A systematic review of algorithm aversion in augmented decision making

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.     

Moral uncertainty and permissibility: Evaluating Option Sets

In this essay, we explore an issue of moral uncertainty: what we are permitted to do when we are unsure about which moral principles are correct. We develop a novel approach to this issue that incorporates important insights from previous work on moral uncertainty, while avoiding some of the difficulties that beset existing alternative approaches. Our approach is based on evaluating and choosing between option sets rather than particular conduct options. We show how our approach is particularly well-suited to address this issue of moral uncertainty with respect to agents that have credence in moral theories that are not fully consequentialist.     

Ease‐of‐retrieval effects on procedural justice judgements under conditions of informational and personal uncertainty

This study tests whether individuals' reliance on ease‐of‐retrieval processes when forming procedural justice judgements are moderated by informational and personal uncertainty. In Studies 1 and 2 we examined the predicted effects of informational uncertainty. Results indicated that participants in information‐uncertain conditions relied on ease‐of‐retrieval, whereas those in information‐certain conditions relied on content information to make procedural justice judgements. In Study 3 we examined the combined effects of informational uncertainty and personal uncertainty on reliance on ease‐of‐retrieval when forming procedural justice judgements. The findings of Study 3 indicated that personal uncertain participants who were in informational certain conditions based their procedural justice judgements on content information, whereas all other participants based their procedural justice judgements on ease‐of‐retrieval. This is the first paper to demonstrate that the joint effect of informational uncertainty and personal uncertainty on reliance on ease‐of‐retrieval is different from the two uncertainties acting alone.     

Uncertainty and group identification: Moderation by warmth and competence as cues to inclusion and identity validation

Drawing upon uncertainty-identity theory, we argue that self-uncertainty creates a motivated mindset in which people desire to have their social identity validated thorough inclusion, and that this transforms the meaning of group warmth and competence to cue inclusion. US College students (N = 289) participated in an on-line experiment in which self-uncertainty and the perceived warmth and competence of their college were primed, and group identification was measured as the dependent variable. As predicted, the three-way interaction significantly affected group identification. Specifically, uncertainty weakened identification with cold-but-competent groups. We also found that uncertainty strengthened identification with cold-and-incompetent groups. This research highlights the preference for certain group attributes under uncertainty.     

Clarifying the unique associations among intolerance of uncertainty,anxiety, and depression*

Increasing evidence suggests that intolerance of uncertainty (IU) may be a transdiagnostic factor across the anxiety disorders, and to a lesser extent, unipolar depression. Whereas anxiety inherently involves uncertainty regarding threat, depression has traditionally been associated with certainty (e.g. the hopelessness theory of depression). Some theorists posit that the observed relationship between depression and IU may be due to the relationship between depression and anxiety and the relationship between anxiety and IU. The present study sought to elucidate the unique relationships among trait anxiety, depression, and IU in undergraduate (N = 554) and clinical (generalized anxiety disorder; N = 43) samples. Findings suggest that IU may play a larger role in anxiety than depression, although some evidence indicates that inhibitory IU and depression may have a modest but independent relationship.     

Taste Aversion Learning Based on Swimming and Lithium Chloride Injection in Rats: Implications From Cross-familiarization Tests and Stimulus Selectivity1

In rats, swimming causes avoidance of the taste solution consumed immediately before the swimming. Several lines of research have shown that this taste avoidance reflects Pavlovian conditioned aversion based on correlations between the taste and swimming-induced nausea. The present research compared swimming-based taste aversion learning (TAL) with conventional TAL based on nausea-inducing lithium chloride (LiCl). By exploiting cross-familiarization techniques, Experiments 1A and 1B suggested that different physiological states are induced by swimming and LiCl. This claim was supported by Experiment 2, which reports stimulus selectivity in saccharin and sucrose aversions based on swimming and LiCl.     

The BDNF Val66Met polymorphism impacts victim's moral emotions following interpersonal transgression

Immoral behaviors make individuals abominate and punish transgressors. Inspired by the associations between the Val66Met polymorphism of brain‐derived neurotropic factor (BDNF) gene and emotional responses following negative events, we investigated whether this polymorphism was also associated moral emotions such as punishment and forgiveness following interpersonal transgression. To do so, we categorized 340 individuals according to the BDNF Val66Met and assessed moral emotions by using 12 hypothetic scenarios in different conditions of intention and interpersonal consequence. The results indicated that this polymorphism was significantly associated with moral aversion and punishment towards transgressors. Victims with the Val/Val genotype expressed less aversion and punishment than the Met carriers, regardless of intention and interpersonal consequence. Moreover, this polymorphism was associated with forgiveness. Victims with the Val/Val genotype expressed more forgiveness than the Met carriers. Taken together, these findings highlight the importance of the BDNF Val66Met to moral emotions.     

得失不对称：跨期选择中自我-他人决策差异

采用跨期选择任务范式,考察得失情境下自我–他人决策差异。结果发现：（1）为自己决策比为他人决策更偏好于选择即刻选项;（2）损失情境比获益情境下更偏好于选择即刻选项;（3）获益情境下为自己决策与为他人决策在选择即刻选项上不存在显著差异,而损失情境下为自己决策比为他人决策更偏好于选择即刻选项,表明得失情境下自我–他人决策差异存在不对称性。     

Better Safe than Sorry: Risking Irresponsibility by Seeking Uncertainty

Research findings differ as to whether choosing a risky option is an efficient strategy for decision makers seeking to avoid responsibility for potential failures. A risky choice may leave the final outcome to chance factors, but the decision maker can still be held responsible for choosing risk. Further, it is unclear whether a risky choice is a responsible choice. The present article investigates the putative relationship between risk‐taking and responsibility by drawing a distinction between being responsible for the outcome (R1) versus acting responsibly (R2). Four experiments were performed, in which participants were presented with scenarios describing decision makers facing a choice between a risky (uncertain) option and a riskless (certain) option, framed in terms of losses or equivalent gains. The results showed that decision makers who chose the risky alternative were judged to have acted in a less responsible manner (R2), while still being held equally responsible for the outcome (R1), unless they were ignorant of the risks involved. Choosing risk did not absolve decision makers from blame, despite being less causal and less in control than those who chose the riskless option. Risky decision makers were also judged to be more personally involved. The dissociation between R1 and R2 ratings confirms earlier findings and serves to clarify an alleged relationship between risky choices and responsibility aversion. Framing effects for own choices were found in both scenarios. In contrast, responsibility ratings were only slightly affected by frame. Copyright © 2017 John Wiley & Sons, Ltd.