An approach to logical counterfactuals inspired by the Demski prior — AI Alignment Forum