A Critical Look at Mark Kaplan’s Decision Theory as Philosophy

John Shoemaker

In Decision Theory as Philosophy, Mark Kaplan reissues a number of perennial questions within decision theory and epistemology, particularly regarding the relevance of decision theory to epistemology and the scope of an epistemology informed by a “modest” Bayesian decision theory.  Much of Kaplan’s book represents a challenge to what he calls the “Orthodox” Bayesian theory of decision and evidence.  His arguments turn positive in the fourth chapter, in which he argues for the “Assertion View” of belief – an attempted reconciliation of the categorical notion of belief (as distinct from disbelief) with that of confidence, which comes in degrees.  The approach to epistemology manifest in Decision Theory, while commendable in some respects, suffers fundamentally from its methodological commitment to the primacy of preference principles over and above distinctively epistemic principles.  But to express this last misgiving is just to doubt whether decision theory has much of its own to contribute to epistemology. 

I have divided this critique of Kaplan’s version of subjective Bayesian epistemology into three sections.  In the first, I defend what Kaplan avows as his “a priorist” approach to describing rational constraints on belief against epistemological naturalism.  In the second section, I uncover some problems (or perhaps, ambivalence) in Kaplan’s discussion of diachronic constraints on con-assignments.  Indeed, we will see that, despite Kaplan’s rejection of them, diachronic constraints are deducible from his decision theory, leaving the reader to wonder all the more at Kaplan’s thoughts on “Better Evidence” (pp. 75-85).  Finally, in the third section, I attack the Assertion View of belief.  Towards this end, I show that the Assertion view does not have the advantage over all other accounts of belief – vis à vis the paradoxes of the lottery and preface – as Kaplan thinks it does.  Following this, I show that the Assertion View cannot mitigate its strangeness by serving as a guard against the skeptical anti-realist argument of Bas van Fraassen.

A Natural Enemy

Kaplan’s approach to decision theory deserves the name “axiomatic,” as he eschews Dutch book vindications of conforming one’s degrees-of-belief [1] to the probability calculus in favor of deriving elementary probability principles from his axioms of preference.  It would be of some philosophical merit, no doubt, if it could be demonstrated that, in virtue of accepting only a few self-evident axioms constraining preferences between states of affairs (including bets), any rational person’s confidence in hypotheses must also satisfy the Kolmogorov axioms of probability. [2]   But Kaplan’s preference axioms (which he calls “Ordering,” “Dominance,” “Confidence,” “Decomposition,” and “Modest Connectedness,” alongside auxiliary assumptions about one’s values which these axioms invoke) are not adequate to that task and require refinements, which are possible only with some rather creative maneuvers. [3]   These maneuvers, in service of what Kaplan describes as an “a priorist” epistemology, are especially apt to arouse suspicion among those self-styled naturalists who are disposed to doubt the value of such “first philosophical” approaches to epistemology. [4]

It is with this antagonism in mind that Kaplan turns to defend his a priori approach to matters epistemological against criticisms typically offered by naturalists.  Three related objections to a priorism are considered and rebutted; however, Kaplan makes undue concessions to the naturalist position.  In what follows, I attempt to defend the legitimacy of Kaplan’s axiomatic, generally a priorist, approach to epistemology against the naturalist objections he reproduces from Kitcher (1992), offering what I hope are more complete, less tentative, responses to these objections.

Let us review each of these objections in order to articulate a stronger defense of a priorism.

“First, a priorism must necessarily proceed in ignorance of the actual cognitive capacities of the human agents whose cognitive activities it seeks to govern.  This renders it singularly ill-suited to determine what cognitive processes are apt to be reliable in actual human agents” (pp. 183-4).

This question-begging first objection is cogent only if one is already disposed to reliabilism – roughly, the view that a belief is justified in virtue of having been produced in an appropriate way by a cognitive mechanism that reliably produces true beliefs.  For those who are not given to reliabilist views (and there are good reasons to doubt its coherence), [5] including Kaplan qua axiomatizer of preference principles from which constraints on rational belief are derived, rationality does not consist in being the proud owner of properly functioning reliable belief-producing mechanisms.  Instead, something of a Cartesian spirit animates Kaplan’s project – where “reason’s requirements” on belief are derived from principles, which cannot be at once seriously entertained and doubted. [6]   Anyone convinced that this a priorist project is fundamentally correct should therefore be indifferent to this first objection.

“Second, having no ingress to what is apt to work in the actual world, a priorism has no choice but to restrict itself to prescribing cognitive strategies that are optimal no matter how the world may be – a quixotic endeavor that leaves aside and (because it sets so high a standard of success) may even frustrate our attempts to find what we really want: strategies that will serve us well in the world we happen actually to inhabit” (p. 184).

This objection reveals a basic misunderstanding of a priorist approaches to justification.  Whereas the naturalist typically begins his investigation with some empirical claims regarding the reliability of certain belief-sources (which secure his “epistemic goals”), the a priori epistemologist need not, and generally does not, begin his enquiry into justification with the goal of discovering epistemic practices that will reliably produce the truth no matter how the world may be.  Rather, he will look for incorrigible principles which guarantee the reasonableness of beliefs; only in narrowly defined cases will these principles make any guarantee that the beliefs they justify are true (or true with some definite frequency). [7]   In Kaplan’s case, these principles are the preference axioms (and perhaps he intends Evidence and Rational Belief to be basic principles of rationality); the principles he offers are not indubitable, but this is an objection to Kaplan – not to a priorism.  The quest for a priori principles that support the rationality of certain inferences and beliefs is decidedly not quixotic.

“Third, a priorism assumes that the fact that a methodology is in accord with our intuitions about what is rational shows that it is worthy of our allegiance…  [On the contrary], that a methodology accords with our concepts of rationality is entirely compatible with its being quite unreliable at delivering us cognitively virtuous states in our world – much less reliable than some feasible alternative” (p. 184). [8]

This last objection goes to the heart of the naturalist’s rejection of the a priori (and with it, the incorrigibility of any principles which are advertised as such): why should we think that our “intuitions” (whatever these may be) are privileged and more worthy of our allegiance (apparently, “worthy of our allegiance” here means “reliable”) than the empirical knowledge on which the naturalist justifies his claims?  After all, don’t our most basic intuitions often clash?  Isn’t natural science the most objective available arbiter of epistemological principles?  Kaplan, though sympathetic to the naturalists’ concerns, responds

“…  I am also struck by the fact that, to show us all this [that the naturalist project is to be preferred to the a priorist], the proponents of epistemological naturalism have offered us a series of arguments…  Like it or not, the naturalists’ attempt to show us the errors of a prioristic methodology depends for its success on our consulting, and finding naturalist arguments in accord with, the very sorts of armchair intuitions whose advice the naturalists would have us ignore” (pp. 184, 5).

Kaplan develops this incisive and potentially deadly response quite tentatively, concluding that until we are able to reliably “tweak our cognitive mechanisms” in ways other than offering arguments, as long as people wish to evaluate the propriety of their beliefs and confidence in propositions, we should continue to honor the a priori constraints on the same.  If that is the right response, then the naturalists would seem to have won the day, so that we wait upon the findings of science to inform us as to the reliability of various “cognitive strategies” (in the meantime, we shall have to make do with principles which accord with our best intuitions).  Moreover, Kaplan implies by his concession that the “profound” epistemological significance of his work comes with an expiration date.  This is disappointing to say the least, and does not represent the full force of Kaplan’s objection.

            Kaplan’s response uncovers the deep weakness which poisons naturalist epistemology – it cannot consistently make a posteriori, empirical hypotheses the ultimate principles of justification, while at the same time licensing arguments which appeal to “untested” non-empirically based intuitions in order to justify the naturalist position to the unconverted.  The naturalist may respond that he has made no such appeal, since what we ordinarily recognize as a priori and non-empirical (logic and mathematics) are simply the ultra-generalized empirical truths Quine held them to be.  In that case, the naturalist’s position becomes even more dire – for every justificatory principle introduced to justify some proposition itself stands in need of justification (in terms of being a principle produced by some properly-functioning and reliable belief-forming cognitive mechanism) – a state-of-affairs appropriately dubbed an infinite “meta-regress.” [9]   A dilemma therefore rears its head against the naturalist – acknowledge the fundamental role of the a priori in justifying knowledge claims and deny naturalism (as formulated by Kitcher), or concede that one’s “reliable sources” cannot, in principle, be vindicated as such. [10]

Decision Theory’s Scope: Diachronic Constraints Revisited

            It is Kaplan’s stated purpose to show that

“…  [T]he Bayesian approach to epistemology…  makes an important contribution to epistemology…  [W]e can still derive results of profound consequence for the way in which we are accustomed to think about (and conduct) the enterprise of inquiry, criticism, and justification” (p. 181).

However, this last claim comes as something of a puzzle in light of his defense of the claim that “… [D]ecision theory does not, of itself, impose any diachronic constraints on our states of confidence” (p. 181) in the second chapter.  But each one of “inquiry, criticism, and justification” presupposes a process of revising confidence assignments. [11]   If Kaplan’s decision theory has nothing at all to say about the way in which a rational person should revise his confidence assignments, wherein lay the “results of profound consequence” for epistemology?

            We can better see the trouble through illustrations: let us imagine that Homer is reviewing his con-assignments (perhaps he constructs imaginary bets for himself to accept) and realizes that they are inconsistent – for some of his con-assignments, con(P) + con(~P) ¹ 1. [12]   If Homer reflects on this state-of-affairs, he will recognize the need to repair this inconsistency, but he will look in vain through the pages of Decision Theory as Philosophy for any guidance whatsoever.  Indeed, any revision at all is compatible with Kaplan’s complete decision theory, just as long as the revision makes con(P) and con(~P) sum to unity. [13]   Perhaps this amounts only to an objection to subjectivism about con-assignments; however, since my present target is not subjectivism, I will turn to a different case that better pinpoints Kaplan’s view.

Suppose that Homer is fooling around with a deck of cards.  He has examined the deck a number of times and his confidence that the deck is standard is quite high.  After shuffling the deck thoroughly, he draws a card from it.  Instead of looking at the drawn card, he hands it over to his honest friend Marge, who examines the card and reports that the card is “red.”  Given all this, and that Homer is aware of his friend’s honesty, [14] he should be twice as confident that the card is a heart as he was before Marge said anything, speaking roughly.  That is,

con(Homer has drawn a heart) = 1/4 ± e₁,

and

con(Homer has drawn a heart ½The card is red) = 1/2 ± e₂,

where e₁ and e₂ are some small margins-of-error.  This conditional con-assignment is related to the first con-assignment, and two other perfectly justified con-assignments (con{The card is red} = 1/2 and con{The card is red½Homer has drawn a heart} = 1)  by Bayes’ theorem.  Moreover, Bayes’ theorem is derivable from Kaplan’s own definition of a conditional con-assignment, as is Jeffrey’s principle.  Of course, as Kaplan would be quick to point out, it doesn’t follow that conditionalizing by Bayes’ theorem – so that Homer’s new con(Homer has drawn a heart) = 1/2 ± e₂ – is compulsory for Homer on that account.  But why not?  For the decision theorist, the answer must lie within the logic of preference.  For many decision theorists, Dutch book arguments provide the motivation for preferences which sanction conditionalization via Bayes’ or Jeffrey’s rule.  Yet, as Kaplan points out, this reliance on Dutch books forgets that a person might harbor no conditionalization strategy at all, so that he is uninterested in such diachronic constraints (p. 71).

A careful examination [15] of Dutch book arguments finds them wanting, but this need not reflect negatively on the epistemic principles which they are intended to support.  There are independent (i.e., non-decision theoretic), a priori reasons, which constrain a rational person to revise con-assignments according to principles like Bayes’ theorem, at least in cases like Homer’s just described.  It is not clear to me that revising one’s state of opinion according Bayes’ theorem or Jeffrey’s principle is always rational.  This much is clear, however, that only Homer’s failure to appreciate his available evidence would result in his being much more (or much less) confident in having drawn a heart than is licensed by the conditional probability of drawing a heart given that the card is red (so that his revised con(Homer has drawn a heart) ¹ con(Homer has drawn a heart ½The card is red)). [16]   Though preference axioms do not guide us toward such a rationale, we hardly need despair that no rational principles for belief-revision can be found, only we must acknowledge that preference principles are not suitable guides for the search.  As Kaplan himself acknowledges,

“Decision-theory-plus-Evidence does not exhaust what reason has to say about what is evidence for what.  Rather it provides a formal constraint that any full and adequate account of what is evidence for what (and, indeed, any account of what is better evidence for what) will have to satisfy.  We will have to look to the more usual sources of epistemic insight for the rest” (p. 88). [17]

We tentatively conclude that Kaplan’s point is that satisfying Modest Probabilism (along with Evidence and Rational Belief) is necessary but not sufficient for epistemic rationality (p. 186). [18]   Since Kaplan admits that his decision theory doesn’t exhaust the rational requirements on a person’s state of opinion, it is compatible with his position that conditionalizing à la Jeffrey (where it applies) is rationally required.

            This last observation manifests the minimal scope, on Kaplan’s view, of an epistemology informed only by decision theory.  And yet it seems that he has not convinced himself that decision theory is so modest, for in the light of our “tentative conclusion” above, a reader will once again be puzzled by Kaplan’s concluding section.  In this finale, Kaplan delivers a “methodological moral” with a test case (p. 186): a lie detector test, which spots 95% of liars as such and discerns truth-tellers with the same reliability, is applied to a suspect.  Should you, given that you are originally quite confident that the suspect is not lying, be quite confident that he is lying after the test indicates that he is lying?  The answer is “yes” only if one’s original confidence in the suspect’s dishonesty is sufficiently high, so that conditionalizing on the polygraph-evidence results in a sufficiently high posterior confidence in his guilt.  That is, when assessing the impact of evidence on one’s present con-assignments, a rational person must not neglect to figure in his prior con-assignments.

“What makes Modest Probabilism and Rational Belief so interesting is that they are able to show that discounting your antecedent degree of confidence… is a mistake” (p. 189).  [My emphasis on the word “mistake.”]

But this innocuous observation simply recapitulates the “Orthodox Bayesian” story about conditionalization, according to which decision theory (and not some non-decision theoretic aspect of rationality) does impose diachronic constraints on opinion.  If we suppose that Kaplan does not intend to invoke conditionalization, but only his weaker Evidence principle, it becomes altogether baffling what the sentence quoted above means.  Reading his “surprising conclusion” (p. 188), it becomes clear that the methodological moral is not that evidence is positively relevant if and only if the corresponding conditional con-assignment is greater than its absolute con-assignment.  Instead, Kaplan’s lesson refers us to Bayes’ theorem itself.  Specifically, if one is “prepared to conditionalize on the result [the lie-detector test’s result in this case],” then either Bayes’ theorem or Jeffrey’s principle is the appropriate rule for such revision. [19]

            Pace Kaplan in his concluding remarks (p. 181), sometimes his Decision theory does impose diachronic constraints on con-assignments.  Apparently, these decision-theoretic constraints are in place just in case the decision-maker is “prepared to conditionalize” on some evidence.  Kaplan does not make clear what he has in mind by such preparation, or when a person ought to be so prepared, [20] but the conditions explicit in Decomposition suggest that valuing bets or strategies like those offered by the Dutch bookie suffices for preparation (pp. 10, 68).  This understanding of preparation drives Kaplan’s rejection of diachronic Dutch Book arguments, where he points out that a person need not place value on bets concerning his future con-assignments, and hence need not commit to any revision strategy (pp. 70-1).  However, the number of states-of-affairs to which one assigns value limits even the scope of the synchronic constraints that decision theory imposes on a believer’s con-assignments.  To a first approximation, as go the decision-theoretic diachronic constraints on con-assignments, so go the synchronic constraints. [21]   Unless Kaplan wishes to make a special plea for decision theory’s synchronic constraints, he should admit the place of diachronic constraints within his decision theory.

The Assertion View

            Kaplan introduces the now well-worn paradox of the preface and lottery paradox in order to show that ordinary talk of “belief” as categorically opposed to “disbelief” faces a dire challenge (he calls it “the Bayesian Challenge”).  Our more-or-less pre-theoretical notion of belief is somewhat ambiguous and lands us in contradictions. [22]   Thus, for the sake of consistency and clarity, we must refine our concepts of belief or else throw them out and speak only of “degrees-of-belief.”  Unlike some Bayesians, however, Kaplan argues that there remains a distinctive role for belief to play in an epistemology informed by decision theory.  The Assertion View, as he calls it, describes this role for belief as a person’s disposition to assert, deny, or remain silent regarding propositions within a context of inquiry where truth is the target.  He proceeds in sections V through VII of chapter four to provide necessary conditions on rational belief (those things, which by one’s own lights, are worth asserting).

For the sake of clarity, let us trudge through the reasoning of the lottery paradox in order to see whether Kaplan’s solution is the only one, or the best.  We begin by considering a fair lottery, in which each ticket is as likely to win as any other.  If T₁, T₂, … T_n stand for the propositions expressed by “Ticket one is a loser,” “Ticket two is a loser,” … “Ticket n is a loser,” respectively, then for a fair lottery with n tickets (where F says that the lottery is indeed fair, and Prob(T_i½F) means “the probability that the i^th ticket will win, given that it has been selected at random from n others in a fair lottery”):

(A)  Prob(T_i½F) = 1/n [23]

and

(B)  Prob ({$i}{~T_i}½F) = S_n Prob(T_i½F) = 1

With (A) in mind, we would be justified in saying about any single ticket (or a relatively small proportion of tickets in the lottery) that it is almost certainly a loser.  Hence, it is ordinarily acceptable either

(C) to believe that any single ticket (or small proportion of tickets) is a loser;

or

(D) to believe that the probability that any single ticket (or small proportion of tickets) is a loser is very high;

or

(E) to believe (with a high degree of certainty) that any single ticket (or small proportion of tickets) is a loser.

Note that (C), (D), and (E) carry distinct meanings.  It is the thought embodied in (C) that is exploited to create the lottery paradox.  Given the following indubitable principle,

(F) If S believes that p and S also believes that q, then reason demands that S believes that p and q, on pain of contradiction, [24]

(C) produces a contradiction.  For S will believe that the first ticket is a loser, and that the second ticket is a loser, and the same for the third, etc.  It follows, by (F), that S believes both

(T₁&T₂…&T_n)

and

            ~(T₁&T₂…&T_n)  [on account of believing that one ticket will win].

Hence the paradox.  Therefore, when we are being explicit, it would best suit us to abandon the language of (C).  (D) and (E) remain as candidates: (E) is urged upon us by eager subjectivists who speak of “degrees-of-belief” and the like, but this view is not endorsed by a phenomenology of belief, or out of respect for ordinary language conventions – which are ambiguous and cannot judge between the language of (D) and (E) – but rather supported by (I daresay) operationalist considerations, by which probability and/or confidence, taken loosely, are defined in terms of some individual’s valuing (or, by being disposed to value) bets on propositions.  It is not surprising, then, to find that Kaplan (a subjectivist) falls into the trap of reasoning that since (C) is for our present purposes untenable, only (E) remains.  In fact, either (D) or (E) can pass the test posed by the paradoxes given; however, (D) does not invoke degrees-of-belief.  Instead of casting belief as a matter of degree, we can picture belief – itself bivalent or “dichotomous” – as having as an object a proposition concerning some probability. [25]  Kaplan has somehow failed to appreciate this alternative, but this confusion is only partly responsible for his asserting the Assertion View.

            Kaplan’s promise for the Assertion View of belief is twofold: first, the Assertion View recaptures an epistemologically significant sense of “dichotomous belief” that has no plausible alternatives compatible with probabilistic reasoning (p. 142); second, the Assertion View represents a cogent answer to the skeptical anti-realist (van Fraassen, in particular), since it shows how we can believe that which is “improbable.” [26]   The first claim is false: we have seen already the beginnings of an alternative account of belief compatible with probabilistic epistemology.  According to dichotomous (D), beliefs do not have degrees (though one may have stronger feelings about some propositions than others), but the probabilities at which beliefs are directed do.  In this case, our lottery presents no problem at all – S’s believing that

[Prob(T₁) = 1/n] & [Prob(T₂) = 1/n] …  & … & [Prob(T_n) = 1/n]

does not have the undesirable paradoxical consequence.  Of course, S’s believing this conjunction entails that he believes, or should, that these probabilities (and not degrees-of-belief or con-assignments) sum to one, by definition of our “fair lottery.”  I do not mean to argue that the Assertion View (hereafter, “AV”) is inferior to the just described dichotomous view (hereafter, “DV”); however, in what follows I hope to show that there is little reason to hold AV, as its first claim to epistemic significance is false, and the second, we shall see, is hardly less dubious.

            According to Kaplan, though “we have no business betting on the strict and literal truth of any ambitious theory” (p. 119), we may rationally assert them (call them “global theories” – p. 137), and thus believe these improbable theories. [27]   This puzzling claim – a consequence of AV – he attempts to mitigate by citing what he takes himself to have shown already: that no coherent account of belief can satisfy the demands of Deductive Cogency and Modest Probabilism while remaining faithful to commonsense views about belief.  Kaplan then attempts to transform this disappointing result into a great virtue by offering it as an antidote to the anti-realist argument given by Bas van Fraassen (mentioned on p. 118).  On the contrary, as I understand “probability,” a hypothesis’ probability just is a measure of the degree to which a person’s actual evidence supports that hypothesis. [28]   Roughly, the more reasons/evidence I have for some hypothesis the greater probability it has for me (or, for the subjectivist, conditionalizing on what I regard as positive evidence for some hypothesis, my degree-of-belief increases). [29]   Now I may believe all sorts of improbable hypotheses (things for which I lack evidence) on account of my intellectual proclivities; however, I take it that what the skeptic wanted was not a psychology of people’s theory preferences, but whether such ambitious beliefs are rationally defensible.

When Kaplan discusses the method by which we can adjudicate between theories, he raises two criteria which may ground a justification of believing “improbable” theories (pp. 132-6): first, as alluded to already, we prefer more ambitious theories to less, second, we prefer theories that more accurately explain things than less.  These two criteria are often in tension – more ambitious theories tend to “fit the world” more loosely than less ambitious theories, though this is not a necessary truth.  Moreover, it follows from elementary probability theory that (as with Kaplan’s own theorem T2 – see p. 212) when P entails Q – so that P is at least as ambitious as Q, [30]

(T2’) Prob(Q) ³ Prob(P) [where P entails Q].

And yet some ambitious theories may be well confirmed by the available evidence – T2’ does not indicate otherwise.  I believe Kaplan misses a crucial point here (perhaps as a consequence of his waffling views about the impact of evidence on opinion): initially improbable does not entail forever improbable!  Indeed, at least a few fairly ambitious scientific theories have much in the way of evidence to support them.  This evidence raises the probability of our ambitious theory in question.  But our skeptic is not satisfied – he argues that the claim of empirical adequacy for a theory will always be somewhat more probable, and hence that mere empirical adequacy is all that can claim our confidence.  As I argue elsewhere, this anti-realism is a mistake; whatever is evidence for empirical adequacy is also evidence for its realist truth as well.  Moreover, it is always possible to weaken a proposition until it is completely uninformative so that its probability will be 1.  But the hypothesis of theory T’s empirical adequacy and that of its truth concerning unobservables are not incompatible propositions, and unless we wish to subject ourselves to another form of lottery paradox, it will not do to ignore the high probability of the realist hypothesis.  That is, maintaining either realism or anti-realism as dogma about some theory T amounts to a leap beyond what evidence warrants.  Though I agree with Kaplan in general that anti-realism is wrong, his AV does not offer a cogent rebuttal to it. [31]

Kaplan’s defense of “believing the improbable” is certainly motivated by a desire to respond to van Fraassen’s anti-realist arguments (essentially, that the constructive empiricist’s beliefs are guaranteed to be more probable than the logically stronger realist beliefs).  In addition, should AV constitute a legitimate justification of realism, other philosophers might count that as a point in favor of accepting AV.  But, as above, Kaplan’s “solution” to the problem – AV, in this case – is actually less plausible than either the dichotomous view I’ve defended or the rejection of categorical belief in favor of degrees-of-belief only.  In summary, there is little reason to accept the Assertion View of belief: first, there is plausible alternative to AV compatible with the paradoxes and quite a bit in accord with “commonsense” views; and second, I’ve shown that while AV is a dubious response to skeptical anti-realism, more traditional probabilistic (even Bayesian) considerations may be sufficient to defuse that problem. [32]

Conclusions

In the space of this thematic review of Kaplan’s book I have at times defended, at times attacked, his decision theory, as well as the claims he makes concerning its relevance to theories of knowledge.  On the one hand, he quite correctly approaches matters of rational belief from an “axiomatic,” a priorist standpoint – as opposed to the naturalist approach.  All I have done is to marshal a better apology for this a priorism.  On the other hand, Kaplan’s stated position regarding the dynamic role of evidence and its measured effect on an individual’s state of opinion gives rise to his wavering over diachronic constraints and to an ill-founded defense of holding to “ambitious” hypotheses. 

Though having left much unsaid, both of praise for his book and further rebuttal, I have attempted a fairly wide-ranging response to Decision Theory as Philosophy.  However, I hope the reader will have detected a thread of criticism running through each section, unified by its simultaneous embrace of the logic of probability with its great significance for epistemology, and a considered suspicion directed at certain key aspects of subjective Bayesianism (defining probability as betting-odds, in particular), along with the problems spawned by its fairly extreme form in Kaplan’s decision theoretic epistemology.  We just do not need the preference axioms to arrive at a sound probabilistic epistemology, nor does decision theory offer anything else to epistemology that could not otherwise be had.  Indeed, our epistemic intuitions are if anything clouded by framing epistemic questions as decision problems (i.e., the betting scenario with which Kaplan commences his argument for Modest Probabilism – see “A Decision Problem,” pp. 1-2). [33]   If nothing else, I hope to have convinced the reader that in order to find cogent reasoning about the important epistemological concepts of justification and probability, he must look elsewhere than Decision Theory as Philosophy – whatever the book’s other merits.

Works Cited

Christensen, David.  “Dutch-Book Arguments Depragmatized: Epistemic Consistency for Partial Believers.”  The Journal of Philosophy.  93;9 (1996): 450-79.

Kaplan, Mark.  Decision Theory as Philosophy.  New York: Cambridge University Press, 1996.

Kennedy, Ralph and Charles Chihara.  “The Dutch Book Argument: Its Logical Flaws, Its Subjective Sources.”  Philosophical Studies.  36 (1979): 19-33.

Kitcher, Philip.  “The Naturalists Return.”  The Philosophical Review.  101;1 (1992): 53-114.

Kyburg, Henry, Jr.  Probability Theory.  Englewood Cliffs, NJ: Prentice-Hall, 1969.

McGrew, Timothy and Lydia.  “What’s Wrong with Epistemic Circularity.”  Dialogue 39 (2000): 219-39.

Schick, Frederic.  “Dutch Bookies and Money Pumps.”  The Journal of Philosophy. 83;2 (1986): 112-9.