A systematic review of algorithm aversion in augmented decision making

Abstract

Despite abundant literature theorizing societal implications of algorithmic decision making, relatively little is known about the conditions that lead to the acceptance or rejection of algorithmically generated insights by individual users of decision aids. More specifically, recent findings of algorithm aversion—the reluctance of human forecasters to use superior but imperfect algorithms—raise questions about whether joint human‐algorithm decision making is feasible in practice. In this paper, we systematically review the topic of algorithm aversion as it appears in 61 peer‐reviewed articles between 1950 and 2018 and follow its conceptual trail across disciplines. We categorize and report on the proposed causes and solutions of algorithm aversion in five themes: expectations and expertise, decision autonomy, incentivization, cognitive compatibility, and divergent rationalities. Although each of the presented themes addresses distinct features of an algorithmic decision aid, human users of the decision aid, and/or the decision making environment, apparent interdependencies are highlighted. We conclude that resolving algorithm aversion requires an updated research program with an emphasis on theory integration. We provide a number of empirical questions that can be immediately carried forth by the behavioral decision making community.

Justice perceptions of artificial intelligence in selection

“…Organizations might consider embedding a certain amount of human interaction within their AI‐based staffing systems even at early stages, in order to help ensure that applicants feel they are valued as individuals rather than as data points feeding AI, thereby reducing applicant reliance on negative fairness heuristics that may impact outcomes further into the selection process. This suggestion is supported by Burton, Stein, and Jensen’s (2019) suppositions that human‐algorithm decision systems outperform single‐system decision makers, largely by mitigating aversion and mistrust of AI. One way to achieve this human‐AI blend can be combining the characteristics of AI interviews with that of video‐conference interviews, such that the interview is performed by HR personnel through video chat software, while the responses are still recorded and processed by the AI.…”

Changing the means of managerial work: effects of automated decision support systems on personnel selection tasks

“…On the one hand, it has shown that mechanical gathering and combining of information (e.g., combination using ordinary least squares regression) can improve decision quality when compared to clinical gathering and combining of information (e.g., intuition-based combination of information) (Kuncel et al, 2013). On the other hand, this literature found that people are skeptical of the use of mechanical gathering and combining of information (Burton et al, 2019). Some researchers even refer to the reluctance to use mechanically combined information as algorithm aversion (Dietvorst et al, 2015;but cf.…”

“…Some researchers even refer to the reluctance to use mechanically combined information as algorithm aversion (Dietvorst et al, 2015;but cf. Logg et al, 2019) and indicate that using automated systems within jobs might affect people's behavior and reactions to the job (Burton et al, 2019).…”

“…In a review on the topic of algorithm aversion, Burton et al (2019) argue that using automated support systems might have a variety of effects on decision-makers. For instance, decision-makers might feel reduced autonomy when receiving decision support, expectations towards systems might not be met by respective systems (Dietvorst et al, 2015;Highhouse, 2008), and certain design options within such systems (e.g., how and when to present a recommendation) might not match to human information processing, thus contributing to reluctance to use such systems.…”

“…For instance, Campion et al (2016) showed that automated processes might help to assess motivational letters, and Naim, Tanveer, Gildea, and Hoque (2018) showed that automated processes can be used to evaluate job interviews. Another stream of research investigates if and how people use decision support by automated systems (Burton, Stein, & Jensen, 2019) and discusses whether people are averse to using recommendations generated by automated systems (Dietvorst, Simmons, & Massey, 2015) or if they appreciate such recommendations (Logg, Minson, & Moore, 2019).…”