Max Forster
Stanford University
How should agents adjust their credences when faced with higher-order de-
feat? According to level-splitters, agents may rationally retain their first-order
credence, while lowering their higher-order confidence that their credence is
the right one to have. Steglich-Petersen (2019) endorses a version of level-
splitting, but adds an important requirement: higher-order defeat rationally
requires agents to lower their credences’ resilience (Skyrms 1977), thereby dis-
posing them to update more readily in response to future evidence.
This proposal faces a serious problem, which I call the problem of nor-
mative over-determination. The challenge is this: If first-order updating is
governed by Bayesian conditionalization, how can higher-order evidence im-
pose additional rational constraints without generating a threat of normative
over-determination: a situation in which both first-order Bayesian norms and
higher-order constraints on resilience independently dictate how an agent should
respond to new evidence?
In this paper, I develop and defend a unified account of resilience adjustments
within the standard Bayesian framework. Drawing on a Gaussian model of
agents’ epistemic self-assessments, I propose an updating rule that generalizes
Bayesian conditionalization to a range of higher-order evidential contexts.
This generalized updating rule preserves the core normative commitments
of Bayesian conditionalization while providing a precise formal interpretation
of Steglich-Petersen’s resilience requirement. The result is a framework that
vindicates level-splitting, respects Bayesian norms, and offers a unified account
of how higher-order defeat rationally constrains agents’ updating.
Chair: tba
Time:
Location:
SOPhiA 2026 