Scholarlink

<p>When robots are given unethical commands, they must respond in effective, yet appropriate ways. In previous work, Mott et al. presented experimental evidence arguing that robots must use <i>bounded proportionality</i> when responding to norm violations, in which they offer effective, yet appropriate responses by limiting themselves to direct, formal language over indirect, informal language. Yet Mott et al.’s insights were drawn from a small group of university students, and leveraged only quantitative results. In this work, we thus perform a mixed-methods replication of Mott et al.’s work with a large, diverse set of online participants (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\textrm{n}=200\)</EquationSource> </InlineEquation>). Our results not only support Mott et al.’s findings, but provide stronger and clearer evidence thereof. Our qualitative results further emphasize that indirect linguistic cues are perceived as uncanny, irritating, or conflicting with other norms of collaboration. Moreover, our qualitative results highlight key technical, sociotechnical, and power-laden concerns held by participants that reveal important insights for the future design and deployment of morally competent robots.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

We Don’t Ask Such Things: A Mixed-Methods Reassessment of Bounded Proportionality in Robotic Norm Violation Response

  • 期刊论文

  • 发表日期：

    2026年1月20日

查看全文
  • Terran Mott,
  • Cailyn Smith,
  • Aaron Fanganello,
  • Tom Williams

摘要

When robots are given unethical commands, they must respond in effective, yet appropriate ways. In previous work, Mott et al. presented experimental evidence arguing that robots must use bounded proportionality when responding to norm violations, in which they offer effective, yet appropriate responses by limiting themselves to direct, formal language over indirect, informal language. Yet Mott et al.’s insights were drawn from a small group of university students, and leveraged only quantitative results. In this work, we thus perform a mixed-methods replication of Mott et al.’s work with a large, diverse set of online participants ( \(\textrm{n}=200\) ). Our results not only support Mott et al.’s findings, but provide stronger and clearer evidence thereof. Our qualitative results further emphasize that indirect linguistic cues are perceived as uncanny, irritating, or conflicting with other norms of collaboration. Moreover, our qualitative results highlight key technical, sociotechnical, and power-laden concerns held by participants that reveal important insights for the future design and deployment of morally competent robots.