by Brian Tomasik
First written: 30 June 2013; last updated: 8 Dec. 2016

Summary

It seems plausible that evolution should sculpt short-lived animals to be more reckless than long-lived ones as far as taking risks for proximate gain. This could be taken to imply that short-lived animals feel less fear and maybe less pain than long-lived ones, although one could also interpret it as saying that short-lived animals feel more pleasure upon success than long-lived ones. Calibrating the hedonic scales between short- and long-lived animals is undetermined purely by evolutionary considerations. Because death is a one-time event, the pain it entails is not constrained by fitness considerations and hence could still be very high even for reckless animals. That shorter-lived animals have less need for learning may be weak evidence to reduce their probability of sentience relative to long-lived animals.

Finally, keep in mind that this single model of hedonic rewards is highly theoretical, and we shouldn't jump to conclusions on this basis alone; I suggest examining actual data on risk tolerances by short- vs. long-lived animals. It's distressing to think about the magnitude of insect suffering in the world, and it's tempting to look for excuses why we can ignore it. The extremely rough speculations proposed in this piece should not be regarded as license to substantially downweight the importance of insect suffering.

Introduction

"A lifetime isn't forever, so take the first chance, don't wait for the second one! Because sometimes, there aren't second chances!"  --C. JoyBell C.

Evolutionary perspective is helpful in assessing whether and how much organisms feel pain. For instance, one of several arguments against plant suffering is that plants would have little evolutionary benefit from fear or pain because they can't run away. (Of course, they can marshal internal responses to stress/injury and occasionally signal danger to neighbors.) Another example: The reason that the worst suffering people experience is far worse than the best pleasure is good is that bad injury can destroy a person's entire future fitness (having and caring for multiple kids), whereas the best momentary experience one could have is, say, some small chance of starting one pregnancy with a good partner (and, considering the rates of infant and child mortality in the ancestral environment, even this was far from equivalent to having one new successful child).

Ultimately fitness only comes from creating viable offspring, but learning is more successful when intermediate states are also given value, or else the organism is unlikely to reach the end states. For instance, when playing chess, only winning at the end matters, but in order to make computation more feasible, you can assume that intermediate states with more players standing are more valuable, that the queen is more important than a pawn, and so on. Similarly, we can say in rough terms that food is good for fitness, a small injury is slightly bad, losing a child to a predator is very bad, etc. A more successful organism will have its reward/punishment weights assigned in such a manner that the weights for a given outcome approximate the intermediate fitness impact of the outcome. Empirically it seems that a state's degree of subjective pleasure/pain is roughly proportional to this reinforcement-learning reward/punishment value, although in the brain it appears that "liking" is distinct from "learning," and if so, this correspondence needn't be exact.

Live fast, fear less

When we consider organisms with different ages, we notice something interesting: Those organisms with more life ahead of them have reason to be more cautious than those that will die soon. This is because if the long-lived organisms mess up, they'll lose potential future fitness for a long time to come, whereas if they're near the end of their lives, it's less severe for them to permanently damage themselves. Thus, longer-lived organisms should only take risks for higher rewards and should only go after a given reward when its risk is small, relative to what short-lived organisms would do. I explore a concrete demonstration of this in "Stylized Model for an Agent's Risk Aversion with Respect to Lifespan," but the elaboration there is ponderous, and the basic point is intuitive enough on its own.

This theory predicts that organisms should become more reckless with old age, but I don't think that's empirically true.a One reason could be that organisms don't actually have a definite endpoint to their lives, so they never reach a time when they'd be as reckless as you might be if you knew the world was ending tomorrow. Another reason is that an organism's fitness per unit time in later years may be lower than in earlier years, especially after it can no longer bear children. As a result, it has less to lose by being reckless earlier. (Of course, it can stick around to take care of its grandchildren, although it also needs to worry about wasting food and other resources that could have gone to the grandchildren themselves.) A final explanation is just that evolution wasn't sophisticated enough to make fear, pain, and pleasure weights depend on an organism's seniority, so it fixes them at a single value.

Short-lived animals, like animals near the end of their lives, have only a few opportunities left for reproduction. Individuals of some species only reproduce once before death, so their eggs are "all in one basket." In this case, we would expect short-lived animals to have less fear/inhibition relative to longer-lived animals. Insofar as pain due to injury, stress, and other non-lethal forms of suffering also inhibit risk-taking in the future, we might expect shorter-lived animals to have lower intensities of these relative to longer-lived animals. Hedonic Treader proposed this idea in a Felicifia thread.

Application to insects?

Insects live short lives relative to humans, so if this abstract model does apply to insects, we might expect, for instance, that they would fear dangerous situations less than a human would. Of course, it's not known if insects can feel fear or pain at all. However, if they can, we would expect its intensity to vary in the way suggested above. Similarly, they should be less bothered by injuries, because they don't have time to wait for healing, and they should not be inhibited from doing risky things in the future. Could this help explain why insects don't seem troubled by certain forms of tissue damage, like a broken leg?

The fitness model I proposed is highly abstract, and it's not clear to me exactly how often insects would encounter situations where they'd risk future fitness in order to gain short-term reward. What might be examples of this? Maybe they'd be more willing to go near a predator in order to capture food than a longer-lived animal would?

One real-world example is Brazilian fig wasps, whose adult lifespans are only 1-2 days. Due to high levels of competition for figs in which to lay eggs, these wasps fight to the death for figs, since their short lifespans don't allow for waiting.

On the other hand, there are many instances where insects display similar levels of aversion as we would expect from larger animals. For example, when a shock is paired with an odor, fruit flies avoid that odor in the future. Situations that threaten all future fitness may still be highly aversive and are not uncommon in nature.

How about pain of death?

Consider a fly trapped in a spider's web. If it doesn't escape, it will lose all potential future fitness. Now consider a human being chased by a lion. If it doesn't escape, it will lose all potential future fitness. In each case, the cost relative to the total value of the organism's life is the same, so it seems the amount of distress evoked in each case should be similar.

Death itself is not subject to learning, because it happens only once. Any feedback signal would be ineffectual, since the soon-to-be-dead animal could never use the signal to change its behavior in the future. Hence, evolution has no pure "death" punishment feeling. That said, the lead-up to death is often accompanied by body destruction that hurts for other reasons: either because of learned aversion to smaller injuries or because of hard-wired triggers that fire in response to tissue damage. While the moment of death has no reinforcement signal, many of the things that typically accompany the trajectory toward death are built to hurt a lot, to dissuade the organism from continuing down that path and, more importantly, because they also cause fitness damage even if they don't kill the organism. Insofar as more reckless animals should be willing to tread farther down the near-death path than more cautious ones, we might expect that the lead-up to death might hurt somewhat less for them, but it's not clear how broadly this would be true.

Also keep in mind that insects also have orders of magnitude more offspring per parent pair than do big animals. So even if the per-child pain of death were orders of magnitude lower for an insect (which seems dubious but not impossible), there are orders of magnitude more child deaths. Even in this model, it's still plausible that the total suffering due to all the deaths of the offspring of an insect pair exceeds the suffering entailed by the death of one human offspring.

Hurting yourself more

If shorter-lived animals are more reckless because they have less fear and maybe less pain, does this mean they suffer less? Maybe, but remember that being more reckless means they injure themselves more often. This is reminiscent of compensation effects in economics, such as when a decline in the marginal-cost curve increases quantity demanded, which pushes the marginal cost back up a little bit.

We can develop an economically inspired optimality model for motivational tradeoffs. In particular, consider the x-axis to represent the "quantity demanded" of resources by the animal, i.e., how much food, territory, mating opportunities, etc. it seeks out per day. It seems that the marginal benefit of these additional resources either declines (in the case of food) or stays constant (in the case of mating opportunities for a male), so in aggregate, we can represent marginal benefit by a declining curve:


Note that this is consistent with our psychological perception of hedonic rewards: The Nth banana is less pleasurable than the first one, and so on.

At the same time, the cost of acquiring extra food, territory, etc. seems to increase as one tries to gain more and more, because one has to search farther away for food, protect a much greater area of land, and so on. As far as finding new mates, you might pick up the easiest ones first, and later ones could get harder and harder. So we expect the marginal-cost curve -- how much fitness you have to sacrifice to get a given level of resources -- to be at least flat or maybe increasing:

Evolutionarily optimized organisms will have hedonic systems that drive them to acquire resources until the marginal fitness benefit drops below the marginal fitness cost. The marginal fitness cost should be proportional to suffering, so the total suffering these organisms endure is the area under the marginal-cost curve up to the equilibrium point:

Now, suppose we consider a short-lived species where there's less opportunity cost to getting injured because the total future fitness lost thereby is smaller. In this case, the marginal-cost curve is lower. The result is less suffering for a given level of resources, but the equilibrium point also shifts out toward demanding more total resources:


It's not completely obvious whether the net result is less or more total suffering, though it seems intuitive that the increase in total suffering due to seeking out more resources might be less than the reduction due to being willing to take more risks.

The interpersonal-scaling problem

The discussion so far has one big hole in it. How much longer you have to live affects how reckless you are, but it doesn't say whether that's manifested as less fear/pain or as more excitement/pleasure. I could have just as well said that short-lived animals feel vastly more pleasure upon success than longer-lived animals, because they have few chances left, so it's worth risking terrible pain for rewards. In a species that mates only a few times before death and doesn't care for its offspring, a single orgasm might be like the sum of years of orgasms and child raising for a human, because a large fraction of the organism's fitness is contained in that action.

If instead of downweighting the pain of short-lived animals, we upweighted their pleasure, we would conclude that short-lived animals suffer equally with long-lived ones per injury, and because they risk injury more, they actually suffer more in total!

The marginal benefit/cost graphs from the previous section can capture the ambiguity about whether greater recklessness is due to lower suffering or merely higher pleasure. Instead of having a lower marginal-cost curve, we could say that short-lived animals have a higher marginal-benefit curve for a given level of resources because the percentage of their total fitness that comes from a given bite of food or mating event is much higher:

This is nothing other than the generic problem of scaling interpersonal utilities when we're taking the utilities to be based on motivational tradeoffs. We have to pick exchange rates among various organisms, but it's not obligatory that we set equal insect pleasure with human pleasure and downweight insect suffering. We could instead set insect suffering equal with human suffering and then upweight insect pleasure. Or we could do something in the middle.

Carl Shulman suggests (personal communication) that it's more plausible to say that insects suffer less rather than that they feel more pleasure because simpler animals can get by with fewer gradations of hedonic experience, so it's easier if evolution just uses less intense emotions. Of course, the choice of interpersonal weighting is still fundamentally up to us, but this provides some intuition for taking insect experience to be less intense. In contrast, others have made the reverse claim: Because simpler animals have fewer gradations, it's easier for them to get close to their worst possible experiences. If we took this view, we could regard insect suffering as more morally problematic than that of a larger animal with a broader range of experiences.

If we did consider injuries by more reckless animals to be less painful than those of bigger animals, it's interesting to ponder to what extent we could capture this through their reduced brain sizes. For instance, maybe an animal twice as reckless has half the size of pain neural regions but the same size of pleasure regions as a cautious animal. Of course, if the reckless animal is really small, then its whole brain would also be scaled down: e.g., maybe its pleasure regions would be 1/100th the size of the bigger animal's pleasure regions and the pain regions 1/200th the size of the bigger animal's pain regions. Note that it's not obvious we should weight ethically by brain size, and even if we do, we might give smaller brains more weight than their size implies due to much higher efficiency. But it's plausible that within a brain, weightings are roughly linear in size (or, at least, number of inputs to decision regions of the brain).

More caution if you're more likely to win

Suppose you're a member of a species where only 10% of young adults actually go on to reproduce. You find yourself near a fertile and successful mating partner, but you're not in the safest of locations. Do you take a risk and try to reproduce here, or do you wait for better conditions? Given that your odds of reproducing at all aren't so great, it makes sense to take the risk and mate now. If you wait, your chance of mating again will be less than 100% (but probably not as low as the 10% baseline because the fact that you could mate at all probably means you're above average). In contrast, if you're a member of a species like humans where many of the adults succeed in mating at some point, you'd better wait.

So fitness tradeoffs are affected not just by how much longer you have to live but also how likely it is you'd succeed in having children if you do live. As it becomes less likely that you'll reproduce even if you survive, you should be willing to take bigger risks. For example, gorillas are polygynous, and low-rank males by default may not produce any offspring. Thus, I would expect low-rank male gorillas to take bigger risks than more monogamous male gibbons. Of course, in this case, it's not at all clear that we should chalk up greater risk tolerance to reduced suffering; it's plausible that male gorillas are motivated to take risks by greater potential pleasure rather than less potential pain.

More r-selected animals should be more willing to take risks because on any given occasion, the expected fitness if they wait is lower. If this is manifested through reduced pain intensity, then we should expect this to somewhat offset the fact that r-selection multiplies the pain animals experience by creating many more non-surviving individuals. Insofar as the greater willingness for risk is manifested through greater pleasure upon success, r-selection's tendency to multiply suffering in nature remains true.

Note also that as an organism matures into adulthood, its probability of going on to reproduce increases significantly, because most non-survivors die before adulthood. Therefore, it should be more risk-averse as an adult than as a juvenile, in contrast to the observation from before that, at least in theory, older animals should take bigger risks because they have less future fitness to lose. In particular, I guess the conjunction of these two points predicts that children of r-selected species should take huge risks because they have such low odds of reproducing at all, young adults should be very cautious because they now have better odds of reproducing but haven't yet consummated any of that fitness, and older adults should once again be more inclined toward risk once they have less future fitness to lose by getting injured?

A similar conclusion by Ng (1995)

Ng (1995) draws a similar conclusion as proposed in this section: "the result [...] is interesting since it means that there is a kind of God-made (or evolution-created) fairness between species. Those facing higher probabilities of success will enjoy less relative to the suffering of failure [...] than those facing lower probabilities" (p. 277). However, Ng's argument is a bit different than mine, and I'm not sure it makes sense, since on my reading of p. 277, Ng treats evolution as minimizing the total cost of hedonic experience in a population of organisms, summed over both winners and losers. But I don't see why evolution would care about that directly. Evolution does select for organisms that incur lower metabolic costs per unit of behavior accomplished. But what matters is that the winners of evolutionary competition be efficient, not that the population as a whole be efficient with its total metabolic consumption? I'm a bit confused on this. But I also feel as though metabolic cost isn't the most useful lens for thinking about rewards and punishments. For instance, when designing a reinforcement-learning agent, stronger rewards don't directly have different "metabolic costs"; rather, different rewards favor different kinds of behavior, and rewards are chosen to optimize behavior.

Pleasure and pain don't perfectly track fitness

One Reddit commenter made the following suggestion:

I think happiness and suffering can be meaningfully defined as an organism's response to positive or negative utility over evolutionary time.

So, to say the pain of death outweighs any benefit to life doesn't make any sense to me because it seems to claim that the small animal was not able to procure enough utility to outweigh it's dis-utility.

To say the total dis-utility outweighs the utility would imply that the line of animals was not able to survive and it went extinct.

Given that the animals are not extinct, and that evolution will align happiness and suffering with utility and dis-utility, then I think it follows that there is not net suffering in lines of animals that are not extinct.

One immediate problem with this argument is that it considers only the individuals of a species who reproduce (have positive fitness), when in fact, most individuals of most species die before reproducing, many of them as infants. This is the foundation on which Ng (1995) based his "Buddhist Premise" that suffering exceeds happiness in nature (p. 272, Proposition 3).

Another problem is that evolution doesn't care about the experiences an organism has once it can't reproduce any longer, so there's no direct reason that the pain of being eaten can't vastly outweigh any pleasure that an animal had previously felt from reproduction. (Of course, insofar as the pain of being eaten triggers similar aversive reactions as are triggered by other "slings and arrows of outrageous fortune" during life, the pain of death will tend to be constrained somewhat by the badness of other pains.)

Finally, there's a difficulty that also affects the discussion in my piece so far: While happiness and suffering roughly track evolutionary fitness in proportional terms, they don't track fitness perfectly, and it's not obvious what fraction of experiences will be negative vs. positive. Evolutionary fitness doesn't even really have negative values -- either you reproduce (positive fitness) or you don't (zero fitness), and intermediate actions can only make reproduction more or less likely/frequent/successful.

So even if hedonic experience perfectly tracked fitness, we still wouldn't know where to set the zero point. For an organism whose experiences perfectly tracked changes in expected future fitness, it could be that everything in its life was pleasurable to different degrees, or that everything was painful to different degrees.

While eating is typically pleasurable, going hungry is typically painful. Our brains can induce the fitness-enhancing activity of eating just using pain (hunger). And for some people, especially those with depression, the best points in life are merely relief from what is otherwise suffering; that is, depression corresponds to "gradients of suffering", even when the organism engages in fitness-enhancing activities. Of course, one could claim that depression is abnormal and that healthy individuals of a species will enjoy net positive experiences when they increase fitness in certain ways.

Eating when hungry increases expected future fitness, yet I personally find that the annoyance of hunger outweighs the pleasure of eating. I'd rather never get hungry and never eat than do both (not counting other benefits, like saving time). I think it's nicer not to have any cravings at all than to have cravings that get satisfied, at least when it comes to "brute" pleasures. (In contrast, with intellectual pleasures, I find that their pleasantness tends to outweigh the unpleasantness of craving them.)

In conclusion, appeals to fitness can only be approximations to a deeper, more complex story about what an organism's actual emotions are.

For what it's worth, the Reddit commenter explained that s/he was proposing a non-standard definition of "utility": "I subscribe to an existentio-physical moral philosophy that defines utility as something that confers the ability to exist." However, the commenter also believed that "in most cases happiness and suffering will be accurate representations of existentio-physical utility."

Less value for learning?

One additional thought that might arise is that if short-lived animals have less of a future, then learning is less important for them. As Hedonic Treader noted:

Parts of what our pain does is to provide learning and aversive memory for future behavior. A short-lived organism has a lot less need for that, since the value of learning in that lifetime is smaller. Some functions of acute suffering could be replaced by aversive reflexes rather than suffering [...].

In an article on insect pain, Debbie Hadley makes the same observation:

Insects are pre-programmed to behave in certain ways. The insect lifespan is short, so the benefits of an individual learning from pain experiences are minimized.

It seems to me that this argument doesn't suggest evolutionary pressure away from learning, just that there was never much pressure toward it. If you already have learning machinery, there's probably little harm in keeping it, even if you die a few weeks from now, assuming it doesn't add significant overhead in terms of energy expenditure or risk of interfering with reflex behaviors.

Moreover, this argument does not concern the intensity of pain that short-lived animals would experience conditional on them being sentient, only the probability that they are sentient at all.

Model uncertainty

Keep in mind that all of this discussion is pure speculation on the basis of fitness-learning considerations. It needn't be manifested in the real world. Even if we wanted to take this discussion to suggest that insects may feel less fear and possibly less pain -- rather than taking it to imply that humans feel less pleasure -- we would still have high uncertainty about whether this held true in practice, and as a result, we should maintain a decent probability against accepting these ideas at face value. We should still be very cautious with possible suffering by any given short-lived animal. And of course, the sheer numbers of them imply that even if any given one did suffer less, the aggregate would still be substantial.

Clock speed doesn't change sign of net hedonic balance

I typically suggest that the lives of short-lived animals are worse per unit time than those of long-lived animals because short-lived animals experience painful deaths after a limited amount of life, so even if life prior to death is net positive for them on average (which is questionable), the pain of death can swamp that. For example, suppose (generously) that life is +1 per week, and death is -20. An insect that lives only 10 weeks would have a net balance of +10 - 20 = -10. In contrast, an animal living 100 weeks would, using the same numbers and ignoring the argument discussed previously in this piece, have a net balance of +100 - 20 = +80.

Sometimes it's objected that subjective time might pass slower for small animals, so that their apparently short lives are not so short to them. But this doesn't change the balance of happiness vs. suffering. If a fly experiences its life N times slower than a human would, then it also experiences its death N times slower than a human would. The value of its experiences are all multiplied by N, but the sign of the net balance of happiness vs. suffering remains unchanged.

It's worth pointing out that insects may take much longer to die even in terms of wall-clock time than bigger animals in response to injuries. One reason is that while big animals have a concentrated circulatory system that can cut off oxygen to the brain and other organs within seconds, most insects breathe through tiny openings in their bodies. This article explains why insects are hardier than vertebrates and adds that "Headless roaches are capable of living for weeks." Of course, this hardiness may also modify our assessments of which kinds of injuries cause how much pain to insects.

Acknowledgements

This piece was inspired by comments from Carl Shulman on a Facebook discussion about learning constraints on suffering vs. happiness in wild animals. Hedonic Treader's observation as quoted in this paper was also an idea I had kicking around in my brain.

See also

A Quora answer from June 2016 makes a similar point as my piece about why short-lived insects may feel less pain (relative to typical intensities of other experiences) from certain injuries than longer-lived animals would.

However, I disagree with the simplistic language that the Quora answer uses. Insects clearly do feel pain as a result of a variety of noxious stimuli and can even learn to avoid predictors of noxious stimuli. What's different between insects and humans is, rather, the types of stimuli that cause pain, how protracted the pain is, and so on. It's true that insects should generally not waste weeks recovering from an injury, but they should still feel bad in response to some forms of injury. And they should struggle as hard as they can to escape a dangerous situation, such as a spider's web or a bird's beak.

Imagine an injury that reduces 10% of a primate's expected future fitness -- e.g., maybe breaking off a finger. That event would be extremely painful to the animal. Now imagine an injury that reduces 10% of a cockroach's expected future fitness. A cockroach might also find such an event extremely painful (although whether and to what extent cockroaches react to particular injuries is an empirical question). Sure, a cockroach might be more willing to risk losing 10% future fitness for the sake of short-term benefits because it has less of a long-term future, but that just means the overall stakes per day are higher for the cockroach. If we take evolutionary fitness to be one measure of an organism's implicit utility function with respect to states of the world, then a 10% fitness loss is a loss of 10% of an organism's implicit evolutionary utility whether that organism is a primate or a cockroach.

Footnotes

  1. Indeed, in humans it seems that recklessness decreases with age. This may reflect the fact that young, childless individuals can take risks needed to acquire good mates, but once they have kids, they need to provide a more stable nurturing environment. In this case, the reduced recklessness reflects differential robustness between adults and children (with children being more sensitive), which is an orthogonal dimension for recklessness relative to the risk of losing all future fitness as discussed in the main text.

    Humans do seem to adjust their emotional processing with age -- not for greater recklessness but for more "exploitation" (happy experiences with trusted partners) rather than "exploration" (gathering knowledge) and investment for the future. If this theory is true, it at least suggests that emotional processing in humans can change somewhat with age in ways that make sense. It's not clear if insects can also make such adjustments or if higher-level, rational reflection is required.  (back)