Thoughts on the foundations of consciousness

Last updated Jan 5, 2025.

How can we make sense of the existence of subjective experience? I’m not really sure how to rigorously define this, beyond just pointing at the raw first-person appearances we have all the time. I can gesture at other terms people have used for subjective experience: “what it’s like” to be something, “phenomenal consciousness,” “qualia,” etc. (I’m sure there are subtle distinctions I’m glossing over.) And I can gesture at the kinds of consciousness I’m not talking about: attention, self-awareness, perception, information processing, verbal reports of having subjective experiences, etc. See footnote for more attempts at clarification, if you’d like.¹

Probably nothing I’ll say here is “new,” and I can’t promise it’ll be super comprehensible, given how much ground I’ll cover. But I’m sharing this because I think consciousness is an area of philosophy where it’s easy to make pretty silly-in-hindsight mistakes, and this is the sort of resource I wish my past self had. On the other hand, even if individual views about consciousness have different practical implications, I’d guess we should be so persistently uncertain over (most of) these views that there might not be much practically at stake overall. Still, when we try to hedge our bets against this uncertainty, we should properly understand the implications of each view, not a caricature of it.

Here are some takes, in decreasing order of confidence. Unfortunately this will require a lot of definitions — boring, but necessary for clarity about a topic where people so often talk past each other.

Rejecting illusionism

For the purposes of this post, by illusionism I mean the view that judgments of whether something has subjective experiences cannot be factually true. So this includes both the view that all such judgments are false, and the view that all such judgments are relative, in the same way that categorizing something as a “table” or not is relative.

Illusionism is a position I truly don’t know how to engage with. In philosophy, I’m very hesitant to say “[view] is obviously false,” given how wrong people have historically been about all kinds of things that they considered obvious. As we’ll see below, all the alternatives to illusionism seem to have nontrivial bullets to bite as well, which I suppose is the main motivation for illusionism. Yet, as is often said, the claim that I have subjective experiences seems to be the most undeniable claim I can think of, so I can’t endorse illusionism. I’ve tried talking to illusionists about this, but, as is also often said, we didn’t seem to make any progress in understanding each other.

Does this practically matter? I find it hard to say because I struggle to imagine what it would be like, pun absolutely intended, to endorse illusionism. But here are some musings.

First, to foreshadow the rest of this post, rejecting illusionism might eventually lead us to views with more practical implications(?), like panpsychism or dualism, just by process of elimination.

Second, illusionism seems to lend itself more (though not necessarily!) to the view that our welfare-based moral judgments are entirely pinned down by third-person-observable information. Whereas if there’s some fundamentally not-directly-verifiable morally relevant information out there, i.e., the character of subjective experiences, then many of our welfare-based moral judgments could be indeterminate. (There’s a lot to unpack about that, elsewhere.) For example, let’s say I want to know how much I should care about an insect dying in the wild. Even if I know a ton of details about this insect’s behavior and neurobiology, and even if my moral theory gives equal consideration to suffering regardless of species, for me “how much I should care” is still an open question because I don’t know what the insect is experiencing.²

Rejecting materialism

I used to be pretty sympathetic to (non-illusionist) materialism, the view that subjective experiences, while real, are themselves fundamentally physical properties (e.g., patterns of neuronal activity or perhaps computational activity). Like how water is identical to H2O. It’s hard to non-circularly define “physical,” but I’ll take physical properties to be those that can be expressed as third-person propositions in some sense. So the kind of materialism I have in mind would imply: If we knew and fully understood every third-person fact, then there would be no further question as to why subjective experiences exist, and why any particular configuration of matter does or doesn’t have them. We could logically derive these first-person facts.

As evidence for their view, materialists often point to 1) the long track record of success of scientific reductionism at explaining things, and 2) the well-established relationships between brain states and subjective experiences. That’s what I found compelling about materialism for so long.

But reductionism’s success at explaining third-person facts tells us very little about whether it will, or possibly could, explain first-person facts. And strong relationships between our brain states and our subjective experiences don’t imply that information about brain states alone explains subjective experiences. Here’s the contrast:

• Material explanations of material phenomena are so enormously successful because the only thing to be explained is a pattern of third-person-observable events. If you know how, say, electrical forces work, the chemical structure of DNA (a third-person fact) adequately explains the biological functions that DNA serves. Because these functions are themselves third-person facts. (Idealists, see footnote.³)

• On the other hand, there’s no particular reason that physical facts should suffice to explain subjective experiences, which by definition are not third-person-observable. This is totally consistent with believing that subjective experiences have physical causes.

So for me, (1) and (2) fail as motivations for materialism. From there, I can moderately confidently reject materialism by observing that first-person experience simply isn’t a third-person property. This is what thought experiments like Mary’s Room and the P-zombie argument get at. (I was aware of these thought experiments for many years as a materialist, but my sympathy to (1) and (2) made me dismiss them.)

Okay, so what’s at stake here?

First, above, I said that rejecting illusionism tells us there’s some not-directly-verifiable information relevant to our moral judgments. If we accepted non-illusionist materialism, we might still be optimistic that we (or our superintelligent successors?) can indirectly verify the character of any subjective experience we like, by figuring out neuroscience. Non-materialism firmly shuts that door, by saying that subjective experience is fundamentally inaccessible via third-person observations. So, back to our insect example, suppose we want to compare how hedonically bad dying of predation is for that insect vs. for a gazelle. Then we have to rely on non-empirical premises about how to extrapolate from our own subjective experiences.

I think this is probably the most important upshot of anything in this post, and it’s worth letting it sink in. Even assuming some fixed moral theory of the goodness/badness of different subjective experiences, how good or bad our decisions are for other sentient beings is underdetermined by the empirical data, perhaps radically so.

Second, non-materialism has implications for personal identity. Or as I prefer to call it here, first-person identity, to avoid confusion with stuff like the identity of the Ship of Theseus. If you’re a materialist, it’s natural to think that once you’ve told me which physical things and subjective experience-moments exist in some possible world, you’ve exhaustively described it. Any additional pointers to “me” in such a description would be redundant. Whereas if we reject materialism, there’s nothing metaphysically suspect about further asking, “Where am I [meaning, this first-person perspective] in this world?” Which opens the door to anthropic reasoning.⁴

And from there, there’s also nothing immediately metaphysically suspect about rejecting Parfit-style reductionism about first-person identity. Why couldn’t a sequence of subjective experiences bear some relationship to each other that doesn’t reduce to third-person facts about memories? Of course, this observation isn’t enough. We still need a positive argument for continuous first-person identity, and I’m currently rather confused about where I stand on that. (Very briefly, I think it’s pretty difficult for Parfit’s view to make sense of the extremely commonplace concept of “anticipating” future experiences. More in the footnote.⁵) But my impression is that much of people’s attraction to Parfit’s view on identity stems from thinking that materialism rules out continuous identity.

Rejecting epiphenomenalism?

If subjective experiences are not explainable in terms of third-person structural relationships, then it looks like we’re left with two (or at a more fine-grained level, three) possibilities:

1. Epiphenomenalism: Subjective experiences have no causal power on third-person properties.

2. Mental causation: Subjective experiences do have causal power on third-person properties, despite not being themselves reducible to third-person properties. We can break this down into:

   1. Interactionist dualism: Subjective experiences are a different kind of substance from the kind that has third-person properties. I’ll unpack below what this cashes out into. (If you’re like me, you might be surprised this isn’t the only option if we accept mental causation!)

   2. Dual-aspect monism: There is just one kind of substance, but it has, or at least can have, both a first-person and a third-person perspective.

I don’t think any of these options are clearly absurd, but they each have more or less serious challenges. I’ve slowly been dragged kicking and screaming toward option (2), especially (2b).

Against epiphenomenalism, subjective experiences do seem as if they have causal power. Indeed, as many have pointed out, the reliability of our philosophical beliefs about subjective experiences seems to require that subjective experiences affect our beliefs.⁶ E.g., my motivation for rejecting illusionism is that I immediately grasp the fact that I have subjective experiences. Can we make sense of this, if my subjective experiences themselves have no effect on my belief that illusionism is false?

A reply I used to give to this sort of argument was, “Well, my beliefs and the subjective experiences have a common cause. So my beliefs can be reliably correlated with my subjective experiences, without being caused by them.” I’m not sure this reply works, since my belief in question is about the subjective experiences themselves, not about the common cause. Subjective experiences wouldn’t be evidence against illusionism, much less the decisive evidence I think they are, if I didn’t have direct causal access to these experiences. Importantly, this doesn’t imply that any being who “believes” they have subjective experiences actually does have subjective experiences.

Further, my justification for deeply prioritizing reducing suffering is that I apprehend the nature of suffering from direct experience with it, and want there to be less of it. If suffering, in itself, has no causal influence on my moral beliefs, these moral beliefs are completely arbitrary. I expect the previous argument to be more widely compelling to others, since all epiphenomenalists want to reject illusionism. But given how important my ethics are to me, and how I take my ethics to be grounded in acquaintance with subjective experience, this second example turns the screws pretty hard for me.

On the other hand, on its face mental causation seems pretty spooky. Like anti-scientific, supernatural hocus pocus. Should we believe in souls, per interactionist dualism? Really? Or should we believe everything has subjective experiences, per panpsychism, which (as I’ll discuss later) follows naturally from dual-aspect monism?

Less glibly, it’s not clear where the room is in our best physical and neuroscientific theories for subjective experiences to play a causal role. This is the problem of the so-called “causal closure of physics.” Interactionism predicts that neuroscience alone, no matter how much progress we make, will fail to explain the third-person fact “I believe I have subjective experiences [operationalizing ‘belief’ as a pattern of neural activity].” Which does seem like a bold bet against the reductionist track record. This is importantly different from the prediction that neuroscience would fail to explain why we have subjective experiences, which is much easier to buy for the reasons I sketched above.

Dual-aspect monism doesn’t have this problem, since it says the substance that causes my belief can have both the properties described by neuroscience (via the third-person aspect) and the properties of subjective experience (via the first-person aspect). Though, as I’ll discuss below, it’s tricky to tease apart how exactly this view avoids slipping into materialism.

We might also worry that if we reject epiphenomenalism, the zombie argument against materialism fails. After all, if subjective experiences have causal power, shouldn’t it be impossible for a perfect third-person replica of me to not have subjective experiences? But this isn’t a problem, because the relevant kind of “impossible” is metaphysical impossibility — i.e., impossibility given the fundamental nature of how subjective experiences relate to the third-person facts. Whereas the zombie argument only requires that it be logically coherent that subjective experiences could be removed without changing the third-person facts.

Anyway, I won’t get into the whole debate here. Suffice it to say, I tend to agree with Green⁷ here that the objections to epiphenomenalism are qualitatively similar to those against illusionism. I.e., I seem to directly perceive mental causation, so it’s pretty hard to give up, in a way that doesn’t just come down to counterintuitiveness. So I set aside epiphenomenalism, though more tentatively than illusionism and materialism.

As for practical consequences, I suppose that for epiphenomenalists, ethics would be even more severely underdetermined than non-materialism alone implies. A system’s external behavior wouldn’t provide any particular evidence about its subjective experiences. To be clear, even without epiphenomenalism, the evidence we get from behavior is very inconclusive in absolute terms! We still need to rely on some kind of Occam’s razor and abductive arguments, since the only subjective experiences we can directly access are our own. It’s just that things look relatively hard given epiphenomenalism.

What is dual-aspect monism?

At this point, I’m especially at risk of saying amateurish things, since I’ve only recently learned about dual-aspect monism.

That disclaimer out of the way, it’s worth unpacking what it means for there to be just one kind of substance that has both a first- and third-person aspect. How is that different from saying that subjective experiences are physical, or that subjective experiences per se are causally redundant? To review, here are the key claims about subjective experience I’ve discussed, and how I think a dual-aspect monist would respond:

1. Do we have subjective experiences? “Yes, this is immediately apparent.”

2. If you knew all the third-person facts, would you know why subjective experiences exist and what kinds of experience exist? “No. Subjective experiences are first-person, so the third-person facts don’t logically imply whether subjective experiences exist and what their character is.”

3. Do subjective experiences have effects on third-person properties? “Yes, e.g., the fact that I’m writing arguments against materialism is caused by me having subjective experiences. Well, strictly speaking, it’s caused by stuff that both has a third-person aspect (neural activity) and a first-person aspect (my subjective experiences).”

4. If you knew all the third-person facts, would you know why we believe we have subjective experiences? Or more generally, why we behave the way we do? “Yes, because there’s no separate first-person substance that exists independently of the substance with the third-person aspect. So our beliefs and behavior, as part of the third-person aspect, could be entirely explainable by stuff with a third-person aspect.”

Now, where I’ve gotten stuck in trying to understand dual-aspect monism is an apparent tension between (2), (3), and (4). If the fact “I believe I have subjective experiences” is entirely explainable by neural activity, per (4), then what causal work are the subjective experiences doing? Per (2), the dual-aspect monist is committed to the view that subjective experiences are first-person, hence not logically identical to neural activity. So this looks contradictory with (3).

I suspect the key is to note the difference between (a) “third-person properties are sufficient to explain other third-person properties” and (b) “the substance that causes certain third-person properties to hold [e.g., those of the belief ‘I have subjective experiences’] must only have third-person properties.” (Chalmers makes a similar distinction in Ch. 4 of The Conscious Mind.) Rejecting interactionist dualism only commits us to (a), not (b). Which means we can still reject epiphenomenalism: If the substance upstream of my belief “I have subjective experiences” didn’t have a subjective aspect, that substance would also have a different third-person aspect inconsistent with producing the belief. (Again, as noted above, this is consistent with P-zombies being logically possible.)

That is, we can say that the cause of the belief “I have subjective experiences” has a subjective aspect to it, without saying every explanation of the belief needs to refer to that subjective aspect. The reason we think the cause of the belief “I have subjective experiences” has a subjective aspect to it is that, well, we directly perceive that subjective aspect, and respond to it by making our beliefs cohere with that data. So there’s still a meaningful sense in which the subjective experience has causal power, even if it’s not a separate substance.

That’s pretty nifty as far as it goes. Though, I’m not totally convinced we’ve avoided epiphenomenalism. And we might worry that, by admitting that third-person facts suffice to explain the third-person fact “I believe I have subjective experiences,” we’ve still implicitly conceded something to illusionism. I’m not too troubled by this, however, because our reason for rejecting illusionism was to explain first-person facts, not third-person facts! As Chalmers puts it in Ch. 5 of The Conscious Mind, subjective experiences aren’t an explanation in themselves, tacked onto the third-person facts. They’re the thing to be explained.

Overall, dual-aspect monism seems pretty promising to me, though I’ll reiterate that I’m new to this and I don’t doubt there are some serious difficulties I’m missing. It’s just that these difficulties don’t seem to be as severe as those of the alternatives.

The specter of panpsychism?

A common worry is that dual-aspect monism is a slippery slope to panpsychism. Why? In principle, it’s possible for some-but-not-all stuff to have a first-person aspect, while other stuff only has a third-person aspect. But:

1. We might think this is metaphysically rather arbitrary, and it’s simpler to suppose that all of reality has both aspects (see, e.g., Sec. 5 of Strawson (2006)).

2. If we were to say that stuff only has a first-person aspect under conditions like “neurons [or silicon ‘neurons’] fire in such-and-such ways,” or “material processes occur that we could model as certain kinds of ‘computations’,” we’d have to explain what it is about these processes that suddenly leads to a qualitative change, the existence of a first-person aspect. This is the problem of so-called strong emergence. Strong emergence is problematic in a way that weaker kinds of emergence, like biology arising out of chemistry, aren’t, because biology is merely a different way of modeling things. It’s not a fundamentally new way for things to be. And weak emergence pertains to “fuzzy” properties that admit degrees. But the presence of subjective experience is not fuzzy — it’s either there or it isn’t.

I’m kinda suspicious of metaphysical arbitrariness arguments. Some examples that would take more context to fully unpack: I’m not that compelled by Parfit’s argument that in fission thought experiments, it would be arbitrary for exactly one of the copies to be “you.” Or by the mathematical universe hypothesis, which seeks to explain why our world is real by positing that all possible worlds (well, computable worlds) are real. The argument from avoiding strong emergence does seem compelling, though.

But is panpsychism so metaphysically objectionable, anyway? Several standard reactions to it seem pretty unwarranted to me. Anticlimactically, I’ll set aside the most serious objection, the “combination problem,” because it’s above my pay grade. My rough understanding is that something like this problem isn’t really unique to panpsychism anyway, since everyone has to give an answer as to how our subjective experiences are unified in a sense, though this might be naïve. (Added Jan 5, 2025:) Upon closer reading, I think the problem is that panpsychism is motivated by avoiding strong emergence, yet it’s unclear how you can get individual subjects out of particular patterns of stuff (manifesting as brain activity) without strong emergence.

One common reaction to panpsychism is sheer incredulity. It seems pretty bizarre on its face to suppose that fundamental particles have a first-person aspect. I suspect this reaction comes from conflating “first-person aspect” with “human- or animal-like subjective experience.” The latter is a pretty specific thing that does seem like it should be rare a priori, and arguably would manifest in third-person-observable behavior. Which is presumably why people think it’s obvious that particles don’t have any subjective experience. But the mere existence of something-it’s-like-to-be a particle isn’t that specific at all. And it’s entirely consistent with particles not displaying complex behavior, or having sense perception, etc.

Relatedly, there’s the worry that we don’t have direct evidence that everything has a first-person aspect, so panpsychism seems like a very strong claim. But it’s not as if we’re postulating ubiquitous first-person-ness for no reason. Rather, we’re inferring that first-person-ness is ubiquitous because the only alternatives are less plausible. I’d be pretty sympathetic to a weaker version of this critique that goes, “We don’t know that the set of fundamental theories of subjective experience we’re aware of is logically exhaustive, so maybe we should just say ‘none of the above’,” but that seems to be a rarer position. Further, we don’t have direct evidence that anything lacks a first-person aspect! As Green notes,⁸ the only direct evidence relevant to this question we have is that we have both a first-person and third-person aspect. So the burden of proof arguably lies with those who’d claim that there is stuff out there that doesn’t have a first-person aspect.

1. For example, I asked the language model Claude if there are any differences between subjective experience and phenomenal consciousness. A part of its answer was that subjective experience “encompasses a broader scope including perceptions, feelings, thoughts and interpretations.” This is not what I mean by subjective experience. Perceptions, thoughts, and interpretations can all be reduced to third-person structure and patterns, so they’re not “subjective” in the relevant sense.

Illusionist philosopher Keith Frankish says phenomenal properties are “sometimes characterized as simple, ineffable, intrinsic, private, and immediately apprehended.” I’m not sure about “simple” (I might misunderstand the term), but I would indeed agree with the other characterizations of the thing I’m referring to when I say “subjective experience.”

2. More thoughts on this:

• As an illusionist, you could still think your moral judgments with respect to welfare ought to be indeterminate. Not because of any uncertainty about some non-normative fact of the matter about which things suffer more or less, but because that’s just how your ethics works.

• What I mean when I say “it’s wrong to cause unnecessary suffering” is “it’s wrong to unnecessarily cause there to exist subjective experiences of feeling bad.” I appreciate that we sometimes realize that, whatever we thought we meant when we said “[ABC],” we actually meant “[XYZ]” all along. Or even if we did really mean “[ABC],” we didn’t realize that “[ABC]” in practice coincided with some other “[abc]” that is still true. But I see no reason to think that’s what’s going on in the case of welfarist ethics and subjective experiences. A claim like “even if you don’t have a subjective experience of pain, the pain is still bad” is incomprehensible to me, precisely because (hedonic) welfare and subjective experience are so fundamentally entangled as far as I’m concerned.

3. For idealists, who definitely aren’t materialists, third-person facts are really just first-person facts. But I think my argument here is compatible with idealism, because the distinction is between 1) facts about how my subjective experiences are structured, and 2) facts about the subjective aspects per se of the experiences.

4. In the jargon, “anthropic reasoning” here means anthropic views other than the minimum-reference-class self-sampling assumption. I guess materialists can also endorse such views, but the metaphysical motivation is less clear to me.

5. More thoughts:

• Carlsmith says: ”In my own attempt to inhabit Parfit’s view, it has seemed to me that one alteration it requires is to our normal notion of “anticipation.” Suppose that you’re about to be harmlessly brain scanned, such that an upload will one day be created on the basis of the scan. It’s natural to feel like you should only anticipate the experiences either of your future biological body, or of the upload (or neither) … I now think this much more likely to be the wrong way to think about the situation, and that the right way is to “anticipate” being both people; both the future biological body, and the upload.”

  I truly don’t know what this could mean. You’re not going to be in a superposition of experiences, surely?

• Parfit’s view implies that you have no reason to feel concerned about your future welfare except in an altruistic sense. You can aspire to the most impartial altruism you want, as do I! But that’s consistent with acknowledging that caring about others’ welfare is more of a “leap of faith,” in some sense, than caring about your own.

6. If I understand correctly, this is closely related to the motivation for some arguments for materialism (and indeed illusionism). Yudkowsky takes it to be undeniable that the thing we call consciousness is what’s causally responsible for us talking about consciousness. Of course, even if interactionist dualism is unacceptable (as Yudkowsky thinks), this doesn’t suffice to get us to materialism, if dual-aspect monism is also an option.

7. I’m indebted to many of his writings and podcasts for making me aware of the considerations in this section of the post.

8. Citing Bertrand Russell, but I can’t track down the quote.