A Summary of my Dissertation: "The Social View of Evidence"

What makes it the case that something is evidence for a hypothesis? Many people think that evidential relations, e.g. “e is evidence for h”, are out there in the world for us to discover. By contrast, my dissertation develops a social view of evidence, on which what is evidence for what is decided by groups in the process of social deliberation.


Conversations about evidential relations happen everywhere. In science, for example, they take place in peer-reviewed journals, conferences, casual conversations, etc. These conversations are regulated by norms of deliberation, such as norms about which dialectical moves are permissible and when deliberation ought to end. The main thesis of my dissertation is that when a social deliberation abides by a set of these norms, the conclusions that emerge have epistemic implications for those who accept them. For example, after a period of debate and discussion, scientists concluded that red spots are evidence for measles. To the extent that this process satisfied the scientific norms of deliberation, red spots became evidence for measles for those who accept these norms.


I support the social view of evidence by showing that it is fruitful: it explains key phenomena related to evidence. One of them is the phenomenon of resistance to evidence. We know this phenomenon from the public sphere, as politicians and members of the public resist evidence on climate change, vaccines, and so on. We also know this phenomenon from feminist science, as feminist scientists push back against long-standing consensuses in medicine, anthropology, and other fields. The social view explains how resistance to evidence works and distinguishes epistemically good from epistemically bad cases of resistance, which other accounts of evidence cannot do.


To explain resistance, I use the fact that individuals can hypothesize about evidence by thinking through whether the claim “e is evidence for h” would survive the right kind of social deliberation, e.g. by running objections and responses in their heads. Resistance occurs when individuals favor their hypothetical over the outcome of social deliberation. We mark resistance as epistemically good when the hypothetical conforms to our norms of deliberation to a high degree, e.g. when the resisting agent can respond to dialectical moves in permissible ways. Feminist scientists are well-positioned to conform to the scientific norms to a high degree given their training. Laypersons, such as some anti-vaxxers, are often ill-positioned. If these anti-vaxxers accept the scientific norms of deliberation, which they seem to be if they accept some scientific evidence, then their denial violates even their own norms.


An implication of the social view, which I develop, is that evidence is political. For example, scientists used to test medication only on men. This inclination reflects norms of deliberation which deprioritize objections about women’s health and is tied to women’s political status. In general, views on which objections we must entertain before concluding the deliberation depend on the group’s priorities, which are influenced by the power structures in and around the group.


You might wonder whether some norms of deliberation are universally better than others no matter what the priorities of the group are. I explore a theorem from machine learning, called the “No Free Lunch Theorem” (NFL), which suggests otherwise. According to NFL, no learning algorithm universally gives more accurate predictions than any other, including all forms of social deliberation. Some algorithms do better than others once we restrict the set of problems we are trying to solve. But no set of norms of deliberation can get us from data to accurate predictions without restricting the set of problems of interest. I use NFL to bring out non-epistemic aspects of accuracy and other measures of goodness of theories, which helps let go of the (misleading) intuition that data can be used to choose between theories in purely epistemic ways.

Recent Posts

See All

How can AI cause harm, and which harms are the most important? In particular, which AI risks are end-users most concerned about? Results fro