De Se and De Dicto

Several philosophers I spoke with at the Pacific meeting are working on the issue of the confirmational relevance of de se information. I’m working on it as part of a defense of what I’ve called propositionalism (which turns out now to be ill-named), Dave Chalmers has a new piece on it (I think he’s using it for the online philosophy conference), and Mike Titelbaum, one of Branden Fitelson’s students at Berkeley, is writing a dissertation on how to model the relevance of such information. (Dave’s paper is here and Mike’s is here.)

To begin addressing the issue, one needs a clean characterization of the difference between de se and de dicto information. David Lewis says that all beliefs are de se and the de dicto ones limit the class of worlds to which their subjects belong, whereas irreducibly de se beliefs indicate a subject’s identity or location in space or time. The distinction is thus a distinction between beliefs about logical space and those about “location” within a world. I find this characterization to be unhelpful, but perhaps I don’t understand it. Suppose I believe that tomorrow I will win the lottery. There are worlds in which I will win the lottery tomorrow and worlds where I don’t; so this belief appears to limit the class of worlds to which their subjects belong, and is thus de dicto. But it clearly is a de se belief, as opposed to the belief that JK will win the lottery on April 14, 2006 (the indexical character of calendar claims ignored for present purposes).

Perhaps the problem is that of trying to extend the concept of de se attitudes to encompass more than beliefs involving self-awareness.

The original use of the language of de se attitudes was used to talk about self-awareness, whereas the confirmational problem is one that arises for any informational content expressed using indexicals or demonstratives, or put more vaguely, any content expressed by sentences involving contextuality.

Thought of in this way, the problem of de se information is just a special case of a more general problem, the problem often labelled since Frege as the problem of cognitive significance. If we think of propositional content in Russellian terms, we will think of propositions as having individuals, properties, and relations as elements. Since Frege, however, it has been relatively well-known that confirmation can’t depend solely on relations between Russellian propositions for the simple reason that cognitive significance can’t be explained solely by appeal to Russellian propositions. So the problem of de se information is best thought of in terms of the more general phenomenon of trying to account for cognitive significance. In this way, de dicto information is just propositional information, and if one assumes the Russellian standpoint, propositional information can’t account for epistemic support relations. What is needed for that is content-plus-sense, or content-plus-mode-of-presentation. If you’re a Fodor fan, you might say: content-plus-sentence-in-the-language-of-thought (requiring of course that LOT has indexicals, demonstratives, contextual terms, and all the other things that allow different sentences in English to express one and the same Russellian proposition).

One might try to develop a non-Russellian account of propositional content, holding out the idea that cognitive significance and confirmational considerations are the best guides we have to the notion of intentional content. There are well-known problems here, however, so I’ll drop that suggestion for now, at least.

What’s troubling about the endpoint here, however, is that it is not a neutral characterization of the problem in question, since it assumes that Russellian propositions are the (alethically relevant) contents of intentional states.


Comments

De Se and De Dicto — 10 Comments

  1. Jon,

    I think the problem of cognitive significance may be a bigger one than we need to solve to understand confirmational relevance.

    Take your lottery example. Suppose that throughout the example you are certain that you are JK and that today is April 13, 2006. If you are rational, learning that you will win the lottery tomorrow has the same effects on your doxastic state as learning that JK will win the lottery on April 14, 2006. Thus strictly from a point of view of confirmational relevance, the belief that you will win the lottery tomorrow and the belief that JK will win it on April 14 seem to have the same significance under these conditions.

    Would we want to say that these two beliefs always have different cognitive significance, even when you are certain throughout that you are JK and that today is April 13? If so, then the concept of cognitive significance in play seems motivated by more than just confirmational considerations.

    Early in your post, you write that “To begin addressing the issue [of the confirmational relevance of de se information], one needs a clean characterization of the difference between de se and de dicto information.” Later, you suggest that “the problem of de se information is just a special case of a more general problem… the problem of cognitive significance.”

    I believe there are correct, general principles for modeling rational agents’ responses to learning new information that make no reference to a de se/de dicto distinction or to a notion of cognitive significance. Drawing a de se/de dicto distinction is useful to help introduce those principles and mine intuitions that support them. But in the end, the confirmational relevance of de se information is just an application of the principles to a particular domain.

    I agree with you that Lewis’s conception of de dicto information and the general notion of cognitive significance are philosophically difficult. But as for your entry point to the topic, I’m not convinced that we need to understand cognitive significance before we can tackle confirmational relevance.

  2. Mike, maybe that’s right, your point about confirmational relevance, but if so, you’ll have to divorce confirmational relevance from justificatory relevance. If I’m amnesiac, learning that JK has a certain feature won’t justify the same beliefs that learning that I have that feature will justify. Same for claims about tomorrow versus claims about a certain date, if I’m lost in time.

  3. If you’re an amnesiac, learning that you will win the lottery tomorrow will have a different effect on your doxastic state than learning that JK will win the lottery on the 14th. So if you’re an amnesiac, the two beliefs have differing confirmational relevance, to go along with their differing justificatory relevance and their differing cognitive significance.

    I focused on the case where you’re certain who you are and what day it is because that’s the case where confirmational relevance and cognitive significance might come apart. In that case, learning either of those beliefs has the same effect on your doxastic state (if you’re rational), so the two beliefs have the same confirmational relevance. However, they might still differ in their cognitive significance. Notice that in this case as well, justificatory relevance and confirmational relevance go the same way. If you are certain who you are and what day it is, learning that you will win the lottery tomorrow will justify the same beliefs as learning that JK will win on the 14th. So I don’t think confirmational relevance and justificatory relevance have to come apart.

  4. Mike, I think that the notion of differing in cognitive significance is a modal notion, one concerning the possibility of two contents playing differing roles, epistemic and otherwise, in different contexts. If I know I’m JK, I’m not sure how, in that context, the roles of beliefs about me and beliefs about JK might come apart. Except, of course, for not taking into account that I’m JK for the moment, and things like that, which are just special instances of the modal property displayed more generally in cases of amnesia.

    I sense we’re not on the same page, though, about the notions in question here, so maybe I’ll try to say, very fast, what I’m assuming. Contents C and C’ have the same justificatory relevance iff they epistemically support believing the same things in all contexts. So beliefs about me and beliefs about JK do not have the same justificatory relevance. I was also assuming that confirmational relevance tracks justificatory relevance, so they wouldn’t have the same confirmational relevance either. Finally, cognitive significance is, I take it, a matter partially constituted by functional role. If it’s more than that, I’m forgetting for the moment!

  5. Going back to your comments about Lewis’ analysis, when you believe what you would express by saying “I will win the lottery tomorrow”, you locate yourself within a certain population of possible individuals. You locate yourself among those possible individuals that are lottery winners (I, too, am ignoring the tensed nature of your belief). In other words (since for Lewis, properties are just sets of possible individuals), you self-ascribe the property of being a lottery winner. If you believe what you would express by saying “JK will win the lottery tomorrow”, you locate yourself within a population of individuals who inhabit a world in which JK wins the lottery. In other words, you self-ascribe the property of inhabiting a world in which JK wins the lottery. So the properties that are self-ascribed differ in the standard de se case vs. the standard de dicto case. In the former case you locate yourself among a population of lotto winners and in the latter case you locate yourself among a population of inhabitants of worlds in which JK wins the lotto.

    If you’re an amnesiac, you might self-ascribe one of the properties without self-ascribing the other. Since they’re different properties that you believe, they will play different roles in the justification of other beliefs.

    So when you believe what you would express by saying “I will win the lottery tomorrow” you are doing more than locating yourself in a particular world, you are locating yourself among a sub-population within a world: namely those lucky lotto winners.

  6. Jon, I think I now understand the difference between the notions you and I were using. I was discussing the (non-modal) question of whether two beliefs have the same confirmational relevance in a given context, while you were discussing the (modal) question of whether for two given beliefs there can be any context in which they differ in confirmational relevance. (The same goes for justificatory relevance.) Of course, before we can determine whether there are any contexts in which two given beliefs differ in their confirmational relevance, we first need a way to determine whether two beliefs differ in their confirmational relevance in a given context. And for that project, I don’t think we need an antecedent notion of cognitive significance. But more broadly, I wonder whether it’s the case that (there exists a context in which two beliefs have different confirmational relevance) just in case (those two beliefs have different cognitive significance). [Parentheses added for clarity.]

  7. Stephen, thanks for the gloss on Lewis’s perspective on attitudes. The problem I have here is that Lewis’s view is simply not acceptable, and the language he uses can’t be divorced (as far as I can tell) from the details of the unacceptable theory. In particular, the idea that one is somehow self-ascribing anything when characterized by a de dicto attitude is mistaken, especially when the attitude is simply a dispositional one. So what I’m looking for is a theory-neutral characterization of the de se/de dicto distinction, rather than a theory-laden one of the sort Lewis provides.

  8. Mike, I’ve been thinking about your last sentence concerning the link between confirmational relevance and cognitive significance. I take it we are quantifying here over all possible individuals, wondering whether for any possible individual, is it the case that p and q differ in cog. sig. for S iff p and q differ in conf. rel. for S (for short, ∀S (CSpqs iff CRpqs)).

    Whatever we say, we better get that A=A and A=B differ in cog. sig., and if that case provides a decent model for how such differences will work, I think the left-to-right direction is safe (i.e., CSpqs → CRpqs). I know this is a crazy hasty generalization, but I don’t see how differences in cognitive significance won’t end up making some confirmational difference. The more interesting direction is the other one: CRpqs → CSpqs. One’s answer here strikes me as revealing how subjective one is about confirmation. Since I like strongly subjective accounts of justification, and I think confirmation had better be a closely related notion, I’m inclined to like this direction as well. But if one resists such subjectivity, it’s easy to see why one would want to reject the claim. Does that seem right to you?

  9. Jon, I’d like to hear more about what you mean by being “subjective” about confirmation. The modal notion of confirmational relevance we’ve been discussing depends on whether there exist contexts in which particular beliefs are confirmationally relevant to particular others. That, in turn, seems to be understandable in terms of whether the doxastic states of rational agents in particular contexts would change upon those agents’ gaining particular beliefs. So how could that be anything but a “subjective” question? I guess what I really want to know is how one could *resist* subjectivity about confirmational relevance.

    As for the left-to-right direction, there might be problems with tautologies. There certainly seem to be tautologies that vary in cognitive significance, but in at least some cases it’s unclear whether they should enter into different confirmation relations. And then there are beliefs like “I am here now,” where it’s unclear whether such a belief ever confirms anything because it’s unclear whether such a belief is ever really *learned*.

  10. Mike, here’s what I was worried about. Different functional roles imply differences in cognitive significance. So, suppose you have a theory of confirmation that allows for the possibility of people being systematically mistaken about what confirms what. Then you get the following possibility: p and q might have different functional roles (since S systematically infers different things from each) and hence differ in cognitive significance, and yet p and q have the same confirmational relevance. Subjective theories will, at the very least, minimize such possibilities and perhaps eliminate them entirely.

Leave a Reply

Your email address will not be published. Required fields are marked *