arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2212.13995cs.CY
收藏

成为数据驱动规则例外的权利

The Right to be an Exception to a Data-Driven Rule

Sarah H. Cen, Manish Raghavan

摘要:

数据驱动的工具越来越多地用于做出重大决策。他们已经开始建议雇主面试哪些求职者,法官批准哪些被告保释,贷款人向哪些房主提供贷款等等。在这种情况下,不同的数据驱动规则会导致不同的决策。问题是:对于每一个数据驱动的规则,都有例外。虽然数据驱动的规则可能适合某些人,但它可能不适合所有人。随着数据驱动的决策变得越来越普遍,在某些情况下,有必要保护那些并非由于自己的过错而成为数据驱动例外的个人。与此同时,不可能仔细检查越来越多的数据驱动决策中的每一个,这就引出了一个问题:何时以及如何保护数据驱动异常?在这篇文章中,我们认为个人有权成为数据驱动规则的例外。也就是说,不应该假设数据驱动的规则-即使是具有高准确性的规则-适用于任意的决策主题。相反,决策者只有在(相对于伤害风险)采取了应有的谨慎和应有的注意,排除了决策主体是数据驱动规则例外的可能性时,才应适用该规则。在某些情况下,伤害的风险可能很低,只需粗略考虑。尽管在人类驱动的决策环境中,应用应有的谨慎和尽职调查是有意义的,但对于数据驱动的规则来说,这意味着什么还不清楚。我们建议,确定一个数据驱动的规则是否适合于一个给定的决策主体需要考虑三个因素:个性化,不确定性和伤害。我们详细地解释了这一权利,提供了一个评估数据驱动规则的框架,并描述了在实践中调用这一权利意味着什么。

英文摘要:

Data-driven tools are increasingly used to make consequential decisions. They have begun to advise employers on which job applicants to interview, judges on which defendants to grant bail, lenders on which homeowners to give loans, and more. In such settings, different data-driven rules result in different decisions. The problem is: to every data-driven rule, there are exceptions. While a data-driven rule may be appropriate for some, it may not be appropriate for all. As data-driven decisions become more common, there are cases in which it becomes necessary to protect the individuals who, through no fault of their own, are the data-driven exceptions. At the same time, it is impossible to scrutinize every one of the increasing number of data-driven decisions, begging the question: When and how should data-driven exceptions be protected? In this piece, we argue that individuals have the right to be an exception to a data-driven rule. That is, the presumption should not be that a data-driven rule--even one with high accuracy--is suitable for an arbitrary decision-subject of interest. Rather, a decision-maker should apply the rule only if they have exercised due care and due diligence (relative to the risk of harm) in excluding the possibility that the decision-subject is an exception to the data-driven rule. In some cases, the risk of harm may be so low that only cursory consideration is required. Although applying due care and due diligence is meaningful in human-driven decision contexts, it is unclear what it means for a data-driven rule to do so. We propose that determining whether a data-driven rule is suitable for a given decision-subject requires the consideration of three factors: individualization, uncertainty, and harm. We unpack this right in detail, providing a framework for assessing data-driven rules and describing what it would mean to invoke the right in practice.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑