This uncertainty [about consciousness] gives us some reason to be suspicious of very low estimates of animal consciousness.
I think consciousness is usually defined in a way that is either unfalsifiable or too vague for estimating a probability to be useful (like what-is-it-likeness). I believe it would be better to focus just on what animals can do (for instance, probability of a given animal having a score of at least x % on a given test), which can be be empirically assessed.
If a creature behaves dramatically, as would be expected if it were in intense pain, we should think itâs reasonably likely that the creature does, in fact, feel intense pain [as intense as intense pain in humans].
I agree that is a possibility, but I struggle to quantify how likely it is. Their pain during dramatic behaviours could be much more intense than their typical experiences, but still much less intense than human pain.
A third consideration is evolutionary: there is every evolutionary reason for simple creatures to feel intense pain. Pain in humans causes them to avoid various behaviors that are likely to be harmful. Pain in fish, insects, and decapods serves similar functions. We should expect intense pain in these creatures to be adaptive for the same reason that pain in humans is adaptive. If a creature just felt mild discomfort, that would be unlikely to sustain the kind of behavior needed to safeguard survival. Intense pain could be adaptive for eliciting rapid and dramatic behavior.
Evolutionarily speaking, nothing would change if the rewards and punishment had a value of âalternative valueâ = k*âvalueâ, where k is positive? The expected value of each action would have been k times as large, but the ratio between the expected value of any 2 actions would have been exactly the same. k can in principle be arbitrarily close to 0. So experiences could have an intensity arbitrarily close to 0 while leading to exactly the same behaviour?
A similar argument can be employed against many of the objections to significant pain in animals. Animals have relatively simple brains. But if we came across a human with a comparatively simple brain, who seemed to respond aversively to external damage the way animals do, we would be hesitant to be confident that its conscious experience is only mild. We should have similar skepticism directed towards arguments for such conclusions about animals.
This argument only applies to mammals at most?
The eighth and final argument for taking seriously the suffering of simple creatures: the most detailed report to date which attempted to estimate intensity of valenced experience in animals guessed that even simple creatures were, in expectation, intensely conscious. This report came from the team of Fischer et al. (2024). Using a mixed function taking into account various different behavioral proxies, the final estimate (p.233) was that shrimp suffer 8% as intensely as humans and carp 24% as intensely. The median estimate was that shrimp suffered 5% as intensely as humans.
The estimates for sentience-adjusted welfare ranges in Bob Fischerâs book are calculated for 3 models. The value for shrimps as a fraction of that of humans is 40 % for the equality model, 0.035 % for the neurophysiological model, and 6.6 % for the simple additive model. However, it is very unclear to me how to combine these in a principled way. I would say that the key takeaway is that a best guess equal to any of those 3 values is reasonable. Moreover, the 3 models only cover a very tiny fraction of possibilities. Models leading to estimates of the type âestimate for the neurophysiological modelâ^âexponentâ with high exponent could easily lead to very small sentience adjusted welfare ranges as a fraction of that of humans. For example, 1.23*10^-7 (= (3.5*10^-4)^2) for shrimps for an exponent of 2. How do you rule out such models? I read the whole book, and I did not find any justification.
Using a more integrated model that took into account even more inputs (see Duffy, 2024, table 2) their mean estimate was that shrimp suffered a whopping 19% as intensely as humans, bees 14.8%, and carp 17.5%.
Bobâs book superseded that work. âFor updates, please see this [Bobâs] bookâ.
And when humans lose many neurons, so long as their brains perform similar functions, we donât normally think they are vastly less conscious.
This could be true, and animals still have experiences of astronomically low intensity? Say the intensity of experiences (or the sentience-adjusted welfare range) is proportional to âindividual number of neuronsâ^2. Decreasing the number of neurons by 1 % would decrease intensity by 2 %. Nothing dramatic? Estimates for the number of neurons in humans range from 61 billion to 99 billion. So the intensity of experiences would vary by a factor of 2.63 (= (99/â61)^2) specifically because of differences in the number of neurons, which seems like a reasonable upper bound to me. If so, the experiences of shrimps could have an intensity as low as 10^-12 (= (10^-6)^2) times that of the experiences of humans, as shrimps have 10^-6 times as many neurons as humans.
On the other hand, I am also open to the experiences of shrimps being roughly as intense as those of humans. For intensity proportional to âindividual number of neuronsâ^âexponentâ, values from 0 to 2 cover the best guesses that I consider reasonable.
Shriver (2024) provides a number of reasons to be skeptical of neuron count proxies.
âNumber of neuronsâ^0.188 explains very well the estimates for sentience-adjusted welfare ranges in Bobâs book.
More importantly, the highest estimate for the number of neurons is humans is 1.62 (= 99â61) times the lowest estimate, whereas humans have 1 M times as many neurons as shrimps. So I think the number of neurons explaining very little of the variation in the intensity of human experiences is practically no evidence for human experiences being roughly as intense as those of shrimps.
Hi Matthew. Great post.
I think consciousness is usually defined in a way that is either unfalsifiable or too vague for estimating a probability to be useful (like what-is-it-likeness). I believe it would be better to focus just on what animals can do (for instance, probability of a given animal having a score of at least x % on a given test), which can be be empirically assessed.
I agree that is a possibility, but I struggle to quantify how likely it is. Their pain during dramatic behaviours could be much more intense than their typical experiences, but still much less intense than human pain.
Evolutionarily speaking, nothing would change if the rewards and punishment had a value of âalternative valueâ = k*âvalueâ, where k is positive? The expected value of each action would have been k times as large, but the ratio between the expected value of any 2 actions would have been exactly the same. k can in principle be arbitrarily close to 0. So experiences could have an intensity arbitrarily close to 0 while leading to exactly the same behaviour?
This argument only applies to mammals at most?
The estimates for sentience-adjusted welfare ranges in Bob Fischerâs book are calculated for 3 models. The value for shrimps as a fraction of that of humans is 40 % for the equality model, 0.035 % for the neurophysiological model, and 6.6 % for the simple additive model. However, it is very unclear to me how to combine these in a principled way. I would say that the key takeaway is that a best guess equal to any of those 3 values is reasonable. Moreover, the 3 models only cover a very tiny fraction of possibilities. Models leading to estimates of the type âestimate for the neurophysiological modelâ^âexponentâ with high exponent could easily lead to very small sentience adjusted welfare ranges as a fraction of that of humans. For example, 1.23*10^-7 (= (3.5*10^-4)^2) for shrimps for an exponent of 2. How do you rule out such models? I read the whole book, and I did not find any justification.
Bobâs book superseded that work. âFor updates, please see this [Bobâs] bookâ.
This could be true, and animals still have experiences of astronomically low intensity? Say the intensity of experiences (or the sentience-adjusted welfare range) is proportional to âindividual number of neuronsâ^2. Decreasing the number of neurons by 1 % would decrease intensity by 2 %. Nothing dramatic? Estimates for the number of neurons in humans range from 61 billion to 99 billion. So the intensity of experiences would vary by a factor of 2.63 (= (99/â61)^2) specifically because of differences in the number of neurons, which seems like a reasonable upper bound to me. If so, the experiences of shrimps could have an intensity as low as 10^-12 (= (10^-6)^2) times that of the experiences of humans, as shrimps have 10^-6 times as many neurons as humans.
On the other hand, I am also open to the experiences of shrimps being roughly as intense as those of humans. For intensity proportional to âindividual number of neuronsâ^âexponentâ, values from 0 to 2 cover the best guesses that I consider reasonable.
âNumber of neuronsâ^0.188 explains very well the estimates for sentience-adjusted welfare ranges in Bobâs book.
More importantly, the highest estimate for the number of neurons is humans is 1.62 (= 99â61) times the lowest estimate, whereas humans have 1 M times as many neurons as shrimps. So I think the number of neurons explaining very little of the variation in the intensity of human experiences is practically no evidence for human experiences being roughly as intense as those of shrimps.