Some Metatheory

Blogging with teasers is fun. So here’s mine: how, in discussing epistemological metatheory, am I going to get you to think of Jason Stanley and steroids in the same thought? I will. I promise. (No, not just by getting you to read that question!)

It is a familiar theme in recent philosophy, and epistemology in particular, to distinguish two approaches to the subject. The first is conceptual analysis and the second is naturalized epistemology. The second requires that, in some important sense, epistemology be continuous with natural science; the first asks that we spend our time providing sets of necessary and sufficient conditions that give the content of important epistemological concepts such as knowledge and justification.

I want to explore my preferred alternative here, but first I want to ask where ordinary language philosophy fits into this picture. Think of the debates between contextualists, contrastivists, and invariantists. As far as I can see, it is a stretch to say they are engaging in the project of conceptual analysis. What they are doing is investigating various formal features of linguistic usage in order to generate an appropriate semantical treatment of epistemic language. This isn’t either conceptual analysis or naturalized philosophy.

There are important differences between old-style ordinary language philosophy and this more recent revival, of course. The recent revival is not anti-theory and anti-philosophy the way much of old-style ordinary language philosophy was. And the recent stuff is infused with a sophisticated use of logic and formal semantics. It’s like J.L. Austin on (logico-semantical) steroids. (OK, it only works if you know Jason and his work…)

It is true, however, that ordinary language philosophy leads naturally, even if not inevitably, to conceptual analysis, though it is a bit difficult to say how or why. Even so, no matter what approach one takes to theorizing, there have to be some connections to ordinary ways of talking and thinking, even if theoretical inquiry leads to reform movements of various sorts.

But my interest is not so much with where ordinary language fits into theorizing, but rather with an alternative to the divide between conceptual analysis and naturalized epistemology. An alternative to these approaches is value-driven epistemology. This sort of inquiry tries to determine what is valuable from a purely theoretical point of view, and develops accounts of the items so identified in ways that reveal and explain the perceived value. The intellectual heritage of this approach is Platonic, most expressly in the Meno, the Theaetetus, and the Sophist. In the first, Socrates considers the question of the value of knowledge over true belief, and this examination leads to the issue in the second of the difference between the two, and in the third, to an account of the method of collection and division. This method some have taken as an account of logos, that notion used in the previous dialogue as an account of the distinction between true belief and knowledge. In my view, it is better taken as an account of understanding, since it requires too much reflection to pass the “small children and animals” test for a theory of knowledge, but the important point here is not the results of the inquiry but its nature.

The picture of theory construction that emerges from these Platonic beginnings is an interplay between the valuable items identified and their nature as we learn of it in theory construction. The nature of the thing in question can show that it doesn’t have the value we first thought it did, and that something else instead should be the focus of our epistemology. Moreover, the results of our efforts will have quite significant practical implications. I expect, for example, that a properly developed account of cognitive achievements will tell us something about the adequacy of our educational system and the kinds of goals that are appropriate to it. It will also tell us something about attempts to develop “rules for the direction of the mind.” If there is value in true opinion, we should want rules that help us achieve that. And if the value of true opinion is less than the value of other cognitive achievements, no set of rules for the direction of the mind should aim (only) at true opinion. In addition, our intuitive sense of intellectual character traits can be honed by the results of value-driven epistemology. It is well-known, for example, that acceding to authority was thought a virtue in the middle ages and is thought a vice now; that originality and creativity are thought of as virtues now but not so much in earlier times. Even in philosophy, originality of thought trumps careful defenses and meticulous argumentation (OK, that’s a tendentious remark about recognition in philosophy). A good, value-driven epistemology would give us a vantage point from which to assess these features of contemporary culture.

OK, I’ll now step off the soapbox. Here’s the point. Nowhere in this description of value-driven epistemology do we appeal to conceptual analysis or the project of naturalizing epistemology (or to some type of ordinary language philosophy). It is equally true that nothing in this approach requires washing one’s hands of these other approaches, but the independence of the value-driven approach shows that these other approaches are not the only approaches available. And, of course, I think not the most promising either.


Some Metatheory — 18 Comments

  1. Jon, your description of the value approach makes it seem very desirable indeed. But isn’t the value approach just *good* philosophy? It seems that, neither conceptual analysis nor naturalized philosophy are any good if not in the service of something that can be shown to be worth thinking about. Philosopher often bring to bear their favored methodology on a question that seems not worth asking. I often find myself wondering, after reading a paper, why the person wrote it. Someone who has hold of a genuine philosophical problem and is making headway on it is just a person who has first identified something something of value and hence the investigation of which would be worth pursuing; by whatever means seems appropriate.

  2. Robert, yes it is just good philosophy. I guess the dispute would be something like this. Decide what’s worth having a theory about, and then the argument would be about how best to produce a theory (and maybe the argument will be about what it is to have a good theory of it as well). And who could object to using the best approach you can think of to finding a theory? ­čÖé

  3. Jon,

    This is a very stimulating post. I like to think about the meta-theory of formal epistemology, particularly the intersection of FE and AI. The most promising approaches in this area seem (to me) to be driven neither by methodological (nor substantive) naturalism, nor by conceptual analysis. Do you think that the following example would count as value driven? I’ll set it up by way of a methodological criticism of FE and AI.

    One complaint about formal epistemology and much of logic based AI is that these exercises are fatally compromised by a commitment to psychologism. Everybody recognizes logic as a theory about what follows from what but many people simultaneously think of logic as a part of a theory of reasoning. Which is, as Frege told us long ago and Gilbert Harman has recently taken to reminding us annually, a disaster. Grant this, as I think one should.

    However, one thing we use logic for is to represent arguments–which we do, namely, when what follows form what is at issue. We also, it seems, make and entertain inductive arguments. But in evaluating an inductive argument, what we are interested in is what follows from but-is-not-entailed-by what.

    The task of evaluating (or studying) inductive arguments is quite different from inductive reasoning. Inductive reasoning (or just `reasoning’) has something to do with belief fixation and change and the right and wrong ways of doing this. The tools for evaluating or studying inductive arguments, by contrast, should involve the study of ‘consequence’ relation(s) that are objective and mathematical in precisely the sense that the study of logical consequence is objective and mathematical. We, then, are not interested in psychology or states of belief but, rather, in relations and functions on sentences and sets of sentences. The pre-theoretic notions that guide our study of such structures are epistemic (or evidential) relations.

    Note the following methodological points. We aren’t trying to define the necessary and sufficient conditions of the concept of evidence or epistemic support. Rather, the method presumes a notion, which is described pre-theoretically and defined formally. (Perhaps we get such notions from straight analytic philosophy.) Psychology is irrelevant in the sense that there is no talk of `rational’ or `irrational’ inferences or behavior and no serious need of the results of psychology to carry out our business. (In whatever sense psychology is relevant it may explain why one approach is deemed more useful than another, but this is no more a concern for formal epistemology than psychological studies of the adoption of mathematical techniques in physics is for the conduct of physics.) We are simply looking at relations where, when you satisfy conditions on the LHS, such-n-such is (or usually is) satisfied on the RHS.

    Like applied mathematics in general, a system is motivated by some pre-theoretical idea about how inductive arguments work or how a feature of them works. These pre-theoretic ideas might not pan out mathematically, or might entail sharp constraints on how the framework is used. If so, that much gained in understanding. In so far as a framework fits our pre-theoretic notion, another ingredient in its evaluation is its intended use. One language might be very helpful for humans to represent and evaluate certain sorts of inductive arguments; the technical study of a language might provide theoretical understanding to a common sticking point in correctly evaluating arguments of a certain type (for example, Glenn Shafer’s notion of an ‘incompleteness mechanism’ is a key analytical insight to understanding how the naive application of conditional probability can generate puzzles like the 3-prisoners and Monty Hall).

    In these two cases, the aim is to increase our understanding of arguments of a certain type and of the mistakes that arise in representing a particular type of argument, respectively.

  4. Greg, I really like what you have to say here, especially the Harman remark! He just did it again at REC–which he should continue to do as long as people keep making the mistake…

    Understanding how arguments work fits quite well with the value-driven approach I like, and there is a sense in which it also provides something else beyond what I was posting about. We not only want to understand how arguments work, but also to develop mathematically precise theories that can have application to practical issues such as those involved in AI. I think of FE as sitting on the fence between the purely theoretical project and the applied issues involved in AI and elsewhere. What would be coolest of all is to produce the epistemological equivalent of a hand calculator!

    So it’s not surprising that it’s hard to think of FE in terms of either naturalized epistemology or conceptual analysis. The interesting and important addition that I hadn’t considered is the combination of pure and applied interests that can drive theory construction as well.

  5. Jon,
    (In haste). I’m not sure it makes sense to choose a methodology or a metatheory in advance. I thought the idea was to see which methodology gets one to some conclusions that are surprising, interesting, or important. If it does, then that methodology turned out to be fruitful. I don’t see the coherence, in epistemology or elsewhere, of ruling out some methodologies by fiat before inquiry. Of course, it makes sense to have an educated guess that certain methodologies won’t be very fruitful ones, or to take the results of applying a particular methodology and argue that they haven’t proven fruitful. But that’s different than saying ‘in advance, before inquiry, let’s say that only this sort of focus will end up with interesting results’. For example, I think before Williamson, people would have thought that only reductive analyses of the knowledge relation could yield interesting epistemology. I would never say that conceptual analysis (whether linguistically or logically informed or not) is the only way to do epistemology. There is no unique way to draw interesting conclusions, either in epistemology or in any other domain of philosophy.

    By the way, it’s useful to think of metaethics as another discipline in which one could have these debates. A lot of work in metaethics (the work of my former colleague Allan Gibbard comes to mind) is work on the metaphysics of content, of the sort that you might claim to be conceptual analysis (Gibbard thinks that the fact that normative content motivates can’t be captured by standard theories of content). That sort of methodology has proven very fruitful in that domain, even though value-driven approaches are of course also important.

    After I go see whether the new apartment I’m buying was part of the wall collapse onto the Henry Hudson parkway, I’ll post about why I think some of my work is in fact value-driven epistemology, of the classic kind…

  6. Harman and Kulkarni’s REC paper is very nice. Another gem is the title of Harman’s 2002 reminder, ‘A Logic is Not a Theory of Reasoning and a Theory of Reasoning is not a Logic’, appearing in a Gabbay handbook on practical reasoning. (I love that title!) One point of Harman’s that I don’t accept is his claim that the notion `inductive argument’ involves a category mistake, for reasons based on the distinction sketched in the above post.

    I also like your calculator analogy. This fits with an idea floating around in AI now about “epistemic tools”, understood as epistemically inspired algorithms or devices. This is one of the ideas collected under the umbrella of ‘human centered computing’, which I think was a term coined by Ken Ford’s group, IHMC, in Pensacola. (‘Android epistemology’ is another.) One idea motivating “epistemic inspired algorithms” is that some data-mining tasks involve semantic and evidential relations between the items in the search space—or, intriguingly, might be advantageously thought of in terms of such relations. In so far as we may isolate such structures there might be advantages in practice and in theoretical understanding.

    I had previously thought that this methodological approach–such as it is– to be a case *for* naturalized (epistemic) internalism, in so far as AI is one of the cognitive sciences. It bumps psychology out of the picture and also addresses the anti-science posture that opponents of naturalism are suspected of harboring but are careful to avoid in print. This literature–epistemic naturalism–is a bit confusing in that the main proponents and opponents often shift between talking about cognitive science and cognitive psychology, but I think people are mostly arguing about the merits of cognitive psychology and the merits of conceptual analysis w.r.t. theories of epistemic justification. So, upshot: you can now be an internalist and also pass the science sniff-test. (Note that the externalist conceptions of justification are not ruled out, either: you can build models whereby you’ve access to `both sides of the fence’ and are primarily concerned with correct classification or feature extraction. Again, no psychology; relations and functions on sets or, perhaps, an artificial neural network situated between precisely defined input and output equations; In the latter case, experimental data from running such programs on test data would also be included.)

  7. Jason, I think what you say shows that you endorse the value-driven approach (note your language of interesting and important results). Not surprising, since I think most everyone thinks this way (see Robert’s earlier claim to this effect), but they don’t consciously affirm the view and hence do not practice it as consistently as they should. And by not being a reflective adherent of the view, one can think of oneself as doing something else and losing track by following a different model instead. That’s actually what I’m objecting to–not following the value-driven approach all the way. And of course I have an agenda: I think you end up with quite a different picture of epistemology when you self-consciously follow the value-driven approach.

  8. Jason, I myself don’t see Gibbard as doing conceptual analysis, though his topic is ‘concepts’–normative concepts. He proceeds very much in the way that Jon describes as a value driven approach. Indeed, I think this is one thing that is so striking about ‘Wise Choices’…for lack of a better word, its humanity. What guides the argument at every stages seems to be an eye to what it is important to understand. There are forays into analysis (his solution to Geach) as well as forays into naturalistic accounts of our practices (e.g., his socio-bio account of norm acceptance). But these seem to me to be in service of illuminating something already located and defended as valuable and important ahead of time, our actual normative discourse and practices.

  9. Jason, just to be clear in case anyone wasn’t clear about this. The steroid comment was not a criticism; this is one kind of steroid everyone would do well to use!

  10. Chris, you’re right, and I now remember reading this thread! Oops… Apologies to Fritz for borrowing the metaphor and not crediting him (though it seems to me his career won’t suffer from my failure!).

  11. Let me explain a bit how I see appeals to formal semantics entering into my recent work (I’m speaking of my forthcoming book, and not my prior work on knowing how with Williamson). It struck me quite early on in this project that if we take the Cohen and DeRose ordinary appeals to intuitions at face-value, what one gets is not contextualism, but rather the dramatic thesis that whether or not someone knows that p at a time depends in part upon their practical situation. Then it struck me that this isn’t surprising, if knowledge is conceptually connected to action in certain ways (Fantl and McGrath investigate one way, Hawthorne and I another). I was at Michigan at the time, and learning about decision theory from my colleagues. For them, knowledge wasn’t connected to action, only appropriately confident belief was. But the ordinary cases discovered by Cohen and DeRose suggested that we often assert or deny that someone knows depending upon whether certain actions are acceptable for them to undertake. It seemed to me then (and still seems to me now) that some of the value of knowledge is due to its constitutive connections to action. Replacing knowledge by (say) appropriately confident belief in explanations involves severing these connections.

    It then struck me that it is odd to try to explain away these intuitions by appeal to semantic claims. However, once I realized some of the consequences of taking the intuitions at face-value, Cohen and DeRose’s route was more explicable. On the other hand, I couldn’t convince myself that Cohen and DeRose were right about the semantics; indeed, I convinced myself that they were wrong. Similarly, other semantic maneuvers to get away from these consequences seemed unpromising.

    So, my appeals to natural language semantics (at least in my forthcoming book) are entirely negative. They are intended to show that there is no way out, save by rejecting that we have the intuitions to which Cohen and DeRose drew our attention, from certain drastic conclusions, namely that what makes true belief into knowledge isn’t (to put it tendentiously) a matter just of reliability or supporting beliefs.

    The conclusion sounds bad. But it’s the conclusion we must accept, if we want to preserve the centrality of knowledge to the explanation of rational action. That is my goal in the book. It is also, I think, quite clearly the goal of Fantl and McGrath in their paper. The fact that the resulting view helps with a great number of complex perhaps non-traditional skeptical problems in epistemology (as has been shown by Hawthorne) provides considerably more evidence for it.

    This is no doubt an application of Williamsonian “knowledge-first-ism”. But I don’t see that it is an application of ordinary language philosophy, however conceived.

  12. Jason, this is a very useful summary, one which makes clear the value-driven character of your inquiry. It’s the same thing that attracts me so much to Williamson’s book. What I don’t see, though, and would like to see, is a carrying out of the value-driven approach when anomalies arise. So, what I notice in what you say is the tension between two parts of the summary. First you say,

    It seemed to me then (and still seems to me now) that some of the value of knowledge is due to its constitutive connections to action. Replacing knowledge by (say) appropriately confident belief in explanations involves severing these connections.

    And then at the end, you say,

    The conclusion sounds bad. But it├ó´┐Ż´┐Żs the conclusion we must accept, if we want to preserve the centrality of knowledge to the explanation of rational action.

    From my own value-driven perspective, the resolution of the anomaly should be approached by explicitly considering the interplay between the value and nature of items used in the explanatory scheme. I think the reason you worry that the conclusion sounds bad is because it is hard to defend the high value we place on knowledge if it comes and goes depending on our cares and concerns. From that perspective, we begin the road of theory construction from an assumption about the high value of knowledge and end up having to minimize its value. If that’s right, then we should look for other cognitive achievements and states that are tied to rational action and for which we can find theories that don’t imply that these achievements and states disappear when we need them most.

    That’s not an objection to your book, as I’m sure you realize. It’s an argument for the following conditional: if you’re right, then the domain of epistemology should shift from its historic preoccupation with knowledge. I really like the consequent here!

  13. Jon, if you’re right that epistemology, should be driven by these evaluative questions, it seems to me that it also could take a look on the strong side of knowledge, to certainty, as well as on the weak side of belief. Many of the examples I’ve read in the contextualist literature, for instance, seem to me to call for questions such as ‘Are you certain?’ rather than ‘Do you know?’ (despite the author’s claims) and that seems to me to signal that certainty may be more valuable, in the sense you’re discussing, than some may think. (Here, I’m thinking of the bank case, for instance.) Contemporary philosophy seems to have spent so much energy playing down certainty in order to secure claims to know that it may well have overlooked something just as or even more valuable.

  14. Dear Jon,

    The conclusion of my book is disjunctive. The first disjunct is the one you take — if knowledge is what I say it is, then epistemology should shift from its historic preoccupation with knowledge, and move to justification. This first disjunct is the disjunct to adopt, if the interest-relativity of knowledge does not filter down to the other ordinary epistemic notions (something I don’t argue explicitly for in the book). The second disjunct is that all epistemic notions are similarly interest-relative. This is the disjunct I’m most sympathetic to, since I am so taken with much of Knowledge-firstism (though not all the way to the identification of knowledge with evidence). To quote myself from the conclusion (this is really obnoxious, but it’s easier than summarizing):

    If evidence and related epistemic notions are similarly interest-relative, then there may be no purely epistemic notions, in the sense of a notion that is ├ó´┐Ż´┐Żstripped clean├ó´┐Ż´┐Ż of its ties to the practical interests of epistemic agents. In the face of this situation, one might appeal to a Carnapian ├ó´┐Ż´┐Żrational reconstruction├ó´┐Ż´┐Ż of a purely epistemic notion of justification or of knowledge. Attempts to do so will be hindered by the concern that any such purified notion will not play the familiar roles we ask of our ordinary epistemic concepts. It may turn out that part of the value of these concepts comes in the links they have to our practical interests. Finally, if all of our epistemic notions share the connections to practical interests that knowledge possesses, it will be very difficult to tell when we have successfully produced a notion purified of such links.

    As I said, I leave the conclusion as a disjunctive one, though I am far more sympathetic to the more dire (from a traditional epistemological perspective) of the two disjuncts.

  15. Jason, I’ve been working on a post giving a metatheoretical argument why epistemic justification has to be purified in this way, so I’ll be interested in what you think about it. I’ll probably put it up early next week.

    I think, though, that given your predilections, you’ve got a lot of motivation to think of the subject matter of epistemology as going beyond knowledge and justification. I think epistemology is about a certain class of cognitive achievements, ones that are important or valuable from a purely theoretical point of view (one that abstracts from other interests and concerns, such as practical, moral, political, religious, etc.). If the dire conclusion you point to is one we’re stuck with, then we ought to think about other kinds of cognitive achievements. The one I’m most interested in is understanding, but there are others as well that appear to be quite important even when we don’t assume any connection to knowledge (I’m thinking of Foley’s target of egocentric rationality, what responsible inquiry is like, etc.). I love dire conclusions as much as anyone (remember I think that knowledge is not more valuable than a proper subset of its parts!), and I’m intrigued by the consequences of your investigation–I think they help confirm that epistemologists should broaden their horizons to think about cognitive achievements other than knowledge.

    By the way, I think there’s a key here to understanding Williamson’s opposition to both contextualism and invariantism. I think he perceives both as supporting dire conclusions about the value of knowledge, and value that underlies his “knowledge-first” approach. He’s never said anything quite like this that I know of, but I bet he’d be sympathetic. Do you think so?

  16. Robert, I think the standard assumption is that knowledge does require certainty, just not the Cartesian sort. Peter Klein’s book is the standard here, and Peter thinks of it as something like axiomatic that you have to be certain in order to know. And I think people usually interpret this to mean both psychological and epistemic certainty. Good Mooreans would insist on the point.

    Some of us resist, of course! But even those who do will want decision theory to be sensitive to something gradeable on the cognitive side. Degrees of belief is one such idea, and maybe strength of belief is another. And each of these will have analogous dimensions when we assess the quality of belief from an epistemic perspective. And I’m all for that–knowledge gets more airtime than it deserves!

  17. I agree completely with your interpretation of Williamson; it’s clear that this is why he strongly rejects both contextualism and interest-relativity (and I suspect he would be more inclined to accept a version of contextualism where the denoted knowledge relations are not interest-relative, than interest-relative invariantism). On the other hand, he’s attracted to several theses that make him especially vulnerable to the temptations of either contextualism or an interest-relative account. First, I think that the intuitions that lead one to contextualism or an interest-relative account are of the same sort as would lead one to the knowledge account of assertion. Secondly and relatedly, he’s attracted to principles linking knowledge and action that lead (via various routes, e.g. the one described by Fantl and McGrath) to interest-relativity. Third, he’s inclined not to be a skeptic about knowledge of the future, to deny that we know in lottery type situations, and to want to hold on to some version of multi-premise epistemic closure. I’m inclined to think that Hawthorne has convincingly shown us that the best way to hold on to these commitments is by adopting an interest-relative account.

Leave a Reply

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