Research

• Bayesian update rules and minimal-change preference revision, International Journal of Approximate Reasoning 2026

Synopsis: Pointwise Bayesian conditioning may discard preference judgments that minimal-change revision can preserve, suggesting that imprecise priors with different origins should sometimes be updated differently.

• Preference revision and Bayesian updating, Synthese 2026

Synopsis: Bayesian conditioning can be uniquely recovered from minimal-change preference revision when prior conditional preferences are preserved as reference anchors, while alternative anchors generate principled departures from Bayesian updating.

• Will power-seeking AGIs harm human society?, AI & Society 2025

Synopsis: Arguments that power-seeking AGIs will threaten humanity rely on anthropomorphic assumptions about their world models; non-human world models both weaken familiar arguments and introduce novel, underappreciated risks.