It’s my pleasure to introduce this symposium on Bernard Molyneux’s paper “The Logic of Mind-Body Identification” (in the current issue of Ergo), with commentaries by Liz Irvine (Cardiff), István Aranyosi (Bilkent), and Jonathan Simon (NYU).
Molyneux’s paper introduces a new strategy for explaining why proposed identities between the mental and the physical give rise to “how-possibly” questions, such as “how could this throbbing pain possibly be nothing but some complicated pattern of neural activation?” Why does considering such identifications lead us to these feelings of incredulity? The most common answer in the literature—called the phenomenal concepts strategy—is that such questions arise because of some distinctive features of the concepts that we deploy in thinking about our experiences from the first-person perspective.
Molyneux’s suggestion is that we can explain why how-possibly questions arise without invoking psychological considerations about our phenomenal concepts. He thinks that the residual incredulity can be explained by considering what logic requires of us when we propose an identity between apparently distinct entities. In particular, he thinks that Leibniz’s Law requires that we reconcile their apparent differences in properties. If we think that a and b are identical, but we think that a has property F and b doesn’t, then logic requires that we either start thinking that b has F (what Molyneux calls addition), stop thinking that a has it (what he calls subtraction), or identify F with some property that we think b has (what he calls facilitatory identification). Facilitatory identifications are governed by the same logical considerations. That is, if properties F and G are apparently distinct and we want to identify them, we must reconcile their apparent differences in (higher-order) properties, either by addition, subtraction, or further facilitatory identification.
Molyneux distinguishes between full-solutions and semi-solutions to problems of apparent differences between entities we’re trying to identify. Full-solutions contain only additions, subtractions, facilitatory identifications, while semi-solutions also contain placeholders, where we make a commitment to reconciling an apparent difference in properties without deciding how to do it. A semi-solution is less of an actual reconciliation between apparently different properties, and more of an assurance that some such reconciliation could be carried out. For this reason, Molyneux thinks that semi-solutions provide no satisfactory answer to how-possibly questions.
He uses this framework to demonstrate a striking result for the mind-body case: that when we represent two things as discriminable (in a certain technical sense), a full-solution will require either that we add or subtract some of their properties, or that the solution go on indefinitely, as we make facilitatory identifications between properties at higher and higher orders ad infinitum.
Molyneux goes on to argue that given some fairly plausible commitments, making property additions and subtractions is not an option in the mind-body case. Thus, we are stuck with two possibilities: a semi-solution, where we commit to reconciling the apparent differences between mind and body without specifying how to do it, or an infinite solution. He thinks that even if an infinite solution is correct, we cannot carry it out (or at least have failed to do so yet), finite creatures that we are. Thus, the best we can have at present is a semi-solution, and semi-solutions leave us with how-possibly questions.
I’m very grateful to Bernard, Liz, István, and Jonathan for their great work, to John Schwenkler for his assistance in putting the symposium together, and to the editors of Ergo for their support. Links to the target article, commentaries, and Bernard’s reply to the commentaries can be found below.
Target article: Bernard Molyneux, “The Logic of Mind-Body Identification”
Commentaries:
- Liz Irvine, “Commentary on Molyneux’s ‘The Logic of Mind-Body Identification’“
- István Aranyosi, “Identity, Identification, and Discernibility“
- Jonathan Simon, “Molyneux’s Problem: Regress in the Logic of Identity?“
Reply to Commentaries: Molyneux, “Remarks on an Identity Crisis“
Hi Bernard,
This is one of those papers that makes you think hard, so thanks!
I would like to comment on the exchange between Aranyosi and you, by rephrasing what I take to be Aranyosi’s point (Istvan, if I’m misunderstanding you, apologies!).
On your (Molyneux’s) approach, if I understand it correctly, if we try to identify Q (a phenomenal property) with P (a physical property), then, since they look different, we need to either add Q to P (which looks like dualism), subtract Q (which looks like eliminativism), or locate two higher-order properties, Q1 of Q, and P1 of P, that look different, and then try to identify them, leading to a regress.
Aranyosi’s point, I think, is that there is a *fourth* option (the phenomenal concept strategy option, or PCS), namely to simply hold that Q and P are identical, and explain why they look different by holding that our phenomenal concept of Q is (e.g.) a directly referential concept (which as such reveals nothing about its referent, according to some authors, such as Janet Levin), and the physical concept of P is a descriptive concept.
This looks like a genuine fourth possibility: there is no subtraction and addition here, and there are no higher-order properties involved, hence no regress.
Thus, in order to for your regress story to work, you need to *assume* that PCS is false (you need to assume that our phenomenal concepts are not special in the way PCS say they are). I take it that this is Aranyosi’s point, and at the moment I find it convincing.
But this (even if correct) may not be a problem for you, because you are trying to provide an *alternative* to PCS, hence you are perhaps entitled to assume that PCS is false (but I feel I’m losing track of the dialectic at this point).
Hi Assaf,
Thank you for your kind words. I did think that I might be missing some subtleties in Aranyosi’s response, so yours may be a very helpful comment. Thanks.
I am puzzled as to how a phenomenal concepts response would explain how x and y seem different *unless*, according to that view, we ascribe apparently-distinct properties to x and y. Otherwise, if we do not, then there is no difference in how we represent them to be. So how could they even seem different in the first place?
I think this is true about properties as much as anything else. If you don’t ascribe apparently-distinct second-order properties to P and Q, then how are P and Q represented as different in the first place? (And if they aren’t, then how are they able to distinguish their bearers x and y?)
But if, according to the phenomenal concepts view, they *are* ascribed apparently-distinct properties, then if we are to coherently say that P and Q are identical, we had better do something about those properties. We can add them or remove them. The third option is to identify them, thus regressing to the next level (where all the same demands will reappear, including the ones that make a PC approach ineffective unless it ascribes differential properties at the *fourth* order.)
To deal with your sketch of Levin; one concept, being descriptive, ascribes properties P, Q, R… to its referent. The other, in directly referring, does not. Then the concepts ascribe different properties in the sense I have in mind (namely, there is at least one property ascribed to one of the identificanda that seems discriminable from all the properties ascribed to the other). This particular suggestion has other peculiarities that we could get into, but it is not a counterexample to the claim that, to explain how x and y seem different, a view must show how they are ascribed different properties.
Assaf and Bernard,
Thanks for the comments. Assaf’s summary of my view is correct. As regards Bernard’s response, which involves what he calls “apparently-distinct properties”, I would say that the phenomenal concept based approach would emphasize that we should not reify appearances of properties as bona fide properties. So, for instance, suppose F=G. And suppose they are apparently distinct. What that presumably means is that the appearance of F is distinct from the appearance of G. According Bernard’s approach, we could then find a distinguishing property, one that is not shared by F and G. For example: F has the property of appearing as distinct from G, but G lacks the property of appearing as distinct from G. Hence, the initial identification F=G is problematic in the way Bernard’s framework would explain it (i.e. there are several options: addition, subtraction, regress, etc.)
But one could just deny that there are such properties as “being apparently distinct”. Consider the following case. John wants to be a US dollar millionaire. As a matter of fact, being a US dollar millionaire is the same as being a Lebanese pound billionaire, in terms of net worth. Yet, to John they appear as distinct. He does not desire to be a Lebanese pound billionaire. We could posit properties like /apparent net worth/, but we could equally argue that “apparent net worth” has nothing to do with net worth at all. The prefix “apparent” changes the meaning of “net worth”, so that it is no more about monetary value as such, but about some imagined/conceived monetary value in John’s mind.
Now, the most obvious reply is to say that apparent properties are genuine properties because, ultimately, they will generate instantiations of mental properties (such as John’s mind instantiating his concept of the Lebanese pound’s net worth), and these mental properties are genuine. But if this is the reply, then we are back to the point that the mind-body case is really special: in general, we should not commit ontologically to appearances of properties, yet, when it comes to the domain of the mental we have to, since appearances are instantiations of mental properties. Then the point is, again, that the property identification puzzle specific to the mind-body problem is not a mere application of Bernard’s more general framework, but really a sui generis puzzle, because it is only the mental/phenomenal realm where we are forced to accept appearances of properties as real properties.
Hi Istvan,
Thanks for your comment. When I talked about things being “apparently distinct”, the word “apparently” was just there to remind us that we’re focusing on the epistemology, not the ontology. I wasn’t trying to carve out a special class of properties of the sort apparently-F.
The rest of your comment makes the phenomenal concepts strategy appear revisionist, if not eliminativist. My experience has the property of being delightful. All that shows up on the fMRI are patterns of brain activity. No delight. If we are to identify my experience with my brain state, what are we to do with this (apparent) property of being delightful? You list the main three options I consider (addition, subtraction and identification) but then suggest that there is a additional response that I did not consider, which is to deny that the delight is real. But I would count that as taking the subtraction option. In denying that the property is real, you a fortiori deny that my experience has it.
I also worry that this doesn’t seem to be in-keeping with the spirit of the phenomenal concepts strategy, which I take to be an alternative to eliminativism.
I feel like in the discussion people are just having (in reaction to Assaf’s comment) there is maybe a distinction that should be drawn, between “appearing to be different” and “not appearing to be the same”.
In some version of the PCS account (that I don’t personnally endorse), such as Janet Levin’s, the PC are supposed not to reveal anything about their referents, but to work as pure labels. In such case, indeed, phenomenal properties are “not appearing to be the same” as physical properties, because they are not presented such that they are ascribed the properties that we usually ascribe to physical property.
But, do they really “appear to be different”? After all, if a PC “says” nothing about its referent (if it “says” the same thing that, for example, a proper name applied to an object), then one could be tempted to say that its referent, if it doesn’t appear to be the same as the phenomenal property, doesn’t appear to be different either. In fact, it doesn’t appear to be anything in particular at all.
This is perhaps why Bernard Molyneux is not satisfied with the PCS, and why he insists (rightly, in my opinion), that, in order for two properties to appaar to be different (and not simply not to appear to be the same), they have to be presented in positively different ways.
If you have a phenomenal concepts strategy on which phen. concepts are pure labels, and no properties are ascribed to the experience or items in your phenomenology , then we arrive at the beginning of the problem just the same. You have to decide what to do with all the properties assigned at the physical end of the equation. But I only see a small impediment to just adding them to the state or property apprehended from the phenomenal perspective. So perhaps no regress.
That means that in assuming that our experiences seem to have positive properties that physical things don’t appear to have, I guess I was assuming the falsity of *certain* PC strategies, as suggested further up the thread. But this is only because I am assuming that our experiences seem a certain way. I think the burden is on anyone who would disagree with this.
i’d think the advocate of PCS (and most physicalists) will agree that it seems (de dicto) that mental properties have higher-order properties (e.g. being mental) that physical properties don’t have, while holding that these higher-order properties are themselves physical (or functional or physically realizable) properties. they’ll just deny that the regress is vicious, for much the reason that jon simon suggests: an adequate reduction of mental to physical will reduce all mental properties at all levels at once, with reductions at all levels following from a finite set of basic principles. of course that’s a nontrivial task, but i don’t yet see strong reasons to think that it’s impossible or that it involves any sort of regress.
What would such principles look like? Say we identify experiential state e with neural state n. Say that experiential state e has the property p of being painful. Say neural state n has the property of stimulating parieto-insular cortex plus the property of causing activity in the anterior cingulate. Which of these latter two properties is the correct identificandum for p? Surely that is an empirical matter, not something that would follow from e=n and some basic recursive principles?
not sure why you think it would have to start from e=n. if one is a functionalist about pain, one might start with e.g. being painful = playing such-and-such functional role (and other functional roles for other mental states, and for being mental in general). pains would then perhaps be things that play the role. being would be being . and so on.
sorry, angle brackets in the second last sentence caused most of it to be dropped. it said something like: being (being pain) would be being (playing such-and-such functional role).
I don’t have to start with neural states. That was just an arbitrary example. Instead, let e be a pain and n a functional state. Then some functionalist says that e=n. Trouble is, e seems to have qualitative properties that n doesn’t seem to have. Sure, we can identify those qualitative properties with (functional) properties of n, but now there is the question of which with which? There are multiple qualitative properties, and multiple higher-order functional properties, hence multiple possible mappings from one to the other.
So we get the same problem. The identity e=u underdetermines the higher order identifications, and I am struggling to think of how it might be supplemented by recursive principles.
sorry, my point wasn’t about neural states. it was that we don’t need to start with identities involving specific pains. a functionalist will start with an identity involving the property of being a pain (or perhaps the property of being in pain) and move across levels from there (down to realizers of the functional property, up to higher-order properties involving the property). to be sure coming up with an appropriate set of identities between specific mental properties and specific functional properties is a hard one — but that’s a within-level problem. given a solution to that problem, it seems to me that the cross-level problem can be handled in much the way i suggest.
(This is a response to Dave Chalmers July 16 3:34. I could find no “reply” button.)
If the functionalist identifies the property of (being a pain) with some functional property then, prima facie, the same problems arise. As far as I can tell, it doesn’t matter that we started one order higher, at the level of properties. It is still the case that the functional property does not seem the same as the pain property. They would seem the same, however, if they were themselves represented as having exactly the same (higher order) properties. Which means they are not. So we must figure out the ways in which we are representing them differently and take corrective measures. When we figure out which properties we are ascribing to one but not the other, this will bring us to the beginning of the problem, where we have the options of adding, subtracting and regressing.
Does that seem right?
sure, it seems that the pain property has various higher-order properties (e.g. being mental) that the functional property doesn’t. the PCS theorist and many other theorists will identify that higher-order property with a physical or physically realizable property. that will lead to an infinite series of higher-order identifications but i haven’t yet seen any reasons to think it’s regressive. the functionalist proposal was just one illustration of a way it could go. perhaps you could give a specific example of a second-order property of mental properties for which you see a potentially regress-generating problem.
Sorry in advance for the long response…
Here’s a caricature of what I think you are asking for. The functionalist identifies the property (is an agonizing pain) with the Ramsified functional property (is the x such that, for all y, if y instantiates x then y is caused by tissue trauma and y causes a desire to not be in a state with property x and y produces aversive behavior towards the cause of the tissue trauma and y causes screaming). Call this property F.
It is observed that the property (is an agonizing pain) has some higher-order properties. E.g. it has the property (entails the presence of a sensation in the bodily space)—a property it shares with (is an experience of physical contact) but does not share with (is a feeling of sadness). E.g., it has the property (entails the unpleasantness of the experience that it modifies)—a property it shares with (is a feeling of sadness), but does not share with (is an experience of physical contact). Let’s call this latter property U so we can refer back to it.
The functional property does not obviously have U. So the functionalist must examine the functional property to see which properties it does have that might plausibly be identified with U. I suggest that making this selection will require further philosophical and theoretical work informed by the latest empirical findings. Some candidates will be better than others, but choosing the best option will be non-trivial. I’ll say why in a moment. For this reason, it is unlikely that the choice can be fixed by recursive principles crafted so as to determine the answer in all cases, and hence specified far in advance of any detailed contemplation of this particular problem.
I think you are suggesting, in contrast, that such principles could be specified. Or at least that you see no reason why not. I agree that, if you can have such principles, then you could complete an infinite amount of work in a stroke just by adding the starting identity to your recursive principles and closing under entailment. But that’s IF you can have the principles.
Here’s why I think you (probably) can’t. Suppose we are looking for a property of F that can be plausibly identified with U. What candidates do we have? Well, F has the higher-order property: (entails that its bearer causes a desire to not be in a state with property F). Whether that’s a plausible identificandum for U will depend on other commitments. If you think that simple creatures (mollusks, say) can have unpleasant phenomenal states even while having no cognitive states like desires, then you would have to reject it. (You would on the same grounds reject the conjunctive property: (entails that its bearer causes a desire to not be in a state with property F AND entails that its bearer produces aversive behavior towards the cause of the tissue trauma).)
So perhaps your best option is that U is identical to the property (entails that its bearer produces aversive behavior towards the cause of the tissue trauma). Others may disagree, but let’s say yours is the right decision. Still, this decision emerged from your other commitments concerning (inter alia) mollusks, and those (if they are justified) will depend on scientific theory and empirical fact as well as philosophical considerations. Any set of recursive principles that is not appropriately sensitive to a range of background considerations will be unable to do the job. Any set that *is* sensitive will not allow us to simply perform a few lower order identifications then close under entailment. Such principles will require further relevant background data to motivate the identifications at the higher orders.
Pretty much everyone will reject identifying U with (entails that its bearer causes screaming), since many unpleasant states do not cause screaming (e.g. the taste of horrible food). But even that depends on knowledge of a common-sense truism not likely to be available to a set of simple recursive principles.
Note that there are other properties that F would likely have that emerge from F’s complex interaction with the environment, with other states, and with itself, that may not be easily predicted from the armchair. Any of these could also be possible identificanda for U. Discovering them will require extensive theoretical and empirical work.
So my position, in summary, is that when you identify x and y (whether they are particulars or properties at whichever order), where x and y seem different, that means that x and y seem to have different properties. Assuming you choose to identify away the differences, you must then decide which “difference” to identify with which, selecting from a menu of possibilities. This will involve (i) creating a sufficiently exhaustive and accurate menu of properties to begin with (making sure, e.g. that some complex, non-obvious properties are among the possible selections) and (ii) motivating one selection over the others from this menu. I claim that both are philosophically and empirically non-trivial endeavors.
it seems to me that identifies involving the higher-order properties you mention (e.g. entails the presence of a sensation in the bodily space, entails the unpleasantness of the experience it modifies, entails that it’s bearer causes such-and-such desire) will all be settled by identities involving first-order mental properties (e.g. has a sensation in the bodily space, is an unpleasant experience, has such-and-such desire) perhaps along with some background empirical facts. to be sure, coming up with a complete set of such first-order identities will be a very hard project. but given such a set of identities, i don’t see the special further difficulty in coming up with higher-order identities.
But consider a statement that merely told us that (necessarily?) for all experiences x, if x has the property (is an agonizing pain) then x has the property (is horrible). This statement captures a certain regularity, but in eschewing mention of second order properties it appears to leave something out. Namely, that there is *something about* the property (is an agonizing pain) that *explains* why it is always accompanied by the property (is horrible). On the face of it, “something about the property p” appears to be a place holder for a second order property.
“that’s a within-level problem. given a solution to that problem, it seems to me that the cross-level problem can be handled in much the way i suggest.”
I am not sure why you think this. If property x has second order properties X1, X2 and property y has s.o. properties Y1, Y2, then x=y doesn’t tell us whether X1=Y1 or X1 = Y2. (That’s assuming we should identify at all. )
Perhaps I am missing some subtlety here.
Also, are you still speaking for “most physicalists”, or does this answer require one to be a functionalist?
I wonder if the question of whether Conway’s ‘Game of Life’ can become complex enough to figure out its own neighborhood rules, can be shown to lead to infinite regress.
Uzi.
That’s an interesting thought. It might if it becomes complex enough to acquire two perspectives on the world, and then tries to identify elements in one perspective with elements in the other. If it tries to do that without sacrificing the commitments of either perspective, the elements for a potential regress are there.
In light of the exchange between you (Molyneux) and Chalmers, here is how things appear to me. What I say here is of course tentative, raw, etc. I need to think about it more.
Take an extant reductive theory of pain, for example Cutter and Tye’s representational theory of pain, according to which pain is basically a (sensory) representation of bodily disturbance as bad. The property of representing a bodily disturbance as bad is supposed to account for the painfulness of pain (as well as for the bodily aspect of pain). This account may involve at some point a higher order property (as you suggest), but this is not the crucial point, it seems to me. The crucial point (I think this might be Chalmers’ point, but I am not sure) is that if we agree that Cutter and Tye’s view is correct and exhaustive (i.e., accounts for all the significant/striking properties of pain), then it is reasonable to expect that any remaining higher order property will be a mere logical construct (in a sense I now illustrate) of the properties they’ve already reduced. For example, on the basis of the property of unpleasantness we can logically construct the property “entails the unpleasantness of the experience it modifies”. It seems that if we have already carried the empirical work needed in order to reduce the first property, there is no extra empirical work we need to do in order to reduce the second.
So what I think you need to provide is a specific example of a higher-order property that (a) is *not* on the usual list of properties (and relations between them) that theorists of pain (e.g., Cutter and Tye) explicitly try to reduce, and (b) is not a mere logical construct of properties in the said usual list. In other words, the property should be substantive in the sense that we need further empirical work in order to deal with it. So far it seems that the examples you have provided do not satisfy both (a) and (b). Examples like “entails the unpleasantness of the experience it modifies” do not satisfy (b). The example you gave, of the higher-order property of the property of being an agonizing pain, which explains why agonizing pain is horrible does not satisfy (a), because Cutter and Tye *in any case* need to explain why a state representing bodily disturbance as bad makes the state painful (or horrible), hence it is a part of the “usual list”.
I’m not sure about this at all. Specifically, I worry that the notion of a “usual list” is suspect, perhaps even question begging. In any case, I think that your answer to this could help me understand better what is going on.
Thanks!
Hi Assaf,
Apologies for the slow response. I think that you make a key move here:
“if we agree that Cutter and Tye’s view is correct and exhaustive (i.e., accounts for all the significant/striking properties of pain), then it is reasonable to expect that any remaining higher order property will be a mere logical construct … of the properties they’ve already reduced.”
If we agree that a view is correct and *exhaustive*, then yes, we thereby agree that everything has been taken care of. But I suggest we will struggle to get to a point where a view is regarded as correct and exhaustive, because doing so requires us to complete a potential infinity of steps.
“It seems that if we have already carried the empirical work needed in order to reduce the first property, there is no extra empirical work we need to do in order to reduce the second.”
Perhaps. But that may be because the right way to understand the regress I describe is as one where no step ever gets fully completed because some other task must always be completed first. So to make a full and coherent identification at level 1, you *first* have to make a full and coherent identification at level 2; and to make one at level 2, you *first* have to make one at level 3. And so on. So I agree that, as a precondition of succeeding at level 1, you must have succeeded at all the higher orders. But then this explains, I think, why we struggle to succeed at level 1. (Alternatively, we provide a semi-solution at some level as a stop-gap. But then you don’t get the higher-order identities as an entailment.)
So I think that, with respect to the second half of your post, I am not obliged to satisfy criterion (a). The properties to which the theorist will regress may be among the usual suspects. Or at least I need you to motivate criterion (a) some more.
To apply this general framework to specific accounts, e.g. like Cutter and Tye’s, you first have to decide on what the proposed identificanda are. (A complication with representationalism is whether we should be interested in identifying elements at the representing or the represented end of the equation, but I am not going to get into it here.)
The next question is whether the supposed identificanda seem indiscriminable. E.g. does the property (is unpleasant) seem exactly indiscriminable from the property (is apt to harm you)?
First Horn: It does. Then there is no problem, and we can all go home.
Second Horn: It does not. Alas, then the bad news is that these seeming differences between the identificanda cannot be removed within a finite number of steps without “going revisionist” (which, for the reasons I get into in the paper, will probably mean either providing a super-physicalist ontology or denying the phenomenology) .
The troubles on the second horn are what my paper focuses on. I am, I admit, broadly assuming that we can dispense with the first horn prior to the discussion, since the first and third-person identificanda never seem to be exactly indiscriminable from one another. But that may be where I go wrong.
Hi Bernard,
Sorry for coming late to the party! Here are some initial responses to your reply to my comment.
1) On Recursive Specifiability:
A first point is that we should distinguish between the question of whether it is a supertask to believe a full preservative solution, and the question of whether it is a supertask to know (or be justified in believing) that a given one is correct, and the further question of whether it is a supertask to know one in the ordinary way, ie, via empirical spadework.
As I initially understood you, Bernard, your challenge was that even believing (cognizing) such a solution would be a supertask. But this business about empirical spadework is a red herring as far as that is concerned. Similarly for the mirror question about representing as discriminable. I can represent as discriminable just by taking every property of P to be distinct from every property of Q. Likewise I can represent as identical just by specifying some scheme of identities. The specification can be arbitrary if the task is just to represent , and I am not concerned with getting it right. My initial thought was that, to the extent that I am representing an infinite hierarchy of levels of properties of properties to begin with, ill probably have some recursive way to formulate a scheme of identifications that run all the way up in a recursive manner, justification be damned. Likewise for representations as discriminable.
The question of justification makes matters more complicated. Lets first note that this is you falling back to a weaker claim (though perhaps this is all you ever meant to be defending): it is not representing as identical or different as such that leads to regress but obtaining sufficient justification for believing in such a representation.
Lets also note that if this is our concern then quite clearly, representing as discriminable is every bit as difficult as identifying. Or rather “finding a full solution to a representing as discriminable problem” is every bit as difficult as finding a full preservative solution to an identification problem. For youll need justification for every discrimination you draw at every level of the hierarchy just as youll need justification for every identification you draw (ill return to this below).
Anyway, broadly speaking I agree with Dave here: I don’t think you’ve given us any reason for thinking that we need to gather infinitely much evidence or do infinitely much spadework. Maybe the spadework has to happen at several levels. Maybe it has to happen at 100,000 levels (though I highly doubt it!). But the recursion claim here is just going to be the claim that there is only finitely much spadework to be done at finitely many levels, and then AFTER that we can recursively formulate the rest. No one is claiming that logical tricks can replace the needed empirical spadework.
Certainly, none of your examples offer us any reason for thinking otherwise. Here it might behoove us to get a little clearer on ontology. Are we talking about token mental states (“my pain”) or properties of such mental states (the painfulness of my pain)? Or are we talking about eg the property-of-being-in-pain?
I take you the first way, but note that the painfulness of your pain is a first order property of that token pain, not a higher order property. So your enumeration of higher order properties in your response: “my pain seems horrible, it seems sharp, it seems to be in my foot…” doesn’t even get us to the second order. These are all still first order properties of your token pain. Plausibly the horrible-seemingness of your pain itself has properties (as various accounts of pain like the Tye-Cutter account hypothesize) but it is doubtful that for every level there is further empirical spadework to do to settle things at that level. The onus, in any case, falls on you here.
Lets get even clearer on the ontology. What is your metaphysics of properties? If you think properties are universals, and universals are sparse, then it is likely that there will not, in general, be an infinite hierarchy of properties at increasingly high levels. More generally if properties at sufficiently high levels are grounded in properties at lower levels, it is reasonable to suppose that empirical spadework will only be needed at lower levels. I say ‘reasonable’ – not everyone thinks that grounding entails scrutability, so perhaps you might think there are facts about properties at arbitrarily high levels that are grounded in but not scrutable from the facts about properties at lower levels. But we have yet to see a positive argument for why this might be so.
Of course you are free to argue that this is so, or indeed that there are facts about higher level properties that are not even grounded in facts at lower levels. Then you may be able to defend your claim that there will be spadework of some kind at arbitrarily high levels. Even then you do not get this for free, since as I pointed out in my comment there are still some recursive constructions that might summarize how the identifications go, but if the property hierarchy is sufficiently unruly these constructions wont capture everything. My point here, though, is that this is some pretty speculative stuff you’d have to buy into to support your thesis, and if you are correct I suspect that finding full solutions to identity problems will be the least of our concerns.
2) Reversing the Problem
I take your point that the challenge in finding full preservative solutions to identity problems does not hinge on one having antecedently succeeded in representing-as-discriminable.
This doesn’t really address my concern though, which is that representing-as-discriminable will be just as hard as finding full preservative solutions to identity problems. And this is a challenge to you because you are supposed to be showing us something that draws out what makes the hard problem of consciousness so hard, making use of special features of the case of psychophysical identity. Your suggestion is that there is a certain sense in which cognizing such an identity in full , or finding justification to believe in such, is a supertask. But it looks like, by your lights, cognizing (or justifying) any non-identity in full is equally a supertask. Yet the idea that two things are distinct doesn’t seem to give rise to any hard problem. So something has to give.
Another point: your tree diagram does not necessarily identify any kind of regress. It depends on what else we find out about these properties when we do the (empirical) spadework! For the reasons Ive already given, the onus lies on you to show otherwise, and you have not yet done so.
Hi Jonathan,
I’ve been thinking about your comments for a few days. I’m going to break them up a bit since you made several points.
I agree that, provided you aren’t concerned about justification, you may well be able to recursively specify a solution. E.g. you could define a (silly) routine that defines the alphabetically nth property on one side with the alphabetically nth property on the other, when those properties are given their usual English names. Something like that.
I don’t yet concede that you can recursively specify a routine that you have any good reason to think is correct.
Still, you’ve identified a place where more work is required. And the answers will make a difference to how we might put the account to work. On the face of it, it won’t find it so easy to predict that there will be no conceivable solution. But we might still predict that there will be no conceivable solution that meets a certain standard of justification or plausibility.
You go on to note that:
“it is not representing as identical or different as such that leads to regress but obtaining sufficient justification for believing in such a representation.”
I think I would resist that way of putting it. I’d rather say there is a cognitive regress, but that regress might not be *vicious* if it can be navigated using recursive methods. And there is a justificatory regress, that appears on its face to be outside of what can be conquered using recursive methods, and hence on its face to be vicious.
Now you say that I have not given you any reason for thinking that we need to gather infinitely much evidence or do infinitely much spadework. Ok, I’ll try to do better. But I think it might be handy, first and foremost, to give an overview of the project. Partly because I can conceive of other ways to view the project (that I want to tacitly reject). And mostly because I think it will matter when determining the burdens. Roughly, I see my argument as:
(1) There are certain kinds of regress that can happen with identification problems sometimes, at least in principle. (part 1 of the paper)
(2) There’s some good prima facie reason to think that the problem of consciousness might give rise to such a regress. (given our reluctance to naysay either phenomenal introspection or the scientist).
(3) If this is indeed the problem of consciousness, it may explain why that problem is so tough, why it gives rise to the landscape of attempted solutions that it does, and why it is attended by “how-possibly” questions. (For the last, I show that how-possibly questions might arise from the absence of a finite solution, assuming certain principles about how little we are willing to go against introspection and science).
(4) Therefore this is a plausible and promising account of what the problem of consciousness is, worth further investigation.
I do not see myself as attempting to *prove* that this is the problem of consciousness. Rather, I am motivating the hypothesis, and showing that it may help explain the data. The only place where I might meet a demand for proof is, perhaps, in establishing that regresses of a *certain* kind can attend the attempt to identify things, sometimes, when certain conditions are met. That takes me a long way towards establishing (1). But even there I would not expect to *deductively prove* that the regress is one that involves infinitely many a posteriori justifications.
Let’s file that under “managing expectations”.
With that in mind, I can tell you why you ought to worry that the problem requires infinite amounts of empirical work. Start by noting that the following argument is invalid:
(1) x = y
Therefore, where X is a property of x and Y is a property of y:
(2) X = Y.
For sure, the identity of x and y entails that each property of the one is identical to *some* property of the other, but it does not tell us that any X in particular is identical to any Y. Now I take it that in being invalid, the inference from (1) to (2) is not a priori. So if you want to get from (1) to (2) you will need something a posteriori. In other words, empirical spadework. But this problem will repeat, over and over, at each and every order. So this at least suggests that we might have an infinite empirical mountain to climb.
Still, it’s not a proof, since there is a possibility that hasn’t been ruled out. Specifically, it may be that whatever a posteriori evidence got you to (1) was enough to also get you to (2). If that happens in every case then we have struck lucky. And I have nowhere proven that that won’t happen. But it doesn’t happen at the lower orders. Which (at the very least) means it doesn’t happen universally.
Still, let’s suppose that, *quite often*, when you have a posteriori evidence for x=y, you have the a posteriori evidence you need for X=Y. Well how often? Not *all* the time. We know that. So let’s be harsh on my hypothesis and say 99% of the time. If that distribution is reflected in the regress then, roughly speaking, after every one hundred orders or so we will run into a gap where more empirical evidence is needed. But that still means that there are infinite amounts of empirical work required.
Again, it’s not a proof. But it is some reason to worry that the problem is infinitely hard.
“Lets also note that if this is our concern then quite clearly, representing as discriminable is every bit as difficult as identifying… For youll need justification for every discrimination you draw at every level of the hierarchy just as youll need justification for every identification you draw (ill return to this below).”
Not sure that’s true. It may be that our start state is unjustified (cf. the state of a Bayesian agent prior to any updates; or a Popperian hypothesis prior to any corroboration). It may also be that we default to not representing x and y as discriminable and that, in defaulting, we are justified. (E.g. do I need justification for my not representing Cnut as physically indiscernible from William Wallace?)
Re: Higher order properties.
English isn’t well suited to talking about higher order properties, and we don’t seem to like thinking about them. Moreover, we have ways of shifting things through the orders. We can talk about properties as if they are (Platonic) objects that relate to physical objects via an exemplification relation. Does that mean that, apart from this one relation, everything is an object? That would mean that there is nothing at the first order. But I am not convinced. By the same token, I am not convinced that, just because we can often (always?) redescribe a second-order property as if it were a first, that means there are no second order properties.
Now among the examples I suggested were that the pain was horrible, sharp and in my foot. You said these were first order properties. I agree, I was either being lazy or avoiding the monstrous English that occurs when trying to be more precise. Earlier on the blog, though, I was more careful. Instead of the first order property (is horrible) I focused on the second order property (entails the unpleasantness of the experience that it modifies). That’s a second order property.
Now you might argue that we can do away with this second order property, and make do with the first order property (is horrible). But I suspect we will need to operate at the second order. Why? Because if we do things at the first order we will need to add a law to the effect that when an experience is an agonizing pain it is horrible. But even such a law appears to leave something out. Namely, that there is *something about* the property (is an agonizing pain) that *explains* why it is always accompanied by the property (is horrible). On the face of it, “something about the property p” appears to be a place holder for a second order property.