Reply to Shulman’s “Are Pain and Pleasure Equally Energy-Efficient?”

By Simon Knutsson
Published August 7, 2017; added the section “Other replies to Shulman’s text” September 24, 2022

Carl Shulman published a blog post called “Are pain and pleasure equally energy-efficient?” in 2012 (archived). He makes several points, including the important one that beings could be produced that each experience more pleasure and pain than humans can, which I will not dispute. I will focus on what he writes about the energy-efficiency and intensity of pleasure and pain, the distinction between traditional and negative utilitarianism, and symmetry.

Energy-efficiency of pleasure and pain

Shulman argues that pleasure and pain are equally energy-efficient. More specifically, he argues that the number of positive hedons (units of pleasure) that could be produced per joule or per computation, given computing hardware optimized to produce maximum pleasure per unit of energy, is equal to the number of negative hedons (what he calls ‘dolors’; i.e., units of pain) that could be produced per joule or per computation, given computing hardware optimized to produce maximum pain per unit of energy. Let’s first say what this idea is and is not about.

Statements about quantities of pleasure and pain are in the philosophical literature usually understood as factual statements, as opposed to value judgements, and Shulman appears to focus on this approach in his text.¹And, as usual when hedonism is discussed, pain should be understood broadly as displeasure or unpleasantness. I take it that he treats the question of the energy-efficiency of pleasure and pain as an empirical question about the energy or computation needed to produce specific things with so far undiscovered technology; roughly like a futuristic engineering question.

My main point is that it is difficult to assess his idea about pleasure and pain being equally energy-efficient, because it is so unclear what the positive and negative hedons are supposed to be, and whether such talk is even meaningful. The following is an illustration of what I mean, but I’m not saying it’s exactly analogous. Suppose someone asks which is more energy efficient, fear or pride. We can call a unit of experienced fear a ‘fearon’ and a unit of experienced pride a ‘pridon.’ Is the number of fearons one could produce per joule equal to the number of pridons per joule? An immediate problem is that it is very unclear what we are talking about. What counts as unit of fear and a unit of pride? Without knowing that, we can’t say whether it would require more joules to produce a unit of fear than a unit of pride.

A big challenge about quantification of pleasure and pain is how it could be possible to calculate a sum of pleasure minus pain, or to calculate how many positive versus negative hedons that could be produced per joule, both of which Shulman is concerned with. The challenge is not solved, as far as I know. To explain what I mean, I would say, roughly speaking, that the challenge is not so much about measuring quantities that exist or may exist in the future, but whether pleasure and pain come in the kinds of quantities that allow for addition and subtraction of the two, or for comparisons of units of them produced per joule. If we are going to talk about something as specific as how many positive compared to negative hedons per joule that could be produced in the future with some new technology, we need to have an account of what positive and negative hedons are, especially because it is so controversial whether talk of such quantities of pleasure and pain even makes sense.

For example, one of the main proposals in the academic literature on the quantification of pleasure and pain is to conceive of a unit of pleasure or pain (a hedon) as a just noticeable difference. When it comes to humans, one can perhaps picture some experiment in which subjects report just noticeable differences in moderate pleasure or pain. But even then it’s not precise what a just noticeable difference is, and even less so if we consider, say, fish pain. It becomes even more unclear what a just noticeable difference of pleasure is for future optimized hardware. The question would be how many just noticeable differences of pleasure compared to such differences in pain that could be produced per joule or computation if some hardware was optimized for it. But that presumably depends on specifics about what the hardware needs to accomplish for it to count as that a just noticeable difference has been produced. In addition, although this is not crucial to my point, upon specifying what an optimizer of just noticeable differences would look like, we may not recognize it as experiencing more pleasure or pain (in the morally relevant senses of ‘pleasure’ and ‘pain’) than familiar beings such as humans who experience fewer just noticeable differences. The maximizer of just noticeable differences may strike some of us as having, roughly speaking, cheated, and as not suffering especially severely, despite creating a lot of just noticeable differences.

Overall, I think those who speak about sums of pleasure minus pain, or similar ideas such as the number of positive versus negative hedons produced per joule, should be pressed harder to explain what they are talking about. In the detailed and futuristic case of the energy-efficiency of pleasure versus pain, as in Shulman’s text, there is an especially strong need for more specificity to even start debating which is more energy-efficient. When being more specific, one might conclude that the energy-efficiency of pleasure compared to pain depends on the conceptions of pleasure and pain, and the conception that make the two equally energy-efficient may not be the morally relevant one, or one might conclude that the energy-efficiency question about positive hedons per joule compared to negative hedons per joule is meaningless.²Another point of unclarity is how one is to understand the optimized hardware. Optimized by whom? Optimized in a way that is realistic in the sense of sufficiently likely to occur in the future? Optimized in the sense of what would be physically possible in our universe? Finally, I just want to note that Shulman’s thesis about pleasure and pain being equally energy efficient excludes the cost and difficulty of producing the optimized hardware, including the software used on it. Even if he would be right that optimized hardware could produce as many positive as negative hedons per joule or per computation, creating hardware optimized for pain production may be cheaper or easier than creating hardware optimized for pleasure production.

Distinguishing traditional and negative utilitarianism

Besides arguing that pleasure and pain are equally energy-efficient, Shulman writes, “I would distinguish traditional and negative-biased hedonistic utilitarians in terms of the tradeoffs they would make between the production of hedonium and dolorium.” Dolorium is hardware “optimized to produce maximum pain per unit of energy.”

It is not obvious what he means by ‘negative-biased hedonistic utilitarians.’ One interpretation is that he means weak negative hedonistic utilitarians, who are, roughly speaking, sometimes called negative leaning hedonistic utilitarians. The standard way to distinguish weak negative from traditional hedonistic utilitarianism in the philosophical literature is to say that, according to traditional hedonistic utilitarianism, pleasure and pain have equal value or moral weight, while according to weak negative hedonistic utilitarianism, both pleasure and pain have moral weight, but pain has more weight. This distinction presupposes that pleasure and pain are quantifiable so that we can say that a given quantity x of pleasure and the same quantity x of pain have equal or different value or weight. Shulman seems to assume such quantifiability in his text. But then I don’t see the need to talk about hedonium when distinguishing weak negative from traditional hedonistic utilitarianism, because we already have a simple way to distinguish them. If a given quantity of pleasure and pain are produced by hedonium and dolorium, it is just a special, futuristic case of the more general, standard way to distinguish traditional and weak negative hedonistic utilitarianism. In addition, talking about tradeoffs between the production of hedonium and dolorium when distinguishing traditional and weak negative hedonistic utilitarianism has the drawback of bringing in more potential points of disagreement in a single distinction. For example, two utilitarians who disagree about such tradeoffs can disagree either along the lines of the standard way to distinguish traditional and weak negative utilitarianism, or they might disagree about the empirical question of the energy-efficiency of possible future hedonium versus dolorium, and we would not know where the disagreement lies by looking at the tradeoffs they make between the production of hedonium and dolorium.

A second interpretation is that Shulman is not talking about how to distinguish moral theories like traditional and negative utilitarianism, but rather about categorizing different people, their tendencies, moral outlooks, or the like. He does talk about distinguishing “traditional and negative-biased hedonistic utilitarians” (my emphasis) not utilitarianism. One can do such a categorization if it is useful, which it may be sometimes. One can also categorize people similarly based on other tradeoffs between pleasure and pain such as whether to alleviate, for example, one human’s daylong torture or adding, say, x pleasurable days to the end of y human lives.³This is just a simple choice to illustrate. It would be better to adjust it, make it more specific, and hold other things equal. I most often find it more useful to categorize people or their outlooks along a suffering-focused or happiness-focused spectrum based on such more familiar tradeoffs rather than tradeoffs between the production of futuristic, hypothetical technologies, but which is most useful may depend on for what purpose one wants to make the categorization.

Intensity

It is important to note that Shulman’s text is not about the intensity of pleasure or pain. He mentions intensity a couple of times, but hardware optimized to produce maximum pleasure is not about the most intense pleasure, because if it is optimized from a hedonistic utilitarian perspective, it is optimized to produce the maximum amount (sum) of pleasure; not the maximum intensity of pleasure. It would not be optimized from a hedonistic utilitarian perspective if it created the maximum intensity of pleasure, but could instead create a larger sum of lower intensity pleasure. What ultimately matters to hedonistic utilitarianism (according to standard formulations of utilitarianism) is the sum of pleasure minus pain; not the intensity of either. One can, for example, say, without disagreeing with his text, that optimized hardware could produce equally many positive and negative hedons per joule or per computation, but that pain can be much more intense than pleasure, even if we don’t limit ourselves to humans or other animals. It is an open question whether a greater sum of pleasure or pain could be produced by optimized hardware if it, so to speak, stacks the hedons on top of each other to increase intensity or places them next to each other to get more instances of lower intensity pleasure or pain.

Symmetry

Towards the end of his text, Shulman writes,

Now, the “measurement” of pain and pleasure brings in definitional and normative premises. Some may say they care more about pleasure than pain or vice versa, while others build into their “unit” of pain or pleasure a moral weighting in various tradeoffs. However, if we make use of data such as the judgments and actions of agents in choice problems, quantity of neuron-equivalents involved, and so forth, the symmetry does seem to hold.

Which symmetry is he referring to? The symmetry that he mentions earlier in the text is that “by symmetry, my default expectation would be that H=D.” Note what H and D stand for in his post. Not hedonium and dolorium but “hedons per joule” or “hedons per computation” and “dolors per joule” or “dolors per computation”. So, H and D are not hardware, pleasure, pain, or intensities of pleasures and pains, but rates like miles per hour. So, he is saying that the suggested symmetry that the number of positive hedons produced per joule by optimized hardware is equal to the number of negative hedons produced per joule by optimized hardware seems to hold if we “make use of data such as the judgments and actions of agents in choice problems, quantity of neuron-equivalents involved, and so forth.”

Regarding the quantity of neuron-equivalents involved, we have to be talking about something like the quantity of neuron-equivalents involved in the production of a given quantity of pleasure compared to the quantity of neuron-equivalents involved in the production of the same quantity of pain. But, again, what is a unit of pleasure and pain in this comparison, or, phrased somewhat differently, what is this one and the same quantity of pleasure and pain that we can then look at and investigate how many neuron-equivalents are involved?⁴See the two paragraphs about ‘Science fiction methods’ and hedonometers in my text on measuring happiness and suffering.

Regarding using data such as the judgments and actions of agents in choice problems in support of the symmetry H=D, he presumably means existing data on humans’ or other animals’ judgments and actions.⁵I say that he presumably means existing such data, as opposed to data that does not yet exist about the judgments and actions of agents that may exist in the future (such as artificial agents), because he reports his current belief that the symmetry (H=D) holds if we make use of data, which suggests that this data is available to us. It sounds like he says that he is basing his view about symmetry on data. It does not sound like he is making a prediction about whether data that does not yet exist will support that H=D, but it would be good if he clarified. How does that support that H equals D? I wish Shulman would spell that out.⁶This essay of mine has benefited from discussions with Max Daniel and Lukas Gloor.

Other replies to Shulman’s text

• Vinding, Magnus. (2022). Reply to the “evolutionary asymmetry objection” against suffering-focused ethics