Wednesday, July 29, 2009

Having Your Qualia and Eating Your Physics Too

Can we coherently acknowledge the existence of qualia without being forced into a non-physicalist stance about the contents of the world? I'm back at CTY--as I am every summer--and today our philosophy of mind class got to Jackson's "Epiphenomenal Qualia." I was somewhat surprised, having not read the article since last year, to find that my own views on it seem to have changed considerably. Specifically, while I still agree with the main thrust of Jackson's argument (that is, that qualia exist), I'm much less impressed with the quality of his argumentation and the route by which he arrives at his conclusion; more specifically still, I'm incredibly skeptical that his "what Mary didn't know" argument shows anything like what it is purported to show. Qualia certainly deserve to be included in our ontology, but that emphatically doesn't imply that we ought to reject the physicalist picture of the world. Let me try and show how I think these two statements can be reconciled.

First, I suppose a bit of background is in order. Readers may already be somewhat familiar with the Mary case--Jackson's version of the knowledge argument against physicalism--so I won't waste a whole lot of time detailing the moves. Still, it's worth laying out exactly how the argument is supposed to proceed; as we shall see, the precise wording of one of the premises can make all the difference between soundness and total incoherence. Let's start with the informal presentation. Briefly, the standard presentation goes something like this.

Mary is a gifted neuroscientist who has dedicated her life to studying human color perception. She's learned everything there is to know about the physical process of seeing color: she knows everything about how the surface spectral reflectance of various objects interacts with environmental variables to produce changes in the photoreceptors of the eye, how those changes produce neural excitations, how those excitations are processed in the brain, and so on. She knows all the physical facts about how humans perceive color. Somewhat ironically, Mary herself has never perceived color. Her eyes (say) have been surgically altered so that she is only able to view the world in shades of grey. Nevertheless, her studies have proceeded beautifully, and she is now in a position of perfect physical knowledge. With this complete knowledge in hand, Mary undergoes an operation to reverse her perceptual idiosyncrasy; the procedure to keep her from being able to see color is reversed, and Mary's biology is returned to normal. When Mary awakens from the operation, she is presented with a red rose, and actually sees red for the first time. Does Mary learn something new?

On the standard interpretation, we're now presented with two horns of a dilemma: we're either forced to say that no, Mary has learned nothing new when she first sees color--an ostensibly counter-intuitive position to hold--or we're forced to say that yes, Mary learns something new when she sees the rose. If we take this second horn, though (so the argument goes), we must also admit that there are facts about color experience that are not physical; after all, ex hypothesi Mary knows all the physical facts about color vision--if she learns something new by actually seeing color, that new fact must be a non-physical fact. Therefore, the physicalist picture of the world is, while perhaps not strictly false, incomplete in an important way: it is incapable of accounting for the qualitative character of conscious experience. Thus, we must appeal to more than physics when describing a world that contains conscious creatures.

Here's a more formal presentation of the argument (taken from the SEP):

Premise P1Mary has complete physical knowledge about human color vision before her release.


Consequence C1Mary knows all the physical facts about human color vision before her release.
Premise P2There is some (kind of) knowledge concerning facts about human color vision that Mary does not have before her release.

Therefore (from (P2)):

Consequence C2There are some facts about human color vision that Mary does not know before her release.

Therefore (from (C1) and (C2)):

Consequence C3There are non-physical facts about human color vision.

This is, at first glance, a very plausible argument. Jackson's own conclusion was a version of epiphenomenalism: at the time of the article's publication, he held that whatever non-physical knowledge Mary acquired must lack any kind of causal efficacy, thus maintaining the causal closure of the physical universe. That seems to me to be a pretty desperate move, though, and apparently Jackson eventually agreed--he's since recanted this position, and now holds that there must be something wrong with the Mary case. I'm not sure if he's put any work into figuring out what it is, but other people certainly have. I'm going to more or less ignore all of them, as is my wont.

Here's what struck me when I was reading this argument today while preparing to lecture to the class on it: Jackson is deeply ambiguous, confused, or otherwise mistaken about what he means in (P1). The argument never even gets off the ground just because he's wrong about the kinds of things that Mary would be able to know from her particular position in her gray scale world. Let's tease this apart a little more.

What does it mean to say that Mary knows all physical facts about color perception? Presumably, just this: for every predicate, relation, or process P that relates to human color vision, if P is constrained by the laws of physics, then Mary knows P. This should be relatively non-controversial--"physical facts" are those (and only those) facts that are about the behavior of physical systems (and nothing else). The physicalist position is that the set of these facts is identical with the set of all facts that are necessary to explain the workings of the universe; that is, the physicalist position is the position that knowing all the physical facts amounts to knowing everything worth knowing. More narrowly, the physicalist position vis-a-vis color perception is just that knowing all the physical facts about color perception is both necessary and sufficient to give a complete account of how color perception works.

Good. We're homing in on the problem. The next question that we need to answer is this one: how do we go about learning physical facts? The physicalist "bite the bullet" style response to Jackson's argument just denies that Mary learns anything new when she's exposed to color for the first time--it asserts that if she knew all the physical facts, then she'd know what the experience was like. This is not very intuitive; we have a deep intuition that no matter how much I study some subject (via books, laboratory experiments, and so on), there are just some facts--like what it's like to see color--that just won't be accessible to me. That is, we have an intuition that there are some relevant facts that either can't be written down, or can't be discerned through objective experimentation: the what-it-is-likeness of color experience is, presumably, counted among these facts. This is the intuition that Jackson's argument exploits.

It's worth proceeding carefully here, though. Is saying that some particular fact F can't be written down or accessed through objective, third-person experimentation--that is, can't be described from a "view from nowhere"--equivalent to saying that F isn't a physical fact? Can all physical facts (to put it another way) be written down and accessed from a third-person viewpoint? Recall our definition of 'physical fact' above:

"Physical facts" are those (and only those) facts that are about the behavior of physical systems (and nothing else)
Let's rephrase the question, then: can all the behavior of every physical system be represented in third-person accessible formats? If we answer this question in the affirmative, we've adopted the position that Flannagan, in Consciousness Reconsidered, terms "linguistic physicalism," and there seems to be good reason to think that we've made a mistake somewhere in our reasoning. If we answer the question in the affirmative (that is), we've committed ourselves to the following position.

(LP) What it means for some fact F to be a physical fact is for F to be representable in some observer-neutral, third-person accessible form (e.g. public language).

That's a problem, though. If we adopt (LP), then Jackson's argument collapses into something that's trivially true (if not question-begging!).

(1a) Mary knows all linguistic (i.e. third-person accessible) facts about color perception.
(2a) Mary learns something new about color perception when she sees the rose.
(3a) Therefore, there are some facts about color perception that are not representable linguistically.

Of course this is true: it's part of what it means for something to be qualitative (that is, to be a conscious experience) that it's essentially private--that it's essentially accessible only from the first-person perspective. The question, then, becomes whether or not we are justified in adopting (LP); can we give an account of what's going on that doesn't require us to adopt it? Sure: we just have to allow that there might be some physical facts--facts about the behavior of some physical systems--that aren't capturable in third-person accessible representations. If we make this concession, then explaining what's going on in the Mary case becomes very easy: while black-and-white Mary has learned all the linguistically representable physical facts about color perception, this set of facts is not identical to the set of all physical facts about color perception--that is, there are aspects of the behavior of some relevant physical systems that cannot be captured from the third-person "view from nowhere." These facts, of course, are facts about what it is like to be in a certain physical state. To put it another way, there are facts about the state of Mary's own brain--which is, of course, a physical system--that can't be known from a third person perspective: she actually has to be in that state in order to know everything about it. When she's exposed to red for the first time, then, she's adding another bit of physical knowledge--which just is, recall, knowledge about the behavior of physical systems, which includes her brain--to her knowledge-base: that bit of knowledge, though, is one that is only accessible from the first-person standpoint.

Let me try to put this point as simply as I can. The problem with this thought-experiment is that Jackson is mistaken when he says that black-and-white Mary knows all the physical facts. What he means to say is that she knows all the linguistic physical facts--all the physical facts that can be accessed from the "view from nowhere." What Mary doesn't know is the set of physical facts--facts about the physical system that is her brain--that can only be accessed from the first-person viewpoint; she doesn't know what it's like to be in a particular physical state. That's what she learns when she leaves her black-and-white operating room.

To put it one more way, let me just say this. "Physical facts" is a term that refers not to a set of facts that is defined by a mode of access--that have in common something about how they can be known--but to a set of facts that is defined by the sort of system they deal with--that have in common a subject matter, not a kind of access. Physical facts are facts about the behavior of systems for which that behavior is totally describable in terms of the laws of physics, and it makes absolutely no difference (at least as far as we're concerned here) what the mode of access to those facts is. Some (many!) of the facts are expressable in observer-neutral language. Some are not. What matters is not this mode of access, but rather whether or not what is accessed is information about the behavior of a physical system.

Addendum: Please read the comment thread for more on this. Both Mark and Eripsa have given very insightful criticism and show that this argument needs refining. I've done my best to refine it below, and I might post an updated version later on. For now, though, the discussion in the comments is definitely worth following. Thanks to Lally, too, for providing vehement (and helpful) critiques off-thread.


eripsa said...

So, as the Wiki demonstration was meant to show during evening session, clearly Mary knows things other than just the linguistic facts, if linguistic facts just means 'propositions'. She also knows imagistic facts, facts of similarity, conventional facts, and facts about habits and customs and practices, all of which one might plausibly doubt can be fully captured by a series of propositions.

But who said science is propositional? Look at any science article, you don't get a list of clean, articulated propositions stated in the grammar of FOL. You get prose, and symbolic representations of all stripes; you get models and theories that come in chunks that aren't purely linguistic but are also structured in important ways and are arguably nonpropositional; you get charts and graphs and figures and drawings. You get lots of description of the set up of various experiments, or of some working model built, or whatever. And then you have just reams and reams of data.

None of this can plausibly be called 'linguistic facts' with a straight face. You are arguing against a straw man, or maybe Fodor, the form of strawmen.

Mark said...

I'll admit, this summer has been pretty philosophy-free for me, at least in any sort of collaborative setting, so I'm almost desperate to engage in some back and forth here.

First, I'm uncomfortable with the supposedly uncontroversial definition of the physical you propose ("Physical facts are facts about the behavior..." and so on). In particular, I'm wary of limiting, by definition, the domain of physical facts to those facts describable via the laws of physics. Surely it isn't a necessary feature of physical systems that they exhibit lawful (or even lawlike) behavior. There doesn't (to my mind, that is) appear to be any conceptual difficulty with a possible world composed entirely of physical entities, none of which behave in a lawful manner. That our universe isn't such a world is clearly a contingent fact, right?

None of that does much immediate damage to your argument, as I'm sure you could reformulate your definition of the physical to dodge the difficulty, or even deny my contention that the link between the physical domain and the domain of lawful behavior is contingent (trading on the unreliability of our judgments concerning possible worlds very unlike our own).

Granting the definition of physical facts in terms of subjection to the laws of physics, however, you've got a further difficulty, which I think is primarily epistemic. You write:

"The problem with this thought-experiment is that Jackson is mistaken when he says that black-and-white Mary knows all the physical facts...What Mary doesn't know is the set of physical facts--facts about the physical system that is her brain--that can only be accessed from the first-person viewpoint; she doesn't know what its like to be in a particular state."

I'm wary of this paragraph in general, which is too bad, since it's your conclusion. First, what reasons could we give for thinking that the supposedly 1st-person accessible physical facts Mary learns (e.g., what it's like to see a red thing) are lawful in their behavior, at least between persons? This isn't anything more than the cliched "how do you know the color red you see is the same as the color red I see?" late-night dorm room question. Still, it appears that we have an insuperable epistemic barriers to establishing anything like a physical law describing the interaction of the 3rd-person accessible physical facts and the 1st-person accessible physical facts, or even between the 1st-person accessible facts available to me, and the 1st-person accessible facts available to you. And if we can't establish something like a law to govern such facts, we can't take them as physical (under the proposed definition).

Lastly, I'm nervous about the equivalence of "the set of facts about her brain that can only be accessed from the first-person viewpoint" with "what it's like to be in a particular physical state." The "like" in the latter sentence is troubling, since you haven't provided any argument that I can see in favor of the position that "knowing what it's like to be in state S" and "knowing that you're in state S" are interchangeable. That, after all, is the heart of the matter, as far as I can tell. The thrust of Jackson's argument is that "knowing that you're in state S" is, intuitively, a matter of physical fact, but "knowing what it's like to be in state S" may not be. Until we've got some reason to assimilate the two, I think your argument just begs the question.

Jon said...

Good. So propositional attitudes is clearly the wrong way to go here--you're definitely right about that. Still, I think we can make this more precise and still run the objection.

You said before that the patterns of neural activity that count as beliefs--I guess this means a four-dimensional shape consisting of neuron activation over time?--count as representations in virtue of being isomorphic to some feature of the world. If that's really what beliefs are, then it's clearly wrong to say that beliefs are fictions that lack any kind of causal efficacy: they're physical events just like everything else. They differ from other representations only in the mode of presentation--they are (at least sometimes) present as conscious experiences. BDI psychology is right in an important way, then: you do have beliefs, desires, and intentions that cause you to do things even if BDI is wrong in saying that all those are necessarily propositional attitudes.

Still, then, it's plausible to say that (one of) the mode(s) of representation composed by these neural activations--conscious belief--has features that can't be readily translated into other representational media. That is, all the other modes of representation that Mary has access to are (at least in some cases) irrevocably lossy with respect to duplicating all the content of the neuronal activity based--mental--representation.

The kind of representation that results from activity in brains like ours has certain unique information-bearing features that are not easily duplicated with other media. That's not to say (I suppose) that information like the belief with the conscious content 'this is what it is like to see the color red' couldn't ever be represented in a non-brain medium, but it certainly can't be done with anything like what we've got today; we might succeed in making artificial brains that work enough like ours do to have conscious experiences like we do, but we haven't yet.

Even if we do, though, that wouldn't seem to help Mary: there's just no way to represent what it's like for Mary's brain to represent the belief 'this is what it is like to see the color red' without actually duplicating Mary's brain representing that belief; that's what it means to say that the representation is bearing information that's intrinsic to (or non-separable from) the medium.

What Mary has access to in the lab, then, is just all the information about human color perception that can be represented in mediums other than her own brain. That still leaves something out--the information (that is, knowledge of what a particular quale is like) that can only be represented by her neuronal activity. To get that information, she has to actually be given the right (non-representational) environmental stimuli, which means actually seeing something red.

We'll talk more about this tomorrow.

Jon said...

That was @eripsa, by the way. I'll respond to your comment tomorrow, Mark. Thanks!

Jon said...

Mark - Your point about possible worlds is well-taken. I think you're probably right, though, that I'd take the tactic of just restricting the discussion to the actual world: it doesn't do us much good to speculate about the physics of worlds very different from our own, and what it means to be a physical system in our world is (something like) being describable entirely in terms of these law-like relationships.

You're also right about the correlation (or lack thereof) of mental representations between individuals. I think this raises some of the same issues that Eripsa raised above, albeit from a slightly different angle. Still, I think my reply to him stands here: what Mary learns when she sees color is what it is like for her brain to represent (in a particular--that is, conscious--way) something as being red. I don't think the inverted spectrum worry is a serious problem here: whether or not my brain's representation of red and Mary's brain's representation of red share all features in common shouldn't affect the argument. Indeed, I'm very much inclined to agree with you, Mark, and say that they almost certainly wouldn't--that's why I think hunting for a universal neuronal correlate of consciousness is a bad project.

We can certainly pry apart "knowing what it's like to be in S" and "knowing that you're in S." People knew what it was like to be in a brain state that represents (consciously) a tomato being red long before they knew anything about brain states and representations. Because we're good modern physicalists, the two states are (roughly) interchangable--I know that when I'm seeing red, it's because my brain is in a particular state; that is, I know that my "knowing what it's like to be in S" implies my "knowing that I'm in S."

More to come.

Jon said...

Let me try a better formulation of this. Some (all?) representational modes have intrinsic information (maybe self-referential information is a better way of saying it...maybe) that doesn't translate--that is, doesn't apply--when the representation is moved to another medium. If I copy a book to my hard drive, it no longer makes sense to speak of the information represented as being a certain number of "pages long." If I stand on the corner and shout the contents of the book, it makes sense to talk about the representation as having a "maximum volume at time t." Not so if I transmit it on a telegraph via Morse Code. That's not to say that any of these things are non-physical in any interesting way--they're just facts that are peculiar to one or another mode of representation.

Qualitative experience might well be like this. What Mary learns when she exits her black and white room is a (physical) fact about how her brain represents a part of the (physical) world as being a certain way. That knowledge is peculiar to that particular mode of representation: it can't be captured (so far) in any other medium than Mary's brain. That's the source of the intuition that she's learned something non-physical.

The story that Jackson is telling, then, can be read two ways. One the first reading, Mary literally knows all the physical facts while she's in the black and white room, including the idiosyncratic facts about how her brain represents color, and she knows all that without ever having actually been exposed to color in the world. This might be possible if we had some kind of "qualia machine;" a very good simulation (like The Matrix) might suffice. In that case, she wouldn't learn anything new when she left the room: she'd just get the knowledge through the normal causal chain instead of a deviant one.

On the other reading, Mary knows all the physical facts about color perception that can be represented in a medium independent way. That's a lot of information, but it's not the whole story--Jackson's just wrong when he says that she knows all the physical facts. She learns something new when she leaves the room because she learns facts about how her brain idiosyncratically represents objects as being colored. The source of Jackson's confusion is the hidden (and mistaken) assumption that all physical facts are representable in any medium at all. If we deny that assumption, then we can see where things go wrong.

Mark said...

Awesome, that helps a lot. Two more things:

I'm not as sanguine as you are about restricting the realm of discussion to the actual world, for a couple of reasons. First, surely we could imagine a purely physical non-lawful possible world with Qualia, right? In fact, I'd imagine its possible to have a world where all the 3rd-person accessible facts are lawful, but where (some) 1st-person accessible states (such as color representation) exhibit unlawful behavior. In fact, it's possible (although unlikely) that we do live in such a world, although that just follows from my general worries about the difficulty of establishing the lawfulness of a world.

Second, I have a question (for which I already have a tentative answer, and its in favor of your argument):

Does the position you're advancing here eliminate the possibility of reductionism? Specifically, does the 1st-person restriction on the accessibility of color representations imply that we can't, ultimately, reduce such states to the position and velocity of various particles? Let's make Mary even more fearfully intelligent, and have her know the position and velocity of every particle in the universe (to whatever accuracy possible under quantum physics--or have her know all the relevant quantum states, or whatnot) along with all the relevant laws governing the position and velocity of all the particles. Does pre-surgery Mary know what it's like to be in state S (seeing the color red)?

I don't think that this SuperMary case adds anything new to Jackson's case, but it does sharpen the question of reducibility. If, ultimately, Mary's knowledge of what it's like to be in state S reduces to some facts about particles, then she should already know it. Otherwise, we've given up on the project of reductionism (which, frankly, would be no terrible loss).

My tentative answer to the question is, strangely, the reverse of my intuition about the normal Mary case: I think that SuperMary does know what it's like to be in state S, prior to leaving the black and white room. That is, I think too much is hidden away in the "relevant laws" clause given above: if SuperMary has all the information I've granted her, she should know not only the laws governing particle movement and interaction, but a law along the lines of L(SuperMary's Brain): "If brain B(x) is in state S1, then the correlating mind M(x) will be in representational state RS1," or something of the sort.

Of course, laws like L(SuperMary's Brain) won't be general laws--I'm sticking to my guns here and saying that this law won't (necessarily) describe the relation between other brains and their correlating minds (or brain states and representational states). But if SuperMary's as all-knowing as was stipulated, she just knows a whole bunch of laws governing particular relations. On the other hand: are these sorts of laws "laws of physics?" They lack the generality we usually associate with the physical sciences, and we might want to exclude them, although that would also (I think) entail that Qualia aren't physical (since they aren't describable in terms of the laws of physics). So I guess my worry about reduction boils down to my earlier worry about the association of the physical with the lawful.

Michael said...

Jon, good to read you again! And a great comeback it is!

This Mary seeing red thing rings the Searle's Chinese Room bell in my head. In the Chinese Room, the "knowledge" of chinese is in the books, not in the people in the room. The books give the rules for how to mechanically process input chinese text into output chinese text, described in such a way that English speakers can execute them, presumably without ever cluing in on the underlying chinese meanings.

So it is with red. What B&W Mary knows is precisely what could be written down in books about the perception of red. B&W Mary could build a "Red Room" in analogy with the Chinese Room, where the people in the room recieved information about the spectral content of all the light hitting some eye-shaped-and-sized region outside the room, but all that information was presented numerically and graphically using B&W text and graphic representations. But with what Mary knows, she could describe what those eyes would see, including what parts of it would be red, how red and so on.

But in Searle's Chinese room, getting the output from the room "as if" a fluent Chinese person were responding to the input text is far from equivalent to the Chinese Room "knowing" Chinese. A simulation of consciousness is no more consciousness than is a simulation of a nuclear explosion in New York City an actual disaster.

And getting the output from the Red Room "as if" a regularly sighted person was seeing the world is far from an actual person seeing Red.

My point here is not to agree or disagree with anything anyone said, but rather to connect the discussion to other things we already know about.

In the other noise, I can't recommend highly enough that you and your students all read, or at minimum learn about, Jeff Hawkins' book "On Intelligence." Even as the Red Mary discussion started putting brain states into Physical Reality (really I think all these discussions pick away at the idea that Mind is separate from Physics), Hawkins' entire purpose in life is to gain coherent understanding of how an actual human brain is conscious. He has not accomplished that yet, but what you learn about what he has figured out and what he speculates on I think will gigantically inform discussions of consciousness, qualia, and the workings of mind.

I wish I could be in your class(es) and I certainly don't say that to all philosophers! Alas for the next few years I will be a bit busy making money and supporting my family. Youth is hardly wasted on the young, I think, but you don't realize fully what you had when you are in great educational experiences nearly 24/7 until that is in your past. Sort of the opposite of B&W Mary --> Red Mary :)

Maia said...

I enjoyed your blog post and the resulting discussion on the Red Mary thought game. I'm not an academic, so I can't argue with the same robust intellect that you do, but I'm left with the notion that Einstein proved that each individual has its own intertial frame of reference; therefore, there can't be a law to govern first person accessible facts.

I agree with you when you write, "hunting for a universal neuronal correlate of consciousness is a bad project." Indeed. Gives me the willies just thinking of universal correlates on consciousness because the notion ignores first person restriction. We can speculate, but there is no way you or I can represent Mary's perception, her unique perspective. Each is alone in it's own glorious (or inglorious) universe. If I'm missing the point here, please let me know. Thanks for the brain workout first thing this morning!

Anonymous said...

hi every person,

I identified after previous months and I'm very excited much to commence participating. I are basically lurking for the last month but figured I would be joining and sign up.

I am from Spain so please forgave my speaking english[url=].[/url][url=].[/url][url=].[/url]