Primary and Secondary Norms

I was reading Paul Weirich's _Realistic Decision Theory_ recently which is an attempt to progressively remove idealizations from standard decision theory to make it more applicable to non-ideal agents. He says in several places--beginning with the first page of the Preface--that this project is already implicit in the focus on maximizing *expected* utility as opposed to *fully informed* utility--what is most utile given what information I have, rather than what is most utile give all relevant information possible. Later in the book he makes the following distinction to highlighting the above difference.

I distinguish success goals and cognitive goals for decisions. The goal to maximize fully informed utility, the "factually objective" counterpart of subjective utility, is a success goal. The goal to maximize utility is cognitive. cognitive goals are subordinate to success goals. They express secondary goals aimed at promoting primary, success goals when obstacles arise. (148).

Williamson also uses the notion of primary and secondary norms to defend the knowledge account of assertion (see, e.g., Williamson 200, 257). Additionally it is often said that the subjective norm "Believe what the evidence indicates!" is subordinate to the objective norm "Believe what is true!" I have my doubts about this whole way of thinking.

If I had to chose between

A. being a member of a demon world where, though I get it wrong all the time, I have excellent responsibilist epistemic virtues or have J-ADROIT doxastic practices or whatever, and
B. being a member of a world where I'm just a pervasively lucky guesser and get it right all the time in spite of my doxastic practices.
I would clearly prefer A. Likewise, when it comes to praise and blame, I see someone exemplifying A as 100% praiseworthy (0% blameworthy) and someone exemplifying B as 0% praiseworthy. The same goes for the corresponding cases in assertion or decision theory.

This is not to deny the value of truth or some norm to seek it. Rather, it suggests to me that the so-called "secondary norms" are really the primary dimension of evaluation and that their value is not solely derived from their service to so-called "primary" goals. (Indeed there are skeptical worries they even do.) That is, the "subjective" goals have independent force in their own right and their satisfaction has independent value.

Foley-Rationality Saved from Shope?

I think Foley's original definition of egocentric rationality probably commits the conditional fallacy. Without committing to that or saying how bad a thing I think that is I want to take the moves Kaplan 1996 makes to save his definition of full belief from the conditional fallacy (plus another two moves) and applied them to Foley.
(FR') S's belief B that p is rational iff were S to be sufficiently reflective concerning p, S would prefer to assert p.

Off hand it seems to circumvent the main form of the conditional fallacy, the form it's usually accused of committing. I'm worried about the following putative counter example a bit though. Go to a world in which S is rational but if S were to reflect sufficiently concerning p and evil demon would cause S to prefer not to assert p.

I have doubts about this c-e because I'm not sure that someone can really cause you--from the outside--to have preferences in the relevant sense. Suppose you're worried about retaining water and so prefer not to drink more than a few cups a day, but then I slip a bunch of salt in your meal and you have a burning desire to drink a gallon of water.

My intuition is that though I've given you a kind of desire it is not your preference in the sense in which (non operationist) decision theorists use that term. Perhaps I need to introduce a technical term "settled preference" defined in terms of higher-order preferences since the desire for water in this case is temporary and you would prefer not to have it.
Some work of Frankfurt and Fischer would probably help elucidate the right notion of preference for my purposes.

The Epistemic Use of Simplicity in Theory Construction

I suggest that

(S1) When multiple theories cover the data, the simplest is the most reasonable to believe

is a priori in the sense that it is constitutive of our concept of reasonable belief. I think this is especially clear when it comes to a deontological conception of reasonable belief for the following seems clearly true to me.

(S2) We don't have a right to posit any more (kinds or tokens of) entities than are required to cover the data.

From this it follows that

(S3) To posit more (kinds or tokens of) entities than are required to cover the data is unjustified.

I think these are basic facts about rationality.

This suggests the following account of theory simplicity.

(S4) Theory H1 is simpler than theory H2 iff H1 postulates fewer (kinds and/or tokens of) entities than H2.

I think all these principles have the positive feature of being platitudinous. I think at one time they would have been accepted as platitudes by philosophers and surely are accepted as platitudes by the innocent folk (in this case a plus in my view).

The question of whether simple theories are more objectively likely to be true is, I think, of the same status of the question of the justification of induction and therefore a very very deep and difficult problem. I think the most interesting question in the present approach concerns (S4).

(SQ1) How many new tokens are worth a new type?

(S4) is quite indeterminate w.r.t. answers to (SQ1). The main options are

(A1) It is always better to postulate more tokens than a new type.
(A2) At some (vague) point, it becomes better to postulate a new type than new tokens.

Here's the sort of conflict I envision: H1 covers the data by postulating many billions of particles of a single type, whereas H2 covers the data by postulating, say, *two* token particles of two different types. (A1) strikes me as right at first, but this consideration makes me pause to reconsider (A2).