No abstract

One of the main lines of research in algorithmic fairness involves individual fairness (IF) methods. Individual fairness is motivated by an intuitive principle I call "similar treatment," which requires that similar individuals be treated similarly. IF offers a precise account of this definition using distance metrics to evaluate the similarity of individuals. Proponents of individual fairness have argued that it gives the correct definition of algorithmic fairness, and that it should therefore be preferred to other methods for determining fairness. I argue that individual fairness cannot serve as a definition of fairness. Moreover, IF methods should not be given priority over other fairness methods, nor used in isolation from them. To support these conclusions, I describe four in-principle problems for individual fairness as a definition and as a method for ensuring fairness: (1) counterexamples show that similar treatment (and therefore IF) are insufficient to guarantee fairness; (2) IF methods for learning similarity metrics are at risk of encoding human implicit bias; (3) IF requires prior moral judgments, limiting its usefulness as a guide for fairness and undermining its claim to define fairness; and (4) the incommensurability of relevant moral values makes similarity metrics impossible for many tasks. In light of these limitations, I suggest that individual fairness cannot be a definition of fairness, and instead should be seen as one tool among many for ameliorating algorithmic bias.

Philosophers have recently been interested in the appropriate attitude to have toward a controversial philosophical theory. There are a variety of reasons to doubt that belief is appropriate, as the high epistemic standards required for justified belief would seem to exclude too many philosophical theories as viable options. Relatedly, there are worries about assertions of philosophical claims, since assertion is also thought to be governed by substantive epistemic norms. In light of this, a variety of alternative attitudes have been proposed in order to characterize philosophical commitment and assertion. 1 Alexandra Plakias (2019) has helpfully expanded this discussion to the topic of philosophical publication. Plakias argues that it is permissible for philosophers to write and publish works (e.g., journal articles or books) that include claims they do not believe, a practice she calls publishing without belief (PWB). 2 Her thesis is carefully limited to philosophy, as opposed to publishing in other fields. Plakias' appeals to three helpful cases to provide a compelling argument that PWB is sometimes permissible. However, this raises further questions: just when is PWB permissible? Is it universally or even generally permissible? I argue that only claims of a certain kind are permissible to publish without belief. The picture I propose recognizes two roles for claims made in published works: what I will call the advocacy and evidential roles. Advocacy claims should often be published despite authors' lack of belief in them. Evidential role claims, however, require a higher epistemic standing, one that does include a belief requirement. On this theory, Plakias' cases are permissible examples of PWB because they involve advocacy role claims. I will offer another three cases illustrating that it is impermissible to publish evidential role claims without belief. The resulting account of the norms of publishing is applicable to research areas beyond just philosophy.

I argue that recognizing a distinct doxastic attitude called endorsement, along with the epistemic norms governing it, solves the self-undermining problem for conciliationism about disagreement. I provide a novel account of how the self-undermining problem works by pointing out the auxiliary assumptions the objection relies on. These assumptions include commitment to certain epistemic principles linking belief in a theory to following prescriptions of that theory. I then argue that we have independent reason to recognize the attitude of endorsement. Endorsement is the attitude of resilient and committed advocacy which is appropriate for researchers to have toward their own theory. Recognizing the importance of endorsement, and of its resiliency, gives us reason to deny the epistemic principles that serve as auxiliary assumptions in the self-undermining objection. This defuses the objection, and provides additional support for the theory of endorsement.

No abstract

Scientists, philosophers, and other researchers commonly assert their theories. This is surprising, as there are good reasons for skepticism about theories in cutting-edge research. I propose a new account of assertion in research contexts that vindicates these assertions. This account appeals to a distinct propositional attitude called endorsement, which is the rational attitude of committed advocacy researchers have to their theories. The account also appeals to a theory of conversational pragmatics known as the Question Under Discussion model, or QUD. Hence, I call the theory the EQUD model. Motivating this account is a recognition that the speech act of assertion has two roles to play in research contexts. The first is the advocacy role, in which researchers assert a theory in order to advocate for it. The second is the evidential role, which is used to add to the common stock of information available to a field of inquiry. The EQUD model provides an account of warranted assertion for both these roles in research contexts. This success provides support for the theory of endorsement. It also provides support for information updating accounts of assertion.

One of the main lines of research in algorithmic fairness involves individual fairness (IF) methods. Individual fairness is motivated by an intuitive principle I call "similar treatment," which requires that similar individuals be treated similarly. IF offers a precise account of this definition using distance metrics to evaluate the similarity of individuals. Proponents of individual fairness have argued that it gives the correct definition of algorithmic fairness, and that it should therefore be preferred to other methods for determining fairness. I argue that individual fairness cannot serve as a definition of fairness. Moreover, IF methods should not be given priority over other fairness methods, nor used in isolation from them. To support these conclusions, I describe four in-principle problems for individual fairness as a definition and as a method for ensuring fairness: (1) counterexamples show that similar treatment (and therefore IF) are insufficient to guarantee fairness; (2) IF methods for learning similarity metrics are at risk of encoding human implicit bias; (3) IF requires prior moral judgments, limiting its usefulness as a guide for fairness and undermining its claim to define fairness; and (4) the incommensurability of relevant moral values makes similarity metrics impossible for many tasks. In light of these limitations, I suggest that individual fairness cannot be a definition of fairness, and instead should be seen as one tool among many for ameliorating algorithmic bias.

No abstract