Tuesday, February 03, 2015

Information and Parfit's Fact Stating Argument

In Chapter 26 of On What Matters (vol 2), Parfit sets out his (comparatively neglected) 'Fact-Stating Argument' against non-analytical moral naturalism.  This begins by distinguishing the referential and informational senses of "same fact".  Consider the following three claims:

(J) Shakespeare is Shakespeare
(K) Shakespeare and the writer of Hamlet are one and the same person.
(L) Shakespeare wrote Hamlet.

Parfit explains:
In the referential sense, (J) and (K) state the same fact, since both claims refer to Shakespeare and tell us that Shakespeare has the property of being numerically identical to himself.  In the informational sense, however, (J) and (K) state different facts. Unlike (J), (K) refers to Shakespeare in a way that also tells us that Shakespeare wrote Hamlet. In the informational sense, it is (K) and (L) that state the same fact.

Parfit then goes on to consider the moral naturalist's thesis:
(Q) moral rightness is the same as some particular natural property
Which would be referentially equivalent to the tautology:
(R) this natural property is the same as this natural property.

Parfit writes: “To defend their claim that (Q) states an important truth, these naturalists must therefore claim that, in the relevant, informational sense (Q) and (R) state different facts.”  This then provides the basis for his Fact Stating Argument.

In rough gist, the argument challenges the naturalist to explain what natural fact is informationally equivalent to (Q) -- or to any other true positive normative claim like "Torturing babies is wrong."  If there is no eligible natural fact available, it looks like the naturalist is committed to non-natural facts after all.  If, on the other hand, the naturalist proposes a candidate natural fact, it looks like they're giving up the "non-analytic" part of their view: the information provided in their candidate natural fact would pin down an analytical reduction of 'moral rightness'.  So, non-analytical naturalism is untenable: under pressure it must transform into either analytical naturalism, or non-naturalism.

Can non-analytical naturalists avoid this conclusion? The key move seems to be in the assumption of a correspondence between information and facts.  Informative claims that referentially correspond to trivial facts must also -- informationally -- correspond to some other, less-trivial fact.  This is, in effect, the fact that we learn when we acquire the information in question.  This seems like an innocent enough assumption.  But if we remember the naturalist's commitment to thinking that all facts are natural facts, then we find an immediate conflict between what we might call this information-fact correspondence thesis, on the one hand, and the core commitments of non-analytic naturalism (irreducibly normative claims / concepts / information without any irreducibly normative facts) on the other hand.  Is there any principled case to be made for abandoning the former?

It seems to me that the information-fact correspondence thesis is highly plausible, and provides an important method for disciplining our metaphysical practice. (It's too easy to claim to be a naturalist these days!) This principle places serious constraints on what phenomena naturalists can feasibly hope to accommodate -- which seems only appropriate for what is, after all, a very metaphysically constraining thesis!  A similar argument could easily be constructed to cast doubt on non-analytical ("type B") materialism in the philosophy of mind, for example.

Thoughts?

16 comments:

  1. Hi Richard,

    It's an interesting argument, but I think the natural/non-natural distinction might be a problem here (granted, there might be an easy way out by redefining the terms, but I'd like to ask how they're defined in that case).
    On that note, Parfit says "A fact is natural if such facts are investigated by people who are working in the natural or social sciences"-
    Let's consider antisocial personality disorder (APD). Whether a person has APD, its causes, etc., are all facts investigated by some people working in psychology, so it would seem to qualify. So, it's a natural fact whether someone has APD.

    Yet, APD is defined as a "pervasive pattern of disregard for and violation of the rights of others" (Source: https ://en.wikipedia.org/wiki/Antisocial_personality_disorder ), which seems clearly a definition in moral terms.

    Granted, the definition also gives a list of behaviors as evidence of APD. But it seems to me that there are moral components on the list too - when properly interpreted.

    Purely for example, one may consider conditions 7 ("rationalizing" and "mistreated"), 5 ("reckless") and even condition 1 (e.g., a gay person having a same-sex partner in a place where gay sex is both socially condemned and criminalized would not meet condition 1, it seems to me, due to an implicit moral requirement on the social norms the person violates).

    Granted, one might come up with a complicated way describe all in non-moral terms, but it seems very difficult, and perhaps even not achievable in a lifetime, since that would seem to require a description of what is morally justified or not in zillions of cases.
    It seems to me that some moral judgments would be likely guiding any assessments of which cases meet some of the conditions - and that's a feature.

    Now, that may not seem to be a philosophically intereting issue: arguably, if psychologists use definitions in moral terms, that does not tell us anything interesting about metaethics. But if that is so, I'm not sure how a distinction based (by definition) on what people who are working in the natural or social sciences do, is relevant.

    ReplyDelete
  2. Richard,

    I've been trying to find around the issue of APD, and considered Partif's other definition (of a normative fact that is 'natural' in the reductive sense).

    Parfit does not think that the vagueness of 'natural' would matter because as long as a normative fact can't be reinstated in non-normative terms, it's not natural and that's that. But if one also keeps the general definition of 'natural fact' (the argument seems to require a definition of 'natural', in my view, though I guess this might be debatable), I'm not sure how this would be of help, because the previous definition seems to render some normative facts (e.g., involing APD) natural, even without reinstating them. There seems to be a conflict between the definitions.

    I guess one might simply leave aside science-based definitions and define 'natural fact' as one that can be stated in non-normative terms. I'm not sure this would work, but it seems to me it's the most promising option for Parfit's argument. Please let me know if that's what you have in mind (or which definition of 'natural' you prefer, if you prefer a different one)

    ReplyDelete
    Replies
    1. Right, it's just the reductive sense that's relevant here. It doesn't really matter how we understand 'natural'; the question is rather whether normative facts are primitive, or whether they're reducible in the way that tables and chairs are.

      Delete
  3. Okay, so I would have to give it more thought, but for now, it seems to me that if it's successful, it would seem to work against pretty much any non-analytic reduction without an analytic reduction to connect them.

    For example, if naturalism about water holds that there are irreducibly watery facts (the same for non-naturalism with respect to color, etc), non-analytical naturalism about water could be defeated by an argument as follows:

    (QW) Water is the same as some particular substance S, described in non-watery language (say, H2O).

    (RW) S is the same as S. (or, if you like, H2O is the same as H2O).

    Then, the challenge for the water non-analytical naturalist would be the same as in the moral case.

    A similar challenge can be raised against, say, the non-analytical naturalist about color, illness, etc.

    On the issue of water, Parfit holds that in the case of "water is H2O", the identification is not analytical, but 'water' can be analytically reduced to something else, and he lists a conjunction of properties, like filling rivers and lakes, quenching thirst, etc.
    But it's not clear to me that the analytical reduction holds. Consider, for example, the following statement:

    W1: Rivers and lakes are not filled with water. They're filled with a liquid that, if we were to drink it, would not quench our thirst and in fact would kill us. But if we go get some of the stuff, it won't kill us because the creator of the sim we live in (or the greatest conceivable being, or whatever) will turn the lethal stuff into water.

    As silly as such a claim would be, it's not non-transparently semantically problematic. Maybe one can remove rivers and lakes from the proposed analytical reduction, but it's not easy to find an analytical reduction. For example, even the condition that it quenches thirst is vulnerable: there are possible situations in which it does not (e.g., some sort of virus possibly causes people never to have their thirst quenched). There is also the question of "whose thirst", etc.

    Based on that, a potential objection from a non-analytic naturalist (I'm not saying it works, though; I can see potential trouble) might be to offer something like:

    MW1: The property of behavior that makes adult humans feel guilt when it's their own behavior, and makes adult humans feel like blaming someone when it's the behavior of someone else, etc., is natural property NW.
    MW2: The property of being morally wrong is that property of behavior that makes adult humans feel guilt when it's their own behavior, and makes adult humans feel like blaming someone when it's the behavior of someone else, etc.

    MW1 would be proposed as a non-analytical reduction, and - perhaps - a naturalist might that MW2 is not analytical, either, but it's means of identifying the property in an ostensive way (i.e., to point at it), just like W1 is not analytical, but it can be used to identify the substance. The challenge then would be to ask the non-naturalist to either explain the difference, or accept irreducible wateriness.

    On the informational equivalence issue, a similar reply would say that "Torturing babies is wrong." provides the same info as "Torturing babies has the property that makes adult humans feel guilt, etc", without claiming analyticity, but saying that informational equivalence is achieved by ostensibly pointing to some object. A reply would be that there is no informational equivalence without analyticity(e.g., what if a person does not know about adult humans, etc.?), and a counter-reply would be to mirror this with a watery argument, etc., and that what information a statement gives us seems to depend on who is "us".

    Those are just preliminary thoughts, though - very interesting argument!

    ReplyDelete
  4. I've been thinking about this argument, and I came up with the following objections:

    If there are two terms, say 'P1-ness' and 'P2-ness' that are used to talk about the respective properties, and the terms 'P1-ness' is not analytically reducible to any other terms, then it seems to me that in order to provide the information that P1 is the same as P2 to someone who does not know it already, one needs to either use the term 'P1-ness' or otherwise point at the property without an analytical equivalence. But I don't see why that would give us any metaphysically interesting result, like the existence of irreducible P1-ness.

    However, if the non-analytical naturalist actually is making claims incompatible with the above, then maybe she can concede a little (but not that much) and say something like:

    The property that we call "moral wrongness" is the same as the property of failing to maximize happiness (for example, and to simplify), and also other properties that we usually refer to by moral terms (like "moral badness", etc.), are the same as properties we can state by more or less complex statements without using moral terms.
    Now, I recognize that in order to provide the information that my thesis is true to someone who does not know that but grasps the meaning of moral terms, one needs to say things like 'the property that we call "moral wrongness" is the same', etc., just in order to let that person know which properties the thesis identifies with other properties.
    So, if the fact that one needs to do that is enough to establish that what is called "moral non-naturalism" is true, then so be it, so moral non-naturalism is true, but that is not metaphysically important: it remains the case that moral properties are so reducible. It's just that one needs to use some moral terms just as a means of pointing to the properties in question in order to explain the thesis to someone who does not know it's true but grasps moral terms, and the term 'moral non-naturalism' has been defined in a way that's broad enough for my thesis to fall under it.


    I think an alternative for the non-analytical naturalist might be to try offer a candidate fact stated entirely in non-normative terms (i.e., a natural fact), and question that informational equivalence requires analyticity.

    For example,

    (Q') Moral wrongness is lack of happiness maximization.

    (Q''') The property that is usually associated with feelings of guilt when one's behavior has it, and with punitive sentiments when someone's behavior towards one has it, and which we usually refer to by means of terms that do not mean the same as 'lack of happiness maximization' or any terms transparently analytically reducible to that*, is the same property as the property of lack of happiness maximization.

    Then, she may propose that (Q''') is informationally equivalent to (Q'), in a way similar to the way in which 'the stuff that fills the lakes and rivers, quenches thirst, and falls as rain, is H2O' is informationally equivalent to 'Water is H2O' (i.e., objections to the (Q''')-(Q') equivalence can be mirrored by similar objections to the water case)

    * The non-analytical naturalist holds that there is no non-transparent reduction, either, but she does not need to state that in (Q'''). She may add (after Q'''), something like 'Additionally, there is no non-transparent reduction, either'.

    ReplyDelete
    Replies
    1. (Q''') isn't informationally equivalent to (Q'), because it's quite possible to imagine (as actual) a world where "the property that is usually associated with feelings of guilt [etc etc]" is not the property of moral wrongness.

      Delete
  5. Richard,

    After further consideration, I have this reply to Parfit's Fact-Stating Argument (PFSA): either it's not successful, or a suitable generalization (say, Generalized Fact-Stating Argument, GFSA) defeats all non-analytical property reductions.
    Let's suppose that PFSA is successful, and Bob offers the following non-analytical reduction:

    (Q(1)): Property P(1) is the same as property N(1).

    The GFSA mirror's Parfit's (R), and says that (Q1) would be referentially equivalent to:

    (R(1)): Property N(1) is the same as property N(1).

    So, to defend their claim that (Q(1)) states an important truth, Bob must therefore claim that, in the relevant, informational sense (Q(1)) and (R(1)) state different facts. Now the argument challenges Bob to explain what fact (not stated in the language as P(1)) is informationally equivalent to P(1).
    If Bob provides an analytical reduction, we have:

    (A(1,2)): Property P(1) is the same as property P(2).

    That's analytical, according to Bob. So, Bob is committed to the following non-analytical reduction:

    (Q(2)): Property P(2) is the same as property N(1).

    But now the GFSA mirrors again Parfit's (R), and says that (Q1) would be referentially equivalent to:

    (R(1)): Property N(1) is the same as property N(1).

    Then, to defend the claim that (Q(2)) states an important truth, Bob must therefore claim that, in the relevant, informational sense (Q(2)) and (R(1)) state different facts. Now the argument challenges Bob to explain what fact (not stated in the language as P(2)) is informationally equivalent to P(2).
    After a finite number of steps (say, after reaching P(k)), Bob fails to provide a candidate analytical reduction, because he doesn't know infinitely many analytical reductions.

    Granted, Bob might provide circular reductions. But that's of no help, because we're assuming PFSA succeeds, and so circular reductions are excluded. Else, one may reduce "Torturing babies is wrong.", to "Torturing babies is immoral.", reduce the latter to "Torturing babies is morally impermissible", and then reduce that one to "Torturing babies is wrong".

    This reply doesn't depend on whether the reduction is a claim of "naturalism". Whatever "naturalism" means (I admit I don't know), the reply is based on what you identified as the key move - namely, that "Informative claims that referentially correspond to trivial facts must also -- informationally -- correspond to some other, less-trivial fact."

    In the case above, Bob ends up with an irreducible claim (or a claim he knows no reduction of), namely P(k), but yet P(k) is not an irreducible fact, since it reduces to N(1). That seems to violate Parfit's assumption. I'm not sure the assumption is innocent enough - if it is, though, it seems it defeats all non-analytical property reductions.

    ReplyDelete
    Replies
    1. An objection to my previous argument might be that even if Bob doesn't know how to analytically reduce P(k), he can hold that there is a reduction P(k+1), and then another one P(k+2), and so on. So, even though he's committed to infinite regress of analytical reductions, he avoids the GFSA.
      However, I think this objection fails, for the following reasons.
      1. A committment to infinite regress of analytical reductions is on its own a problem, and would need defending.
      2. It seems extremely implausible that the matter of whether a synthetic property reduction is correct depends on whether there are infinitely many analytical reductions. That would be very seriously weird.
      3. Assuming there is a way around 1. and 2., the non-analytic metaethical naturalist might parallel that reply and say there are infinitely many analytical reductions of "Torturing babies is wrong" - one just needs infinitely many analytical reductions of "Torturing babies".
      Now, those reductions would still use the term "wrong", and PFSA demands one that is stated in non-normative terms, so arguably that's why the move of the non-analytical metaethical naturalist fails.
      But then, for that matter, the infinitely many reductions {P(k): k = 1, 2, 3, ...} also such that none of them uses the word that means N(1), so it seems to me the GFSA - which generalizes PFSA - may similarly say that the reduction fails because none of them uses the term that means N(1).

      From a different angle, one might ask: if having infinitely many analytical reductions P(k) that don't use the word for N(1) or a synonym somehow enables the synthetic reduction (Q(1)) to go through, then why wouldn't having infinitely many analytical reductions of "torturing babies is wrong" (even if all of them use the term "wrong") would not allow a synthetic reduction of "torturing babies is wrong"?
      The weirdness seems as big in the moral case as in other ones.

      Delete
  6. Hi Angra, I agree that this kind of argument (based on the information-fact correspondence thesis) spells trouble for "pretty much any non-analytic reduction without an analytic reduction to connect them".

    I don't think this is an objection, since there are no clear counterexamples, i.e. of uncontroversial reductions that lack associated analytic connections. Take water/H2O. It's admittedly not clear precisely how to verbally explicate "the watery role". But linguistic competence suffices to pick out, for any fully specified possible world, what is playing the watery role. (See this quote from Frank Jackson.) So in that sense it is 'analytic' what functional role is associated with 'water'.

    But there is no such functional role, or associated description, for normative terms, as the open question argument demonstrates.

    ReplyDelete
    Replies
    1. Hi Richard,

      Personally, I'm not entirely sure linguistic competence is enough, but I grant that in this context.
      I realize I misspoke. I meant to say "Q(1)" (and "Q(2)", etc.) instead of "P(1)", "P(2)", etc., in "Now the argument challenges Bob to explain what fact (not stated in the language as P(1)) is informationally equivalent to P(1)" (it should have read "Now the argument challenges Bob to explain what fact (not stated in the language as P(1)) is informationally equivalent to Q(1)"), then the same for P(2) and Q(2), and so on.

      I think that error obscured my argument. I meant to construct it (i.e., if I hadn't misspoken) precisely with the kind of reply you suggest in mind.
      For example, we can apply the procedure to the "water is H2O" example you give. Let's say that, as your reply suggests, there is an analytical reduction from "water"(or "wateriness"), to "functional role (1)" (whatever that turns out to be).

      We then have the following proposed reductions:

      (Q(1)): Property "wateriness" is the same as property "being composed of H2O".

      (R(1)): Property "being composed of H2O" is the same as property "being composed of H2O".

      So, to defend their claim that (Q(1)) states an important truth, Bob (or whoever proposes it) must therefore claim that, in the relevant, informational sense (Q(1)) and (R(1)) state different facts, and the GFSA argument challenges Bob to explain what fact (not stated in the language as "water"/"wateriness" is informationally equivalent to Q(1).
      Going by the analytical reduction you propose, Bob may offer the analytical reduction from "water"(or "wateriness"), to "functional role (1)".
      But then we have the following non-analytical reduction:

      (Q(2)): Property "functional role (1)" is the same as property "being composed of H2O".

      We also have:

      (R(1)): Property "being composed of H2O" is the same as property "being composed of H2O".

      The Generalized Fact-Stating Argument now challenges Bob as follows: to defend their claim that (Q(2)) states an important truth, Bob must therefore claim that, in the relevant, informational sense (Q(2)) and (R(1)) state different facts. So, Bob is pressed to explain what fact (not stated in the language as "functional role (1)" is informationally equivalent to Q(2)).

      If Bob offers another reduction, then he's pressed again, and again, and again, until he has no further reduction to offer. Now, Bob might say "I don't know what the reduction is, but there must be one", but that's the infinite regress I tackle in the second post on my argument.

      Delete
    2. Ah, thanks for the clarification. The informational content of Q(1) is not given by a property identity statement. [Your Q(2), for example, is actually false: the functional role property is a higher-order property, distinct from the property of being water / H2O.] Rather, it is given by:

      Q2*: H2O is what actually fills the watery role. (i.e., the property of being H2O is the property which actually possesses the higher-order property of fulfilling such-and-such watery functional role.)

      No regress follows from this, because it is not an identity claim but an informative relation of two distinct properties.

      Delete
    3. I see. I thought the analytical reduction you proposed was from "X is water" to "X is the thing that fills oceans, or quenches thirst, etc." (which I tend to disagree, but I was granting that for the sake of the argument).

      Given your reply, I'm not entirely sure what the analytical reduction is, but it seems to me it's this: the statement "X is the property of being water" is analytically reduced to "X actually has the higher property of fulfilling such-and-such watery functional role".

      If there were no such analytical reduction, it seems to me it would conceptually possible that a property Z actually had the higher property of fulfilling such-and-such watery functional role, but Z would not be the property of being water. But if that were so, then Q2* would not have the same informational content as (Q(1)).

      But maybe I'm missing something. Is that the proposed analytic reduction?

      Delete
    4. Yep, that's the idea. (I don't think it's conceivable that something *actually* fulfill the watery role without thereby being water.)

      Delete
    5. Okay, thanks.
      Then, it seems to me that the GFST succeeds in defeating an identification between wateriness and H2O, for the following reason:

      Suppose Bob proposes the identity between wateriness and H2O, and then he's pressed to give the statement with the same informational content. Based on your previous reply, we get:

      Q1*: The property "wateriness" (i.e., the property of being water) is the same as the property of being H2O. (I removed the "composed of" to match your reply).

      R1*: The property of being H2O is the property of being H2O.

      Q2*: The property of being H2O is the property which actually possesses the higher-order property of fulfilling such-and-such watery functional role.

      If Q2* were offered as a property identity, then Bob would be asked to provide another informationally equivalent claim, and so on, leading to regress (and regress is a problem in this context for a number of reasons, I think; I argued for that in an earlier post). However, as you pointed out, Q2* is not an identity claim, but an informative relation of two distinct properties, so regress is blocked.

      Yet, the statement "X is the property of being water" is analytically reduced to "X actually has the higher property of fulfilling such-and-such watery functional role".
      Based on that, we have the following property identity:

      S2*: The property of being water is the property which actually possesses the higher-order property of fulfilling such-and-such watery functional role.

      But then, since the property of being H2O is not the same property as the property which actually possesses the higher-order property of fulfilling such-and-such watery functional role (because Q2* is an informative relation of two distinct properties), but the property of being water is the property which actually possesses the higher-order property of fulfilling such-and-such watery functional role, then the property of being H2O is not the property of being water.
      Of course, this does not defeat the scientific discovery about the composition of water, or the claim that in every possible world (imagined as counterfactual), water is composed of H2O, but it seems to me that the property identity is defeated.

      Am I missing something?

      Delete
  7. "since the property of being H2O is not the same property as the property which actually possesses the higher-order property of fulfilling such-and-such watery functional role (because Q2* is an informative relation of two distinct properties)"

    Wait, this is confused. Here are two distinct properties:
    (1) The property of being (composed of) H2O -- call this the H2O property -- and
    (2) The higher-order property of fulfilling such-and-such watery functional role -- call this the role property.

    Now, Q2* relates these two properties. It tells us that the H2O-property actually possesses the (distinct) role property. The H2O property is distinct from the role-property. It is NOT distinct from the property that possesses the role-property, for it IS the (one and only) property that possesses the role-property. But the role-property that it possesses is something distinct from the H2O-property itself. And by relating these two properties -- H2O and the role it actually has -- Q2* gets to be informative.

    ReplyDelete
    Replies
    1. I had misunderstood a previous post, but now I see what you were saying.

      I thought you were saying that Q2* didn't identify two properties, but I see Q2* does identify two properties, namely the property of being H2O and the property that actually fulfills the watery role.

      Delete

Visitors: check my comments policy first.
Non-Blogger users: If the comment form isn't working for you, email me your comment and I can post it on your behalf. (If your comment is too long, first try breaking it into two parts.)

Note: only a member of this blog may post a comment.