The Puzzle of Psychophysical Harmony


In this post I present the puzzle of psychophysical harmony. Dustin Crummett and I discuss the puzzle in a forthcoming OSPR paper (“Psychophysical Harmony: A New Argument for Theism“), where we argue that psychophysical harmony provides strong evidence for theism. The goal of this post is just to explain what psychophysical harmony is, why it seems to cry out for explanation, and why it seems especially surprising from the standpoint of naturalistic atheism.

For an initial presentation of the problem, it will be convenient to make two metaphysical assumptions about consciousness. Later I’ll argue that the problem remains even if we drop both.

The first assumption is dualism. Experiences are not identical with physical processes in the brain, dispositions to behave in certain ways, or anything like that. Nor do the physical truths metaphysically necessitate the truths about experience. Experiences are linked to physical states of matter by fundamental “psychophysical laws.” These laws are metaphysically contingent. There are worlds with no psychophysical laws, such as zombie worlds (worlds physically like ours but without experience), and there are worlds with different psychophysical laws. Some possible laws would yield spectrum inversion, giving us greenish experiences in the physical circumstances that produce reddish experiences in the actual world. Some possible psychophysical laws would have made us “partial zombies,” giving us (say) visual phenomenology but no auditory phenomenology, or neutral sensory experience but no affectively valenced experience, or sensory and affective phenomenology but no cognitive phenomenology. Some possible psychophysical laws would have assigned experience to different physical systems, as in worlds physically like ours where only rocks are conscious, or worlds where elementary particles are conscious but all macroscopic physical systems are zombies. Some possible psychophysical laws draw from a more impoverished palate of phenomenal qualities, associating all physical states with the same boring gray phenomenology, or with a small set of slightly different ringing sensations. Some possible psychophysical laws are only sensitive to micro-scale physical patterns, not higher-level functional/computational patterns, as in worlds where each specific microphysical configuration of the particles in your brain is associated with a slightly different TV-static-like experience.

The second assumption is the causal closure of the physical. The physical events involved in behavior and brain processing have exclusively physical causes. Taken together, our two assumptions amount to epiphenomenalist dualism, a view that many naturalistically inclined dualists have found attractive (e.g., David Chalmers, Frank Jackson, and Thomas Huxley). On this view, experience is a byproduct of physical activity in the brain, but experience doesn’t exert “top-down” causal influence on brain processing or behavior. Since experience is not causally responsible for our behavior, our behavior would presumably have been the same if the psychophysical laws had mapped physical states onto phenomenal states in different ways (as in the alternative nomological regimes described above). These are controversial assumptions, but to reiterate, they aren’t ultimately required to generate the puzzle.

Now for the puzzle of psychophysical harmony. Psychophysical harmony consists in the fact that experiences are correlated with physical states and with one another in strikingly fortunate ways. One kind of psychophysical harmony is normative harmony, which occurs when the descriptive role of experience is harmoniously aligned with its normative role. The most straightforward examples involve affectively valenced experiences like pain and pleasure. The descriptive role of pain includes the fact that it is typically associated with withdrawal from the pain-causing stimulus, avoidance of similar stimuli in the future, etc. This behavior tends to lead to the elimination or reduction of pain when we have it and the avoidance of pain in the future. Pain also has a normative role: it is non-instrumentally bad; it is an experience we have reason to avoid and to eliminate/reduce when we have it. The two roles are nicely aligned. Generally speaking, pain is associated with the behaviors it gives us reason to perform, the very behaviors it justifies or rationalizes. The same goes for pleasure.

Given our dualist assumption, it didn’t have to be this way. Imagine a hedonic-inversion scenario, a world where the psychophysical laws swap the descriptive roles of pain and pleasure. (Here I use “pain” and “pleasure” to denote, respectively, unpleasant and pleasant experiences generally, though “pain” has a somewhat narrower sense in ordinary language.) In this world, bodily injury results in pleasure, which is followed by withdrawal from the pleasure-causing stimulus and avoidance of similar stimuli in the future. Eating glucose-rich berries results in an unpleasant experience, which is followed by behavior that prolongs the experience and a strengthened disposition to engage in the pain-causing behavior in the future. Here the descriptive roles of pain and pleasure are badly misaligned with their normative roles. We systematically avoid an experience we have reason to pursue and pursue an experience we have reason to avoid.

Or consider a scenario where the psychophysical laws pair all brain states with chaotic, affectively neutral TV-static experience (perhaps a slightly different TV-static experience for each possible microphysical configuration of the brain).  In this scenario, we systematically act in ways that promote certain kinds of normatively neutral TV-static phenomenology, and we systematically avoid other kinds of normatively neutral TV-static phenomenology. Here things aren’t as bad as in the hedonic-inversion scenario, but this scenario is still lacking a kind of psychophysical harmony present in our world. 

Evolution can’t explain normative harmony. Evolution can explain why we have a brain state that selectively responds to bodily damage and causes withdrawal, avoidance behavior, and the like. But evolution can’t explain why the psychophysical laws of our universe conveniently map this physical state onto an experience whose normative role matches this functional role, an experience that rationally justifies avoidance behavior, withdrawal, etc. Evolutionary forces can’t affect the psychophysical laws, so it’s hard to see how an evolutionary explanation of psychophysical harmony could even get off the ground. And given our causal closure assumption, it’s not like the physical behavior/functioning of organisms would have been any different under alternative psychophysical mappings. Since natural selection only responds to physical behavior/functioning, we’d expect evolution to produce the same physical structures under alternative psychophysical mappings, resulting in disharmony. (None of this undermines the standard evolutionary explanation of why we feel pain in response to harmful stimuli. Given that pain is lawfully linked to a functional/dispositional profile involving avoidance behavior and the like, it makes perfect evolutionary sense that we would experience pain in response to harmful stimuli. But this evolutionary explanation presupposes normative harmony; it does not explain it.)

Normative harmony seems very lucky. It seems to cry out for explanation. It doesn’t seem to admit of an evolutionary explanation, but perhaps there are other naturalism-friendly responses to the alleged datum of normative harmony. I’ll consider two, the contingent normative roles response and the normative error theory response.

The Contingent Normative Roles Response

A natural response to the puzzle is to say that the normative role of an experience (e.g, the fact that the pain quale is bad, and that we have reason to avoid it) depends on its contingent descriptive role. One might hold that if a different experience had occupied the pain role, we would have intensely disliked that experience, and in virtue of that fact, we would have reason to avoid/eliminate it and it would have been non-instrumentally bad. Conversely, if the sensation we call “pain” had occupied the pleasure role, we wouldn’t have minded this sensation. We would have liked it. And in that case we would have had reason to pursue it, not to avoid it. It would have been non-instrumentally good. This response goes naturally with an attitudinal theory of pain and pleasure. According to the attitudinal theory, the sensation we call pain is an intrinsically neutral sensation. What makes it unpleasant—what gives it negative valence—is the fact that we take a certain attitude toward it: we dislike it. This fact is also what makes it bad, something we have reason to avoid. Likewise, mutatis mutandis, for pleasure. If a different intrinsically neutral sensation had played the pain role, like a buzzing sensation or the actual occupant of the pleasure role, we would have disliked that other sensation. In that case, it would have been unpleasant, and it would have been bad, an experience we have reason to avoid. The general idea behind this response is that, if the normative role of an experience depends on its descriptive role in the right way, then it’s no big surprise that the descriptive role would nicely align with the normative role.

I think this response fails on reflection. We can distinguish two versions of the attitudinal theory. One version has some plausibility but doesn’t explain normative harmony. The other would explain normative harmony but is implausible. Let’s give the name “Q” to the intrinsically neutral sensation that actually plays the pain role. On one version of the attitudinal theory, sometimes called the “felt dislike” theory, what makes Q unpleasant, and what makes it bad, is the fact that we respond to it with a certain phenomenal state of “felt dislike,” an essentially experiential attitude (one individuated by what it’s like to have it), perhaps an instance of cognitive phenomenology. This view has some plausibility, I think, but it doesn’t solve our puzzle. It merely relocates it. Now the question is: why is the badness-conferring phenomenal attitude of felt-dislike linked to behavioral/functional properties (e.g., dispositions toward avoidance behavior) in harmonious ways? Consider a set of psychophysical laws that invert the “felt dislike” and “felt like” experiences. In this world, we have a felt dislike of the sensation caused by glucose-rich berries, and we thereby have reason to avoid/eliminate it, but we are disposed to act in ways that prolong it and cause it to recur in the future. And we like the sensation caused by bodily injury, and thereby have reason to prolong and pursue it, but we act in ways that eliminate and prevent it. Or consider a set of psychophysical laws that gives us no phenomenal attitudes of felt like or dislike. Here we would systematically pursue sensations we have no reason to pursue and avoid sensations we have no reason to avoid.

The second version of the attitudinal theory is what I’ll call the non-felt dislike view. It says that what makes Q unpleasant, and what makes it bad, is the fact that we “dislike” it in a sense that can be ultimately cashed out in non-phenomenal terms—for example, in terms of a functional profile involving the disposition to behave in ways that would eliminate or prevent Q. The crucial claim here is that what makes Q bad is its relation to facts outside the phenomenal domain (patterns of behavior and physical functioning). But this view is implausible on reflection. Imagine someone being tortured. The experience is non-instrumentally bad, an experience the subject has reason to eliminate. According to the view under consideration, what makes his experience bad is its functional connections to things outside the phenomenal domain, such as the physical behavior that tends to be associated with this kind of experience. To see why this is implausible, consider a disembodied Cartesian mind whose overall experience is exactly the same as that of the torture victim. The experience of the disembodied mind has no causal ties to any physical behavior or physical stimuli (because it has no body or physical sense organs). Still, it seems self-evident that its experience is bad. It is an experience the subject has reason to avoid. It doesn’t even seem conceivable that there should be a perfect phenomenal duplicate of the torture victim whose experience isn’t bad in a non-instrumental way. (It may help to consider this point from the perspective of Cartesian doubt. When you’re in intense pain, you may be unsure whether you’re in a Cartesian skeptical scenario, with all your actual experiences but no physical body. But it’s clear that, even if you are in such a scenario, your pain is still bad.) For this reason, we think it is implausible that the normative role of experience is grounded in factors outside the phenomenal domain, such as functional ties to outward behavior.

While I allow that the normative role of a particular experience might depend on its relations to other experiences (e.g., a sensation might be bad in virtue of its association with a phenomenal attitude of felt-dislike), at least some normative features of experience do not seem to depend on anything outside the phenomenal domain altogether. The latter claim is a consequence of two plausible principles: (1) Necessarily, if one has an unpleasant experience, one has an experience that is non-instrumentally bad, an experience one has reason to avoid/eliminate. (2) Unpleasantness supervenes on total phenomenology. That is, necessarily, any two subjects who are exactly alike with respect to their total phenomenology (the “what it’s like” of their overall stream of experience, not just one part/aspect of it) are alike with respect to whether they have an unpleasant experience. And as long as the experiential truths are sufficient by themselves to ground normative truths like “I am in a state I have reason to avoid,” there remains a puzzle about why experiential truths are linked to physical truths in ways that respect rational norms.

The Normative Error Theory Response

The second naturalism-friendly response, which is not so much an explanation of the datum of psychophysical harmony as a denial of it, is to adopt normative error theory. On this view, there is no correspondence between the normative role of an experience and its descriptive role because there are no normative facts at all. Pain is not bad, and does not give one reason to avoid and eliminate it. The main problem with this response is that it is extremely implausible. It seems about as self-evident as anything in philosophy that excruciating pain is bad, or that at least some pleasures are good. If denying these claims is the price of upholding atheism, the price is too high. A second problem is that, while it sidesteps the puzzle of normative harmony, it does nothing to address the other main form of psychophysical harmony we discuss in the paper (“semantic harmony”— see below). It is therefore a partial solution at best.

Beyond Hedonic Experience

So far I’ve focused on one very specific example of normative harmony: the fact (roughly) that hedonic experiences are paired with behavioral dispositions in ways that respect norms of practical rationality. In the paper, we discuss several other examples of normative harmony that don’t involve hedonic experience, including: (i) cases where sensory experience is paired with behavior/functioning in ways that respect norms of epistemic rationality, (ii) cases where sensory experience is paired with “cognitive phenomenology” in ways that respect norms of epistemic rationality, and (iii) cases where cognitive phenomenology is paired with behavior in ways that respect norms of instrumental rationality, among others. 

We also discuss another kind of psychophysical harmony, semantic harmony, which occurs when experiences are linked to physical states in a manner that induces semantic correspondence. For example, a certain brain state disposes us to make structural claims about our experience, such as “I experience a smooth black expanse immediately next to a sharply contrasting smooth white expanse.” The psychophysical laws pair this brain state with an experience of a smooth black expanse next to a sharply contrasting white expanse. The report you are disposed to make is thereby true. But if the psychophysical laws had instead paired your brain state with TV-static phenomenology, or buzzing phenomenology, or uniform gray phenomenology, or almost anything else, your structural report would have been untrue.

Given epiphenomenalist dualism, an especially striking form of semantic harmony has to do with our metaphysical statements about consciousness. (This is closely related to what Chalmers calls the “meta-problem of consciousness.”) We (those of us who share the usual anti-physicalist intuitions, anyway) have a brain state that disposes us to make metaphysical claims about consciousness, like “I currently possess a property (consciousness) that is (e.g.) irreducible, non-physical, fundamental, modally independent of the physical (etc.).” The psychophysical laws associate this physical state with a state that is irreducible, non-physical, fundamental, and modally independent of the physical. Thus, the metaphysical statements we are disposed to make are true, despite the fact that what makes them true (consciousness) is epiphenomenal. Neither the presence of consciousness nor the fact that consciousness is irreducible (etc.) play any causal role in generating our reports about the presence and metaphysical nature of consciousness. No mechanism in the brain is triggered by the presence of an irreducible (etc.) phenomenal state. If we had lived in a nomologically simpler world without psychophysical laws, a materialist zombie world with only our physical laws, these reports would have been untrue. Likewise if the psychophysical laws had only assigned consciousness to non-verbal physical systems.

Psychophysical Harmony is Surprising on Naturalism

All these forms of psychophysical harmony seem like lucky coincidences, at least from the standpoint of naturalist atheism. It seems that most conceivable sets of psychophysical laws— most ways of pairing physical states with experiences—would have failed to generate anything approaching the degree of psychophysical harmony we find in our world. One might respond that, even if most psychophysical mappings would be disharmonious, the simplest mappings are harmonious. In that case, given an a priori bias toward simpler theories, psychophysical harmony may not be especially surprising given naturalistic atheism. But on reflection, it’s not very plausible that the simplest mappings would be harmonious. The psychophysical laws would be simpler if they mapped all physical states onto the same experience of buzzing noise, or if they mapped each physical object onto a grayscale sensation, with phenomenal brightness proportional to the object’s mass, or if they only assigned experiences to subatomic particles, perhaps with a limited palate of simple phenomenal qualities corresponding to basic physical properties like charge or spin. None of these mappings would induce the level of psychophysical harmony we find in our world. Relatedly, it’s striking that the psychophysical laws, unlike other fundamental laws, appear to operate on relatively macroscopic physical states, such as neural firing patterns or high-level information structures in the brain. As J.J.C. Smart remarked, it is this feature of the dualist’s ultimate laws that give them an odd “smell,” unlike anything else known to science. We might have expected the psychophysical laws to instead be directly sensitive to microphysical phenomena, with our conscious experiences somehow mirroring the microphysical structure of our brains. But that would result in a chaotic and disharmonious mess.

So it seems like psychophysical harmony was very improbable on naturalism. Now, this fact by itself is no reason to doubt naturalism. (The exact arrangement of papers on my desk was also very improbable on naturalism.) However, the naturalist should be worried if there is some other hypothesis H incompatible with naturalism such that (i) psychophysical harmony is not extraordinarily improbable on H, and (ii) the prior probability of H is not extremely low. We argue in the paper that theism is such a hypothesis, though I won’t argue for this here. If theism is such a hypothesis, then a fortiori, so is the disjunction of theism with a certain other non-naturalistic views, such as axiarchism (John Leslie, Derek Parfit), value-involving laws of nature (Thomas Nagel, Brad Saad), and some non-theistic designer hypotheses like Draper’s “aesthetic deism.” Psychophysical harmony seems relatively unsurprising on each of these hypotheses, and we don’t argue that psychophysical harmony favors theism over these theism-adjacent views. But we do think that psychophysical harmony provides very strong evidence against naturalistic atheism.

Dropping the Causal Closure Assumption Doesn’t Solve the Problem

We’ve been assuming epiphenomenalist dualism in the background. What if we go in for interactionist dualism, where in addition to the “bottom up” causal influence of brain activity on experience, there is “top-down” causal influence of experience on brain activity? This doesn’t solve the problem. On this view, the psychophysical laws assign pain a specific nomological role involving certain physical causes and certain physical effects. As a toy model, we’ll assume a nomological role for pain where C-fiber firing directly causes pain and pain directly causes avoidance behavior. The role can be represented as follows:

Nomological Role R: C-fiber firing → X; X —> avoidance behavior.

In the actual world, the laws plug in “pain” for X. But there are various alternative psychophysical laws that assign other experiences to be the occupant of role R. If the laws had assigned pleasure, or a ringing sensation, or really almost anything other than pain, to be the occupant of this role, we wouldn’t have the normative harmony we find in the actual world. Similar points apply to other examples of psychophysical harmony.

Dropping the Dualist Assumption Doesn’t Solve the Problem

Our dualist assumption baked in a certain modal assumption, namely, that the phenomenal truths could have varied independently of the physical truths. Physicalist views of consciousness deny this assumption (as do non-standard versions of dualism that take laws of nature to be metaphysically necessary). One might therefore think that the argument fails unless we assume that physicalism is false from the outset. Even if this were correct, the argument would be significant, since many philosophers reject physicalism for reasons independent of their views on theism. However, in the paper we argue that the success of the argument does not depend on a prior rejection of physicalism. The argument goes through even if we assign a substantial prior probability to physicalist views of consciousness, as long as we grant that there is at least an epistemic gap between the physical truths and the phenomenal truths, such that alternative psychophysical correlation patterns are a priori coherent epistemic possibilities (whether or not they’re metaphysical possibilities). Most physicalists nowadays accept such an epistemic gap.

You can consult the paper for the full argument, but here I’ll just give an initial reason to suspect that the Bayesian reasoning from psychophysical harmony to theism doesn’t require any strong assumptions about the metaphysical modal status of the psychophysical correlations. Consider an ordinary piece of Bayesian reasoning from a message-like arrangement of rocks in your yard to the conclusion that (probably) they were intentionally arranged. Presumably this reasoning should be convincing even to someone who is open to the Spinozist doctrine that, of all the epistemically possible ways for rocks to be arranged in your yard, the actual arrangement, whatever it turns out to be, is the only metaphysically possible arrangement. So long as the alternative arrangements are coherent epistemic possibilities, openness to Spinozism doesn’t undermine the idea that the epistemic probability of a message-like arrangement is much higher on the intentional-arrangement hypothesis than on its negation. This analogy suggests that the Bayesian reasoning from psychophysical harmony to theism (or to the falsity of naturalism) should be convincing even if one is open to the idea that, of all the epistemically possible psychophysical correlations, whichever one turns out to be actual is necessary. More generally, it gives us reason to suspect that this reasoning would not be undermined by an openness to physicalism. For a lengthy and somewhat technical attempt to vindicate this suspicion, I refer you to section 3.2 of the paper.

Would love your thoughts, please comment.x