Algorithm Aversion Research Articles

Humans as Creativity Gatekeepers: Are We Biased Against AI Creativity?

With artificial intelligence (AI) increasingly involved in the creation of organizational and commercial artifacts, human evaluators’ role as creativity gatekeepers of AI-produced artifacts will become critical for innovation processes. However, when humans evaluate creativity, their judgment is clouded by biases triggered by the characteristics of the creator. Drawing from folk psychology and algorithm aversion research, we examine whether the identity of the producer of a given artifact as artificial intelligence (AI) or human is a source of bias affecting people’s creativity evaluation of such artifact and what drives this effect. With four experimental studies (N = 2039), of which two were pre-registered, using different experimental designs and evaluation targets, we found that people sometimes—but not always—ascribe lower creativity to a product when they are told that the producer is an AI rather than a human. In addition, we found that people consistently perceive generative AI to exert less effort than humans in the creation of a given artifact, which drives the lower creativity ratings ascribed to generative AI producers. We discuss the implication of these findings for organizational creativity and innovation in the context of human-AI interaction.

Hooked on artificial agents: a systems thinking perspective

Following recent technological developments in the artificial intelligence space, artificial agents are increasingly taking over organizational tasks typically reserved for humans. Studies have shown that humans respond differently to this, with some being appreciative of their advice (algorithm appreciation), others being averse toward them (algorithm aversion), and others still fully relinquishing control to artificial agents without adequate oversight (automation bias). Using systems thinking, we analyze the existing literature on these phenomena and develop a conceptual model that provides an underlying structural explanation for their emergence. In doing so, we create a powerful visual tool that can be used to ground discussions about the impact artificial agents have on organizations and humans within them.

A Cogitation on the ChatGPT Craze from the Perspective of Psychological Algorithm Aversion and Appreciation.

In recent times, ChatGPT has garnered significant interest from the public, sparking a range of reactions that encompass both aversion and appreciation. This paper delves into the paradoxical attitudes of individuals towards ChatGPT, highlighting the simultaneous existence of algorithmic aversion and appreciation. A comprehensive analysis is conducted from the vantage points of psychology and algorithmic decision-making, exploring the underlying causes of these conflicting attitudes from three dimensions: self-performance, task types, and individual factors. Subsequently, strategies to reconcile these opposing psychological stances are proposed, delineated into two categories: flexible coping and inflexible coping. In light of the ongoing advancements in artificial intelligence, this paper posits recommendations for the attitudes and actions that individuals ought to adopt in the face of artificial intelligence. Regardless of whether one exhibits algorithm aversion or appreciation, the paper underscores that coexisting with algorithms is an inescapable reality in the age of artificial intelligence, necessitating the preservation of human advantages.

Algorithm aversion to mobile clinical decision support among clinicians: a choice-based conjoint analysis

ABSTRACT Artificial intelligence has pervaded clinical decision support systems driven by algorithms. In healthcare, algorithmic decision-making is said to outperform clinicians in their treatment choices and recommendations. Yet, relying on an algorithm outcome for clinical decisions comes with uncertainties resulting from the opacity, inconsistency and inaccuracy usurping the epistemic authority of clinicians and ultimately creating aversion. This study explores latent factors that constitute aversion to using algorithm-augmented mobile clinical decision support (mCDS) applications (apps) by clinicians at the point of care. Given that aversion encompasses a tendency to avoid potential risks, we modelled clinicians’ preferences using choice-based conjoint analysis with Hierarchical Bayes (HB) estimation methods to examine 362 clinicians’ aversion algorithm-augmented mCDS apps. In order of magnitude, the findings indicate that clinicians using mCDS apps are averse to algorithms undermining their epistemic authority with persistently inconsistent results at the point of care. In essence, clinicians will at the point-of-care resist algorithms that restrict them from modifying the result in line with their professional responsibilities. The study has several implications for both research and practice by helping improve potential drawbacks in clinical algorithms while promoting effective frameworks to enhance decision-making in medical practice.

Content Analysis of Judges' Sentiments Toward Artificial Intelligence Risk Assessment Tools

Artificial intelligence (AI) uses computer programming to make predictions (e.g., bail decisions) and has the potential to benefit the justice system (e.g., save time and reduce bias). This secondary data analysis assessed 381 judges’ responses to the question, “Do you feel that artificial intelligence (using computer programs and algorithms) holds promise to remove bias from bail and sentencing decisions?” The authors created apriori themes based on the literature, which included judges’ algorithm aversion and appreciation, locus of control, procedural justice, and legitimacy. Results suggest that judges experience algorithm aversion, have significant concerns about bias being exacerbated by AI, and worry about being replaced by computers. Judges believe that AI has the potential to inform their decisions about bail and sentencing; however, it must be empirically tested and follow guidelines. Using the data gathered about judges’ sentiments toward AI, we discuss the integration of AI into the legal system and future research.

Human Workers at the Center: Algorithm Aversion and Appreciation as Reactions to AI in Organizations

Human Workers at the Center: Algorithm Aversion and Appreciation as Reactions to AI in Organizations

The Social and Moral Dimensions of Algorithm Aversion

The Social and Moral Dimensions of Algorithm Aversion

Task-specific algorithm advice acceptance: A review and directions for future research

Due to digitalization resulting in artificial intelligence advice, there are increasing studies on advice taking, exploring individual and task-relevant factors associated with the acceptance of algorithm advice. However, to our notice, there are no reviews of studies on the acceptance of algorithm advice that focus explicitly on a task level that consider methodological features and provide a quantitative measure of algorithm acceptance. Our review closes these research gaps. We evaluated 44 studies, 122 tasks, and 89,751 participants. Our review shows that algorithm aversion is present in 75% of the 122 considered tasks. In addition, our quantified measures underscore some shortcomings by the underrepresented individual, task, or methodological characteristics—for example, the expertise of advice takers and longitudinal studies. Finally, we provide valuable recommendations to continue research on algorithm acceptance.

Choosing between human and algorithmic advisors: The role of responsibility sharing

Algorithms are increasingly employed to provide highly accurate advice and recommendations across domains, yet in many cases people tend to prefer human advisors. Studies to date have focused mainly on the advisor’s perceived competence and the outcome of the advice as determinants of advice takers’ willingness to accept advice from human and algorithmic advisors and to arbitrate between them. Here we examine the role of another factor that is not directly related to the outcome: the advice taker’s ability to psychologically offload responsibility for the decision’s potential consequences. Building on studies showing differences in responsibility attribution between human and algorithmic advisors, we hypothesize that, controlling for the effects of the advisor’s competence, the advisor's perceived responsibility is an important factor affecting advice takers’ choice between human and algorithmic advisors. In an experiment in two domains, Medical and Financial (N = 806), participants were asked to rate advisors’ perceived responsibility and choose between a human and algorithmic advisor. Our results show that human advisors were perceived as more responsible than algorithmic advisors and most importantly, that the perception of the advisor’s responsibility affected the preference for a human advisor over an algorithmic counterpart. Furthermore, we found that an experimental manipulation that impeded advice takers’ ability to offload responsibility affected the extent to which human, but not algorithmic, advisors were perceived as responsible. Together, our findings highlight the role of responsibility sharing in influencing algorithm aversion.

Conservatives endorse Fintech? Individual regulatory focus attenuates the algorithm aversion effects in automated wealth management

Algorithm aversion, which is a widely observed phenomenon, suggests that customers prefer human advisors over robo-advisors in automated wealth management. Studies have shown that prevention-focused (vs. promotion-focused) customers refuse high-tech products; therefore, algorithm aversion should be more pronounced among them. However, our findings challenge this notion. Drawing on the regulatory focus theory, through four studies, we discovered that (1) customers' mindset was relatively more prevention-focused when using robo-advisors and relatively more promotion-focused when using human advisors; (2) perceived flexibility mediated this effect; and (3) the algorithm aversion effects existed in automated wealth management, but vanished among prevention-focused customers because of the fit effects. Our findings identified a psychological boundary condition of algorithm aversion, provided a novel antecedent variable of regulatory focus, and provided managerial implications for robo-advisory firms’ market segmentation strategies.

Perceived corruption reduces algorithm aversion

AbstractScholarship on when and why humans are willing to rely on algorithms rather than other humans has made substantial progress in recent years, although virtually all such research is based on Western, educated, industrialized, rich, and democratic (WEIRD) research participants. This limits efforts to understand the cultural generalizability of attitudes toward algorithms. In this paper, I study algorithm aversion among participants from over 30 countries on all inhabited continents, thereby significantly increasing the diversity of this field's knowledge base. Furthermore, I leverage this diversity to test a theoretically derived prediction: that perceived corruption makes algorithmic decision‐making more appealing. I find that participants who are born or raised in countries with high levels of perceived corruption are much less averse to algorithmic decision‐making (or, in some studies, are not at all algorithm averse), relative to those from countries with low perceived corruption. Furthermore, experimentally varying corruption salience causes a decrease in algorithm aversion. I explore mechanisms and boundary conditions of these effects and discuss the implications in the context of algorithms that can both increase and decrease injustice.

What drives managers towards algorithm aversion and how to overcome it? Mitigating the impact of innovation resistance through technology readiness

Artificial intelligence has been increasingly used in many managerial decision-making contexts for its superior performance. However, many managers are still averse to using AI algorithms in decision-making. The benefits of algorithms cannot be realized optimally if managers are reluctant to use them. Although there are many studies investigating the factors that make an individual averse towards algorithms, studies examining such factors in the managerial decision-making context are scarce. This study, drawing on the innovation resistance theory and the technology readiness concept, proposes a structural model depicting relationships between managers' perceived barriers and algorithm aversion. The model also incorporates the role of managers' technology readiness in influencing these relationships. To test the model, data were collected from 179 managers of 31 banks/financial institutions in Bangladesh through an online survey and were analyzed using partial least squares structural equation modeling. The results indicate that barriers related to value, tradition, and image are significantly associated with algorithm aversion, whereas barriers related to usage and risk have a non-significant impact. Moderation analysis suggests that motivating factors of technology readiness mitigate the impact of value barriers on algorithm aversion.

No Thanks, Dear AI! Understanding the Effects of Disclosure and Deployment of Artificial Intelligence in Public Sector Recruitment

Abstract Applications based on artificial intelligence (AI) play an increasing role in the public sector and invoke political discussions. Research gaps exist regarding the disclosure effects—reactions to disclosure of the use of AI applications—and the deployment effect—efficiency gains in data savvy tasks. This study analyzes disclosure effects and explores the deployment of an AI application in a preregistered field experiment (n = 2,000) co-designed with a public organization in the context of employer-driven recruitment. The linear regression results show that disclosing the use of the AI application leads to significantly less interest in an offer among job candidates. The explorative analysis of the deployment of the AI application indicates that the person–job fit determined by the leaders can be predicted by the AI application. Based on the literature on algorithm aversion and digital discretion, this study provides a theoretical and empirical disentanglement of the disclosure effect and the deployment effect to inform future evaluations of AI applications in the public sector. It contributes to the understanding of how AI applications can shape public policy and management decisions, and discusses the potential benefits and downsides of disclosing and deploying AI applications in the public sector and in employer-driven recruitment.

AI Agents as Team Members: Effects on Satisfaction, Conflict, Trustworthiness, and Willingness to Work With

ABSTRACT Organizations are beginning to deploy artificial intelligence (AI) agents as members of virtual teams to help manage information, coordinate team processes, and perform simple tasks. How will team members perceive these AI team members and will they be willing to work with them? We conducted a 2 x 2 x 2 lab experiment that manipulated the type of team member (human or AI), their performance (high or low), and the performance of other team members (high or low). AI team members were perceived to have higher ability and integrity but lower benevolence, which led to no differences in trustworthiness or willingness to work with them. However, the presence of an AI team member resulted in lower process satisfaction. When the AI team member performed well, participants perceived less conflict compared to a human team member with the same performance, but there were no differences in perceived conflict when it performed poorly. There were no other interactions with performance, indicating that the AI team member was judged similarly to humans, irrespective of variations in performance; there was no evidence of algorithm aversion. Our research suggests that AI team members are likely to be accepted into teams, meaning that many old collaboration research questions may need to be reexamined to consider AI team members.

When Algorithms Err : Differential Impact of Early vs. Late Errors on Users’ Reliance on Algorithms

Errors are a natural part of predictive algorithms, but may discourage users from relying on algorithms. We conduct two experiments to demonstrate that reliance on a predictive algorithm following a substantial error is affected by (i) when the error occurs and (ii) how the algorithm is used in the decision-making process. We find that the impact of an error on reliance depends on whether the error occurs early (i.e., when users first start using the algorithm) or late (i.e., after users have used the algorithm for an extended period). While an early error results in substantial and persistent reliance reduction, a late error affects reliance only temporarily and to a lesser extent. However, when users have more control over how to use the algorithm’s predictions, error timing ceases to have a significant impact. Our work advances the understanding of algorithm aversion and informs the practical design of algorithmic decision-making systems.

The extent of algorithm aversion in decision-making situations with varying gravity.

Algorithms already carry out many tasks more reliably than human experts. Nevertheless, some subjects have an aversion towards algorithms. In some decision-making situations an error can have serious consequences, in others not. In the context of a framing experiment, we examine the connection between the consequences of a decision-making situation and the frequency of algorithm aversion. This shows that the more serious the consequences of a decision are, the more frequently algorithm aversion occurs. Particularly in the case of very important decisions, algorithm aversion thus leads to a reduction of the probability of success. This can be described as the tragedy of algorithm aversion.

Is algorithm aversion WEIRD? A cross-country comparison of individual-differences and algorithm aversion

Although algorithms offer superior performance over humans across many tasks, individuals often exhibit algorithm aversion, resisting algorithmic advice in favour of human recommendations. However, most algorithm aversion studies rely on American samples, potentially limiting the generalisability of the findings. Given the increasing adoption of algorithms globally, we explore if the impact of two crucial factors driving algorithm aversion, uniqueness neglect and familiarity, differ between culturally different countries. Drawing on the individualism-collectivism cultural dimension, we conducted two online studies comparing algorithm aversion between people in India and the United States in medical and financial services scenarios. While our results suggest that there is no difference in the degree of algorithm aversion between Indians and Americans at an aggregate level, we find important cross-cultural differences: Uniqueness neglect strengthens algorithm aversion for Americans more than Indians, while familiarity weakens algorithm aversion more for Indians than Americans. Thus, our results reveal generalisability issues within the algorithm aversion literature, as factors influencing algorithm aversion can be culturally dependent.

Human favoritism, not AI aversion: People’s perceptions (and bias) toward generative AI, human experts, and human–GAI collaboration in persuasive content generation

Abstract With the wide availability of large language models and generative AI, there are four primary paradigms for human–AI collaboration: human-only, AI-only (ChatGPT-4), augmented human (where a human makes the final decision with AI output as a reference), or augmented AI (where the AI makes the final decision with human output as a reference). In partnership with one of the world’s leading consulting firms, we enlisted professional content creators and ChatGPT-4 to create advertising content for products and persuasive content for campaigns following the aforementioned paradigms. First, we find that, contrary to the expectations of some of the existing algorithm aversion literature on conventional predictive AI, the content generated by generative AI and augmented AI is perceived as of higher quality than that produced by human experts and augmented human experts. Second, revealing the source of content production reduces—but does not reverse—the perceived quality gap between human- and AI-generated content. This bias in evaluation is predominantly driven by human favoritism rather than AI aversion: Knowing that the same content is created by a human expert increases its (reported) perceived quality, but knowing that AI is involved in the creation process does not affect its perceived quality. Further analysis suggests this bias is not due to a ‘quality prime’ as knowing the content they are about to evaluate comes from competent creators (e.g., industry professionals and state-of-the-art AI) without knowing exactly that the creator of each piece of content does not increase participants’ perceived quality.

An Experimental Study on Second-Order Trust in Algorithms: How Does Algorithm Aversion Spill Over During the Pandemic?

An Experimental Study on Second-Order Trust in Algorithms: How Does Algorithm Aversion Spill Over During the Pandemic?

Algorithm aversion in delegated investing

AbstractThe tendency of humans to shy away from using algorithms—even when algorithms observably outperform their human counterpart—has been referred to as algorithm aversion. We conduct an experiment with young adults to test for algorithm aversion in financial decision making. Participants acting as investors can tie their incentives to either a human fund manager or an investment algorithm. We find no sign of algorithm aversion: participants care about returns, but do not have strong preferences which financial intermediary obtains these returns. Contrary to what has been suggested, participants are neither quicker to lose confidence in the algorithm after seeing it err. However, we find that participants’ inability to separate skill and luck when evaluating intermediaries slows down their migration to the algorithm.