When Should Recommender Systems Not Act?

Abstract

Recommender systems (RSs) can produce useful outcomes, but they can also cause harm. This raises a basic operational question for responsible recommendation: under what conditions should a system withhold the action it would otherwise select? We define justified non-action as withholding this action in favor of an alternative response or no response. The aim is to protect users, reduce harm, or satisfy normative constraints. Rather than proposing a new abstention mechanism, we present justified non-action as a unifying perspective organized around four questions: detection, alternative actions, evaluation, and governance. We illustrate these questions through three ongoing projects involving refusal and human handoff in a university assistant, non-intervention in shopping assistance, and action selection in a conversational RS for multi-domain price comparison. Together, these projects motivate further research on when RSs should not act.