while most insects are smaller ants and termites, so I will use a figure of 100,000 neurons.

Chicken brain mass estimates for different lineages and studies are often in the range of 2.5-4 g. At the human ratio of  mass to neurons this would suggests ~200 million neurons.

Cow brains are ~100+ times more massive than chicken brains, although likely have fewer neurons per gram than humans or chickens. In her book Herculano-Houzel estimates (apparently by extrapolation from related animals) that a cow brain would have 3 billion neurons, and a chimpanzee 6 billion. I will use the 3 billion number

Now we can compare this groups with linear, square root, and logarithmic functional forms:

# of neuronssqrt(# of neurons)log10(# of neurons)
We can normalize in terms of insects to see how many insects would be equivalent to a chicken or cow with each weighting scheme:

Insect-normalized # of neuronsInsect-normalized sqrt(# of neurons)Insect-normalized log10(# of neurons)

One might then discount for uncertainty about whether insects have morally relevant well-being. Effective altruist and philosopher Pablo Stafforini, along with Aron Vallinder, put together a small expert survey on animal consciousness aimed at philosophers and scientists and got 13 respondents. This survey could surely be much improved on with large samples, analysis of respondent background effects, enlightened preference methodology, and analysis of response bias, but it is a useful starting point.

7/13 experts answered that the probability insects could suffer was between 20% and 50%, plus 3 giving very high estimates and 3 giving very low estimates. Since there are likely hundreds of thousands of ants and termites per farmed land animal (mostly chickens), it seems that all of these functional forms would favor a focus on wild animal impacts of agriculture, not farmed animals. This seems true even for linear weighting (see this thread for some arguments for near-linear weighting in hedonistic calculations), although in less extreme fashion. However, superlinear weighting is also possible.

This conclusion differs from that of most of the people I have heard propose square root or log weightings, who typically did so in the context of policies giving cattle and pigs some modest extra weight (an order of magnitude or less) relative to chickens, but not considering the impacts of farming on wild animals to be more important than those on farmed animals. This suggests some combination of revision of brain-weighting rules-of-thumb and revision of weighting of farmed and wild land animals.


Brian Tomasik said...

It's also interesting to note that there may be about as many nematode neurons as insect neurons in the world, given that nematodes outnumber insects by 10^3 to 10^4 times but have about 10^3 times fewer neurons than insects. (C. elegans has exactly 302 neurons.) Of course, for those who impose a threshold on whether an animal counts at all, it will be less likely that nematodes pass that threshold than that insects do.

Squark said...

I don't understand the reasoning behind these concave weight functions. My intuition is that the weight function has to be convex precisely because decomposing a brain into individual neurons clearly destroys its moral value. Worse, it seems unlikely that any simple weight function is going to work. Imagine growing an artificial brain in a lab which has an enormous number of neurons but they are interconnected randomly and it has nothing resembling consciousness. In other words, measuring moral worth of brains by counting neurons is like measuring moral worth of computers by counting bits (apply this principle to two computers, one of them emulating a human brain and the other running many copies of Tetris).

Brian Tomasik said...

Hi Squark :) I discuss some thoughts on the topic here. I completely agree that counting neurons isn't a sufficient measure in general, but it may be a crudely useful approximation within the set of, e.g., typical animal minds on Earth.

Decomposing a brain into single neurons probably reduces its importance in most people's estimation, but it's not obvious that the trend is monotonic. I think many people would find it plausible that if you hypothetically cut a brain in half and inserted one half into the skull of another (previously dead) human body, the resulting set of two people would have more total moral importance, even if both resulting humans would have significant cognitive disabilities.