On Optimizing Interventions in Shared Autonomy