![Sample our Humanities journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5939/TOC-Subject-Sample-2021-Humanities.jpg"})
ABSTRACT
Is uncertainty about preferences rationally possible? And if so, does it matter for rational decision? It is argued that uncertainty about preferences is possible and should play the same role in rational decision-making as uncertainty about worldly facts. The paper develops this hypothesis and defends it against various objections.
Acknowledgments
I would like to thank Mario Brandhorst, Ralf Busse, Roman Heil, Jakob Koscholke, Thomas Krödel, Patricia Rich, Sergui Spatan and Jacques Vollet as well as the participants of the Humboldt Normativity Conference 2018, a colloquium in Hamburg 2018 and a workshop in Mainz 2019 for various helpful comments. Thanks are also due to an anonymous referee, whose challenging comments led to significant improvements of the paper. This work was supported by the Deutsche Forschungsgemeinschaft under grant SCHU 3080/3-1/2 and was conducted within the Emmy-Noether research group ‘Knoweldge and Decision’.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 The Humean picture is not quite Hume's, for Hume thought that actions ‘cannot be reasonable or unreasonable’ because they cannot be true or false (Hume Citation1739–40, T 3.1.1.10, SBN 458). But this picture takes from Hume the thought that the primary role of reason in action is to select proper means to satisfy one's passions (Hume Citation1739–40, T 3.1.1.12, SBN 459f.).
2 Evidential decision theory and causal decision theory offer more refined versions of expected utility. See Jeffrey (Citation1965) and Lewis (Citation1981).
3 In particular, this is meant to cover non-instrumental (or intrinsic) preferences. See also the discussion on p. 15f.
4 A preference is strong if the difference in utility is large, while a preference is weak if the difference in utility is small.
5 The present discussion is inspired by Williamson's (Citation2000, Chapter 4) Anti-Luminosity Argument. When applied to gradually changing preferences, it delivers the conclusion that at a certain point the agent has no knowledge of whether she prefers a over b. Let me point out that the argument I propose only requires that there are points at which it is rational to be uncertain. If uncertainty is compatible with knowledge, then this conclusion is weaker than what the Anti-Luminosity Argument is supposed to deliver. Cf. Berker (Citation2008). Unnoticeable changes in preferences have been discussed in the context of Quinn's ‘puzzle of the self-torturer’ (Quinn Citation1990), yet to discuss this here would take us too far afield.
6 But see Nissan-Rozen (Citation2010). Nissan-Rozen discusses a constraint like this (under the label ‘Likelihood of Betterness Constraint’) and proves a triviality result indicating that such a constraint cannot be universally satisfied except in very few trivial cases (given certain side assumptions).
7 In the context of Williamson's Anti-Luminosity Argument, DeRose (Citation2002) suggests that concepts may be intransparent at their margins but transparent in a realm of core applications.
8 A caveat: I assume that Katherine has no reason to think that if she has a preference for another child, this preference is weak, while if she has a preference for not having another child, this preference is strong. In such a situation, the risk involved in deciding to have another child could be seen to outweigh the higher probability. As a matter of fact, situation of this kind provide a further reason to suspect that acting on the most likely preference is not always best. See Nissan-Rozen (Citation2010) for discussion (and qualification).
9 Delaying a decision falls within the class of problems studied as search theory in economics and optimal stopping problems in mathematics.
10 This sense of ‘(in)transparency’ is slightly different from how it is used in the discussion about Williamson's (Citation2000) Anti-Luminosity Argument, where transparency means that a subject is always in a position to know whether a given condition obtains. Cf. also fn. 5.
11 Berker (Citation2008) suggests that purely phenomenal conditions such as, potentially, feeling cold may escape Williamson's (Citation2000) Anti-Luminosity Argument. He seems to allow for uncertainty, though (cf. fn. 5).
12 In the more recent debate about the question of whether (certain) mental states are transparent, preferences are, as far as I am aware, not discussed in any detail.
13 As an aside, the same consideration applies to conceptions of preference which link preferences to actual as well as counterfactual choice behavior. As one can be uncertain about how one would choose, one can be uncertain about preferences construed that way.
14 To guard against a possible misunderstanding: I do not wish to suggest that conditions concerning the obtaining of a relation are never purely phenomenal.
15 Such a view could build on Taylor (Citation1985, part I), especially p. 101. Taylor holds that emotions and related desires are partly constituted by episodes of self-interpretation. There is also a rich discussion of the role of self-interpretation and self-determination of one's preferences through deliberative processes in Hurley (Citation1989, Chapters 6 and 15).
16 See also Bradley (Citation2017, Section 6) and Gibbard (Citation1998), the latter building on Gibbard (Citation1990).
17 This does not mean that one has to go as far as what Lewis (Citation1988) has dubbed the desire-as-belief thesis. See Broome (Citation1991) for a viable fallback position.
18 Broome (Citation2006) discusses the further question of whether one can somehow reason with preferences which does not reduce to reasoning about preferences. If this were possible, one might substitute the considerations in favor of an epistemic perspective on preferences by saying that preferences must be occurrent or articulated in order to matter for rational decision. Thanks to Ralf Busse for this suggestion. It remains somewhat unclear, however, how different degrees of uncertainty about preferences could be accommodated on such a picture.
19 There is more than one way to do so; see the discussion in Dietrich and Jabarian (Citation2020).
20 The present formula is designed to cover uncertainty about finitely many utility functions. It would have to facilitate integrals to cover the infinite case.
21 The probability function P is supposed to be the same as the one used to calculate expected value. In both cases P represents the agent's uncertainty (over worldly facts and preferences respectively).
22 The present formula is close in form to Savage's (Citation1954) decision theory (see p. 3). As Savage's theory ignores possible dependencies between actions and worldly facts, the present theory ignores possible dependencies between worldly facts and preferences as well as between actions and preferences. If this turns out to be problematic, one could move the present theory closer to evidential decision theory (Jeffrey Citation1965) or causal decision theory (Lewis Citation1981).
23 Compare the theorem by Good (Citation1967) to the effect that the expected value of acquiring more information is always highest if the information is cost-free and could make a difference for the decision at hand.
24 Nissan-Rozen (Citation2018) worries, in the context of transformative choice and normative uncertainty, that accounts of the present kind might either violate plausible decision-theoretic principles or else associate agents with the wrong kind of motivation. Here I have to confine myself to saying just this: I find the reasoning process as described above entirely natural and not at all suggestive of an objectionable source of motivation.
25 I thank an anonymous referee for alerting me to these challenges.
26 Riedener (Citation2019) provides a helpful classification of solutions to the problem of inter-theoretic comparisons and also offers a (constructivist) solution of his own.
27 To a large extent, this mirrors the account of Ross (Citation2006, 761–65) for inter-theoretic value comparisons. Ross (Citation2006, 765) points out that gradable attitudes like disappointment may be warranted to a certain degree depending on an underlying scale of value.
28 A less committal response may be possible, too. When a single subject is uncertain as to whether she prefers a given option greatly or only slightly, it is plausible that she conceptualize the possible value differences according to a fixed underlying scale. According to this thought, the problem of intra-personal comparisons would vanish on closer inspection because agents evaluate their options all from a single standpoint.
29 There are also long-run arguments (based on the law of large numbers) claiming that agents maximizing expected value will do better in the long run as agents not maximizing expected value. A long-run argument may also be available for maximizing expected expected value (yet a comprehensive discussion must wait for another occasion): in the long run, an agent uncertain about her preferences will approximate a gain close to the expected expected value.
30 Most of the objections are condensed versions of discussions I had at the Humboldt Normativity Conference 2018, a colloquium in Hamburg 2018 and a workshop in Mainz 2019.
31 Zimmerman (Citation2008, Chapter 1.5) accepts this conclusion in the slightly different setting of moral uncertainty. See also Schulz (Citation2017) for a possible connection between higher-order uncertainty and high-stakes contexts.
32 Examples of this kind derive from the debate about the contingent a priori starting with Kripke (Citation1980).
33 For the distinction between extrinsic and intrinsic preferences, see Lumer (Citation1998).