Epistemic Injustice

Epistemic injustice is a wrong done to someone specifically in their capacity as a knower. Miranda Fricker distinguished two forms. Testimonial injustice occurs when prejudice about a speaker’s identity leads a hearer to give her word less credibility than it deserves. Hermeneutical injustice occurs when a gap in a community’s shared interpretive resources leaves someone unable to make sense of, or communicate, an important part of her experience. Later work, including Dotson’s, has extended the analysis to practices of silencing.

The concept bears on artificial systems in more than one way. Systems trained on records of who has historically been believed can inherit that distribution and apply it at scale, without anyone holding the prejudice in a form that could be challenged. Symons and Alvarado have examined how data-science technologies can themselves be sources of epistemic injustice.

A neighbouring wrong appears in my work on algorithmic welfare. A labourer whose worn fingerprint fails to authenticate at the ration shop is not disbelieved. No one hears her and discounts her; the architecture reads the criterion and nothing else, and her manifest entitlement is not among the things it can read. I have analysed this as misrecognition, a status wrong effected through the design of an identification system, rather than as testimonial injustice, since there is no hearer whose prejudice could be corrected. What the two share is that the harm is done to someone as the bearer of a standing that ought to have counted.

An open question is what happens to testimonial injustice when part of the route from speaker to audience is a model that summarises, ranks or paraphrases what people say. The speaker may never learn who, or what, discounted her.