Machine learning systems impact many stakeholders and groups of users, often disparately. Prior studies have reconciled conflicting user preferences by aggregating a high volume of manually labeled pairwise comparisons, but this technique may be costly or impractical. How can we lower the barrier to participation in algorithm design? Instead of creating a simplified labeling task for a crowd, we suggest collecting ranked decision-making heuristics from a focused sample of affected users. With empirical data from two use cases, we show that our weak learning approach, which requires little to no manual labeling, agrees with participants' pairwise choices nearly as often as fully supervised approaches.
翻译:机器学习系统往往对许多利害相关者和用户群体产生不同的影响。 先前的研究通过汇集大量手工标签的对口比较,调和了相互冲突的用户偏好,但这一方法可能成本高,也可能不切实际。 我们如何降低参与算法设计的障碍? 我们建议,不要为人群制造简化标签任务,而应该从集中的受影响用户抽样中收集决策等级累赘。 根据两个使用案例的经验数据,我们发现,我们薄弱的学习方法几乎不需要人工标签,我们同意参与者的对口选择,就像完全监督的方法一样。