Epistemology

The following are some notes from a graduate seminar, mostly consisting of ~novel observations rather than recapitulations of the content (as I had on the old version of my site), although I’ll try to include some minimal exposition. This class was on distinctions in theoretical philosophy, run by Bernhard Salow in Hilary Term 2026.

Inquiry (Week 1)

Friedman (2024) argues that suspending judgement on Q is to be in an inquiring state of mind, which is necessary for inquiring (otherwise, you’re just collecting evidence: consider a cleaner who collects the same things as a detective) but not sufficient for inquiring (you might not collect any evidence). Left-to-right direction is fairly plausible, right-to-left is less clear.

Someone suggests grasping what an inquiring state of mind is supposed to be by comparing to an assertive state of mind (what actors lack, for instance), prompted some discussion about whether an inquiring state of mind carries any commitments. Someone suggests some flavour of possibiility, Bernhard suggests to the presuppositions: say, the question of how many people went to the party, while being unsure about whether there was a party. But suppose I do want to know how many people went to the party, but first need to find out whether it happened. Isn’t the most natural way to describe what I’m doing to the first question suspending judgement? Bernhard explains the point well: Friedman plausibly shows that you can’t or shouldn’t believe that the presupposition is false, but perhaps you don’t need to be all the way on the other side of the spectrum and believe that the presupposition is true.

Friedman notes that sometimes rediscovering something is easier than recalling it, but says that there’s still a basic epistemic conflict or incoherence here. But consider a concrete case: I have some mathematical formula that I can recall (so in particular, that I know), but it’s easier to rederive it every time I need it. This seems completely epistemically kosher—indeed, there would be something wrong with not just working it out again—which tells against the ignorance norm (only inquire if you’re ignorant) that Friedman posits. (It should be guise-relativised anyway, but this would affect a revised version too.) Bernard says it’s a good case, and that although it might seem weird to describe you as (say) wondering about what the formula is, maybe this is just an artefact of it being weird to wonder instead of just working it out, if it really is that quick to work out. But suppose we use a present-awareness (guise-relativised) standard for knowledge, such that facts remembered which aren’t in view don’t count. The norm is just: don’t use the harder route to settle the question. Then, you typically shouldn’t collect new evidence, because it’s easier to recall; but in cases where it’s easier to collect new evidence, you shouldn’t recall. Bernard also has a fantastic objection to Friedman’s otherwise-neat objection to the norm which says to only inquire if you’re aware that you know (namely, that this second-order problem is only plausibly generated by a first-order conflict: “If there’s nothing wrong with inquiring while knowing then why shouldn’t I inquire while knowing, in full awareness?”): this same argument seems to overgenerate, suggesting that Friedman’s ignorance norm is only plausible if there’s a tension between p being true and inquiring whether {p, …}—which would make it such that it’s always bad to inquire.

Kelp argues that the aim of inquiry is knowledge. Someone brings up the case of mathematics. But suppose we don’t guise-relativise: then it is right that you know. Suppose we do guise-relativise: then you’re unsafe with respect to the counterpart proposition that has the same guise. And, in particular, it’s nothing special about necessary propositions. (Consider propositions expressed in terms that would have a slightly different referent in any other world.)

Note 27 January 2026: Kelp gave today’s metaphysics & epistemology talk! I didn’t find it super compelling, though. It’s weird how little awareness there is of (i) the ur-prior approach on which your credences can go down from 1 (because you can lose evidence), although I was happy that Maria Lasonen pointed it out; and (ii) that probability 1 is not the same as guaranteed.

Double-Checking

In our class’s disgression on what checking is supposed to be, Bernhard suggests that checking might involve making belief sensitive in a way not required for knowledge. For instance, you can know that ice left in the sunlight two hours ago has melted, or that your dog doesn’t secretly speak German; although of course, if it wasn’t melted (say, if someone picked it up and put it in the freezer) or you dog did secretly speak German (after all, it would be secret), you would retain the same belief (or belief in something under the same guise, or whatever). But didn’t you check that the ice cube was melted? For it seems that one way of checking that p is by checking that p is locally necessary. (Bernhard’s example: maybe you didn’t know whether it was left in the sun or the shade, and you can check that it was in fact left in the shade, and so in all close worlds must have melted.) But if you can do that, you can do that in the imagination! In particular, you can imagine the ice melting within two hours in the sun, and thereby come to know that the ice would melt, and thereby come to know (in the good case) that the ice did melt. Someone objects that (i) checking that it’s likely that the ice melted doesn’t suffice, and (ii) in the case where this seems plausible you didn’t already believe and so didn’t double check. But what you checked wasn’t that the ice melted in a high proportion of worlds (that you had a low chance of error), what you checked was that the ice melted in every world in this region of modal space and so in the actual world (that you had no close chance of error). Someone else supplies the example of throwing a ball, closing your eyes, and then imagining where it will go based on the trajectory; someone else adds that double-checking by solving dynamical equations should count as double-checking, so using your capacities for rolling forward physical simulations in the imagination should count too. And indeed this example deals with the second objection, too: suppose you somewhat unreasonably believe that you’re about to sink this half-court shot; you shoot, then close your eyes, and confirm after a moment that you in fact would sink it. Bernhard quite liked the observation that you can (double-)check things by just thinking about them! Of course, if your wife asks you to double-check that you turned off the oven, and you just imagine leaving the kitchen and observe that you turn off the oven (without this being an actual memory)…