Taking algorithmic (vs. human) advice reveals different goals to others

People are increasingly likely to obtain advice from algorithms. But what does taking advice from an algorithm (as opposed to a human) reveal to others about the advice seekers’ goals? In five studies (total N = 1927), we find that observers attribute the primary goal that an algorithm is designed to pursue in a situation to advice seekers. As a result, when explaining advice seekers’ subsequent behaviors and decisions, primarily this goal is taken into account, leaving less room for other possible motives that could account for people's actions. Such secondary goals are, however, more readily taken into account when (the same) advice comes from human advisors, leading to different judgments about advice seekers’ motives. Specifically, advice seekers’ goals were perceived differently in terms of fairness, profit-seeking, and prosociality depending on whether the obtained advice came from an algorithm or another human. We find that these differences are in part guided by the different expectations people have of the type of information that algorithmic- vs. human advisors take into account when making their recommendations. The presented work has implications for (algorithmic) fairness perceptions and human-computer interaction.

随着算法在人们生活中的广泛应用，人们越来越倾向于从算法而非人类那里获取建议。然而，从算法（而非人类）那里获取建议对他人而言，揭示了建议寻求者的哪些目标？在五项研究中（总样本量 N = 1927），我们发现观察者会将算法在特定情境下旨在追求的主要目标归因于建议寻求者。因此，在解释建议寻求者后续的行为和决策时，主要考虑的是这一目标，从而为其他可能解释人们行为的动机留下了较少的空间。然而，当（相同的）建议来自人类顾问时，这些次要目标更容易被考虑，导致对建议寻求者动机的判断出现差异。具体而言，根据获取的建议是来自算法还是另一个人，人们对建议寻求者的目标在公平性、盈利性和亲社会性方面的感知存在差异。我们发现，这些差异部分是由人们对算法顾问和人类顾问在制定建议时考虑的信息类型的不同期望所引导的。本研究对于（算法的）公平性感知以及人机交互具有启示意义。