No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Increasing Reliance on Financial Advice with Avatars: The Effects of Competence and Complexity on Algorithm Aversion
Abstract
We find that avatar design can reduce algorithm aversion, which is the tendency of decision makers to ignore advice received from an algorithm after the algorithm makes an error. When the facial features of an avatar exhibit high levels of competence, algorithm aversion can be reduced relative to no avatar or a less competent-looking avatar. Humanizing the financial advice from an algorithm with an avatar that promotes the perception of competence effectively reduces algorithm aversion and can enhance reliance on the financial advice of robo-advisors.
... Yet, a higher level of anthropomorphism can also produce a negative uncanny valley effect (Kim et al., 2019). The underlying mechanism explaining the effect of anthropomorphism on consumer trust is equally unclear (Ganbold et al., 2022;Northey et al., 2022). Existent research suggests examining the effect of anthropomorphism through social presence under the lens of warmth-competence and expertise-trust conceptualizations (da Silva Lourenco et al., 2020;Fiske et al., 2007). ...
... Following a growing trend toward anthropomorphizing robo-advisors, Altrock et al. (2023) called for further investigation into anthropomorphism's characteristics and its effects on human-machine interaction and trust (Li and Suh, 2022;Blut et al., 2021). Using a humanized avatar can indeed reduce algorithm aversion (Ganbold et al., 2022) and increase investment intention (Baek and Kim, 2023), while naming a robo-advisor can lead to increased reliance on its recommendations (Hodge et al., 2021). Other anthropomorph cues such as speech can also increase trust toward the robo-advisor, positive attitude toward a financial services firm, and reliance on financial advice (Hildebrand and Bergner, 2021). ...
... When considering trust, measuring both affective and cognitive dimensions is important, as it is a multi-dimensional concept (Zhang et al., 2021). Hence, to explain consumer trust in a robo-advisory context, underlying psychological mechanisms need to be examined (Ganbold et al., 2022;Northey et al., 2022). While there is limited evidence of such psychological mechanisms, it is suggested to apply competence-warmth dimensions of social cognition and expertise-trust dimensions (da Silva Lourenco et al., 2020;Fiske et al., 2007). ...
Purpose
Financial services industry is increasingly showing interest in automated financial advisors, or robo-advisors, with the aim of democratizing access to financial advice and stimulating investment behavior among populations that were previously less active and less served. However, the extent to which consumers trust this technology influences the adoption of rob-advisors. The resemblance to a human, or anthropomorphism, can provide a sense of social presence and increase trust.
Design/methodology/approach
In this paper, we conduct an experiment ( N = 223) to test the effect of anthropomorphism (low vs medium vs high) and gender (male vs female) of the robo-advisor on social presence. This perception, in turn, enables consumers to evaluate personality characteristics of the robo-advisor, such as competence, warmth, and persuasiveness, all of which are related to trust in the robo-advisor. We separately conduct an experimental study ( N = 206) testing the effect of gender neutrality on consumer responses to robo-advisory anthropomorphism.
Findings
Our results show that consumers prefer human-alike robo-advisors over machinelike or humanoid robo-advisors. This preference is only observed for male robo-advisors and is explained by perceived competence and perceived persuasiveness. Furthermore, highlighting gender neutrality undermines the positive effect of robo-advisor anthropomorphism on trust.
Originality/value
We contribute to the body of knowledge on robo-advisor design by showing the effect of robot’s anthropomorphism and gender on consumer perceptions and trust. Consequently, we offer insightful recommendations to promote the adoption of robo-advisory services in the financial sector.
Robo-advisors, also known as robo-advisory services, significantly reshape customer service in financial advisory industries, transforming retail investor markets by substituting human financial advisory experts with artificial intelligence empowered services. However, existing literature remains scattered across disciplines, with theories on financial customer service predominantly focused on Internet banking, neglecting artificial intelligence empowered interactions. Thus, service providers need a framework for implementing robo-advisors in frontline service and researchers require an advanced agenda to stimulate future research. Through a systematic, interdisciplinary literature review based on Belanche et al.’s service robot framework, this article contextualizes service robot theories into financial advisory services, synthesizing knowledge on artificial intelligence empowered customer service. We contribute to literature on service robots by contextualizing, refining, and extending the original framework by Belanche et al. and by developing a research agenda with critical perspectives. Moreover, the study yields practical and theoretical insights into artificial intelligence empowered financial advisory services.
The increasing complexity of multinational supply networks has generated a new issue (supply chain uncertainty) for today's managers. This article surveys the existing literature on the topic of supply chain uncertainty and establishes the theoretical framework for future study in this area (in addition to supply chain risk). This literature study identifies fourteen potential causes of uncertainty, including both well-studied phenomena like the bullwhip effect and less well-known ones like parallel interaction. Ten solutions try to eliminate the core source of uncertainty, while eleven others aim to adapt to the existence of these unknowns in order to reduce their effects on manufacturing performance. The theory of manufacturing strategy and core concept of contingency and alignment establish a foundation of the supply chain uncertainty framework that is thus establishment using the research findings. More future empirical study is required to discover which uncertainty exists in distinct industrial settings, the effect of suitable sources and management strategies on productivity, and the intricate interaction between management techniques and diverse uncertainty sources.
In five experiments (N = 3,828), we investigate whether people prefer investment decisions to be made by human investment managers rather than by algorithms ("robos"). In all of the studies we investigate morally controversial companies, as it is plausible that a preference for humans as investment managers becomes exacerbated in areas where machines are less competent, such as morality. In Study 1, participants rated the permissibility of an algorithm to autonomously exclude morally controversial stocks from investment portfolios as lower than if a human fund manager did the same; this finding was not different if participants were informed that such exclusions might be financially disadvantageous for them. In Study 2, we show that this robo-investment aversion manifests itself both when considering investment in controversial and non-controversial industries. In Study 3, our findings show that robo-investment aversion is also present when algorithms are given the autonomy to increase investment in controversial stocks. In Studies 4 and 5, we investigate choices between actual humans and an algorithm. In Study 4 -which was incentivized-participants show no robo-investment aversion, but are significantly less likely to choose machines as investment managers for controversial stocks. In contrast, in Study 5 robo-investment aversion is present, but it is not different across controversial and non-controversial stocks. Overall, our findings show a considerable mean effect size for robo-investment aversion (d = -0.39 [-0.45, -0.32]). This suggests that algorithm aversion extends to the financial realm, supporting the existence of a barrier for the adoption of innovative financial technologies (FinTech).
Conversational agents (CAs) are software-based systems designed to interact with humans using natural language and have attracted considerable research interest in recent years. Following the Computers Are Social Actors paradigm, many studies have shown that humans react socially to CAs when they display social cues such as small talk, gender, age, gestures, or facial expressions. However, research on social cues for CAs is scattered across different fields, often using their specific terminology, which makes it challenging to identify, classify, and accumulate existing knowledge. To address this problem, we conducted a systematic literature review to identify an initial set of social cues of CAs from existing research. Building on classifications from interpersonal communication theory, we developed a taxonomy that classifies the identified social cues into four major categories (i.e., verbal, visual, auditory, invisible) and ten subcategories. Subsequently, we evaluated the mapping between the identified social cues and the categories using a card sorting approach in order to verify that the taxonomy is natural, simple, and parsimonious. Finally, we demonstrate the usefulness of the taxonomy by classifying a broader and more generic set of social cues of CAs from existing research and practice. Our main contribution is a comprehensive taxonomy of social cues for CAs. For researchers, the taxonomy helps to systematically classify research about social cues into one of the taxonomy's categories and corresponding subcategories. Therefore, it builds a bridge between different research fields and provides a starting point for interdisciplinary research and knowledge accumulation. For practitioners, the taxonomy provides a systematic overview of relevant categories of social cues in order to identify, implement, and test their effects in the design of a CA.
Competence impressions from faces affect important decisions, such as hiring and voting. Here, using data-driven computational models, we identify the components of the competence stereotype. Faces manipulated by a competence model varied in attractiveness (Experiment 1a). However, faces could be manipulated on perceived competence controlling for attractiveness (Experiments 1b); moreover, those faces perceived as more competent, but not attractive, were also perceived as more confident and masculine, suggesting a bias to perceive male faces as more competent than female faces (Experiment 2). Correspondingly, faces manipulated to appear competent, but not attractive, were more likely to be classified as male (Experiment 3). When masculinity cues that induced competence impressions were applied to real-life images, these cues were more effective on male faces (Experiment 4). These findings suggest that the main components of competence impressions are attractiveness, confidence, and masculinity, and reveal gender biases in how we form important impressions of others.
https://journals.sagepub.com/doi/pdf/10.1177/0956797618813092
Although evidence-based algorithms consistently outperform human forecasters, people often fail to use them after learning that they are imperfect, a phenomenon known as algorithm aversion. In this paper, we present three studies investigating how to reduce algorithm aversion. In incentivized forecasting tasks, participants chose between using their own forecasts or those of an algorithm that was built by experts. Participants were considerably more likely to choose to use an imperfect algorithm when they could modify its forecasts, and they performed better as a result. Notably, the preference for modifiable algorithms held even when participants were severely restricted in the modifications they could make (Studies 1–3). In fact, our results suggest that participants’ preference for modifiable algorithms was indicative of a desire for some control over the forecasting outcome, and not for a desire for greater control over the forecasting outcome, as participants’ preference for modifiable algorithms was relatively insensitive to the magnitude of the modifications they were able to make (Study 2). Additionally, we found that giving participants the freedom to modify an imperfect algorithm made them feel more satisfied with the forecasting process, more likely to believe that the algorithm was superior, and more likely to choose to use an algorithm to make subsequent forecasts (Study 3). This research suggests that one can reduce algorithm aversion by giving people some control—even a slight amount—over an imperfect algorithm’s forecast.
Data, as supplemental material, are available at https://doi.org/10.1287/mnsc.2016.2643 .
This paper was accepted by Yuval Rottenstreich, judgment and decision making.
A long-standing goal of artificial intelligence is an algorithm that learns, tabula rasa, superhuman proficiency in challenging domains. Recently, AlphaGo became the first program to defeat a world champion in the game of Go. The tree search in AlphaGo evaluated positions and selected moves using deep neural networks. These neural networks were trained by supervised learning from human expert moves, and by reinforcement learning from self-play. Here we introduce an algorithm based solely on reinforcement learning, without human data, guidance or domain knowledge beyond game rules. AlphaGo becomes its own teacher: a neural network is trained to predict AlphaGo's own move selections and also the winner of AlphaGo's games. This neural network improves the strength of the tree search, resulting in higher quality move selection and stronger self-play in the next iteration. Starting tabula rasa, our new program AlphaGo Zero achieved superhuman performance, winning 100-0 against the previously published, champion-defeating AlphaGo. © 2017 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
When we interact with others, we make inferences about their internal states (i.e., intentions, emotions) and use this information to understand and predict their behavior. Reasoning about the internal states of others is referred to as mentalizing, and presupposes that our social partners are believed to have a mind. Seeing mind in others increases trust, prosocial behaviors and feelings of social connection, and leads to improved joint performance. However, while human agents trigger mind perception by default, artificial agents are not automatically treated as intentional entities but need to be designed to do so. The panel addresses this issue by discussing how mind attribution to robots and other automated agents can be elicited by design, what the effects of mind perception are on attitudes and performance in human-robot and human-machine interaction and what behavioral and neuroscientific paradigms can be used to investigate these questions. Application areas covered include social robotics, automation, driver-vehicle interfaces, and others.
This article introduces the Special Issue on simple versus complex methods in forecasting. Simplicity in forecasting requires that (1) method, (2) representation of cumulative knowledge, (3) relationships in models, and (4) relationships among models,forecasts, and decisions are all sufficiently uncomplicated as to be easily understood by decision-makers. Our review of studies comparing simple and complex methods—including those in this special issue—found 97 comparisons in 32 papers. None of the papers provide a balance of evidence that complexity improves forecast accuracy.Complexity increases forecast error by 27 percent on average in the 25 papers with quantitative comparisons. The finding is consistent with prior research to identify valid forecasting methods: all 22 previously identified evidence-based forecasting procedures are simple. Nevertheless, complexity remains popular among researchers, forecasters, and clients. Some evidence suggests that the popularity of complexity may be due to incentives:(1) researchers are rewarded for publishing in highly ranked journals, which favor complexity; (2) forecasters can use complex methods to provide forecasts that support decision-makers’ plans; and (3) forecasters’ clients may be reassured by incomprehensibility. Clients who prefer accuracy should accept forecasts only from simple evidence-based procedures. They can rate the simplicity of forecasters’ procedures using the questionnaire at simple-forecasting.com.
The game of Go has long been viewed as the most challenging of classic games for artificial intelligence owing to its enormous search space and the difficulty of evaluating board positions and moves. Here we introduce a new approach to computer Go that uses ‘value networks’ to evaluate board positions and ‘policy networks’ to select moves. These deep neural networks are trained by a novel combination of supervised learning from human expert games, and reinforcement learning from games of self-play. Without any lookahead search, the neural networks play Go at the level of state-of-the-art Monte Carlo tree search programs that simulate thousands of random games of self-play. We also introduce a new search algorithm that combines Monte Carlo simulation with value and policy networks. Using this search algorithm, our program AlphaGo achieved a 99.8% winning rate against other Go programs, and defeated the human European Go champion by 5 games to 0. This is the first time that a computer program has defeated a human professional player in the full-sized game of Go, a feat previously thought to be at least a decade away.
Individuals are quite sensitive to others' appearance cues when forming social evaluations. Cues such as facial emotional resemblance are based on facial musculature and thus dynamic. Cues such as a face's structure are based on the underlying bone and are thus relatively static. The current research examines the distinction between these types of facial cues by investigating the consistency in social evaluations arising from dynamic versus static cues. Specifically, across four studies using real faces, digitally generated faces, and downstream behavioral decisions, we demonstrate that social evaluations based on dynamic cues, such as intentions, have greater variability across multiple presentations of the same identity than do social evaluations based on static cues, such as ability. Thus, although evaluations of intentions vary considerably across different instances of a target's face, evaluations of ability are relatively fixed. The findings highlight the role of facial cues' consistency in the stability of social evaluations.
© 2015 by the Society for Personality and Social Psychology, Inc.
This article introduces this JBR Special Issue on simple versus complex methods in forecasting. Simplicity in forecasting requires that (1) method, (2) representation of cumulative knowledge, (3) relationships in models, and (4) relationships among models, forecasts, and decisions are all sufficiently uncomplicated as to be easily understood by decision-makers. Our review of studies comparing simple and complex methods - including those in this special issue - found 97 comparisons in 32 papers. None of the papers provide a balance of evidence that complexity improves forecast accuracy. Complexity increases forecast error by 27 percent on average in the 25 papers with quantitative comparisons. The finding is consistent with prior research to identify valid forecasting methods: all 22 previously identified evidence-based forecasting procedures are simple. Nevertheless, complexity remains popular among researchers, forecasters, and clients. Some evidence suggests that the popularity of complexity may be due to incentives: (1) researchers are rewarded for publishing in highly ranked journals, which favor complexity; (2) forecasters can use complex methods to provide forecasts that support decision-makers’ plans; and (3) forecasters’ clients may be reassured by incomprehensibility. Clients who prefer accuracy should accept forecasts only from simple evidence-based procedures. They can rate the simplicity of forecasters’ procedures using the questionnaire at simple-forecasting.com.
This chapter outlines the two basic routes to persuasion. One route is based on the thoughtful consideration of arguments central to the issue, whereas the other is based on the affective associations or simple inferences tied to peripheral cues in the persuasion context. This chapter discusses a wide variety of variables that proved instrumental in affecting the elaboration likelihood, and thus the route to persuasion. One of the basic postulates of the Elaboration Likelihood Model—that variables may affect persuasion by increasing or decreasing scrutiny of message arguments—has been highly useful in accounting for the effects of a seemingly diverse list of variables. The reviewers of the attitude change literature have been disappointed with the many conflicting effects observed, even for ostensibly simple variables. The Elaboration Likelihood Model (ELM) attempts to place these many conflicting results and theories under one conceptual umbrella by specifying the major processes underlying persuasion and indicating the way many of the traditionally studied variables and theories relate to these basic processes. The ELM may prove useful in providing a guiding set of postulates from which to interpret previous work and in suggesting new hypotheses to be explored in future research.
Copyright © 1986 Academic Press Inc. Published by Elsevier Inc. All rights reserved.
Research shows that evidence-based algorithms more accurately predict the future than do human forecasters. Yet when forecasters are deciding whether to use a human forecaster or a statistical algorithm, they often choose the human forecaster. This phenomenon, which we call algorithm aversion, is costly, and it is important to understand its causes. We show that people are especially averse to algorithmic forecasters after seeing them perform, even when they see them outperform a human forecaster. This is because people more quickly lose confidence in algorithmic than human forecasters after seeing them make the same mistake. In 5 studies, participants either saw an algorithm make forecasts, a human make forecasts, both, or neither. They then decided whether to tie their incentives to the future predictions of the algorithm or the human. Participants who saw the algorithm perform were less confident in it, and less likely to choose it over an inferior human forecaster. This was true even among those who saw the algorithm outperform the human. (PsycINFO Database Record (c) 2014 APA, all rights reserved).
Clinical predictions made by mental health practitioners are compared with those using statistical approaches. Sixty-seven studies were identified from a comprehensive search of 56 years of research; 92 effect sizes were derived from these studies. The overall effect of clinical versus statistical prediction showed a somewhat greater accuracy for statistical methods. The most stringent sample of studies, from which 48 effect sizes were extracted, indicated a 13% increase in accuracy using statistical versus clinical methods. Several variables influenced this overall effect. Clinical and statistical prediction accuracy varied by type of prediction, the setting in which predictor data were gathered, the type of statistical formula used, and the amount of information available to the clinicians and the formulas. Recommendations are provided about when and under what conditions counseling psychologists might use statistical formulas as well as when they can rely on clinical methods. Implications for clinical judgment research and training are discussed.
People rapidly form impressions from facial appearance, and these impressions affect social decisions. We argue that data-driven, computational models are the best available tools for identifying the source of such impressions. Here we validate seven computational models of social judgments of faces: attractiveness, competence, dominance, extroversion, likability, threat, and trustworthiness. The models manipulate both face shape and reflectance (i.e., cues such as pigmentation and skin smoothness). We show that human judgments track the models' predictions (Experiment 1) and that the models differentiate between different judgments, though this differentiation is constrained by the similarity of the models (Experiment 2). We also make the validated stimuli available for academic research: seven databases containing 25 identities manipulated in the respective model to take on seven different dimension values, ranging from -3 SD to +3 SD (175 stimuli in each database). Finally, we show how the computational models can be used to control for shared variance of the models. For example, even for highly correlated dimensions (e.g., dominance and threat), we can identify cues specific to each dimension and, consequently, generate faces that vary only on these cues. (PsycINFO Database Record (c) 2013 APA, all rights reserved).
Is it possible to increase one's influence simply by behaving more confidently? Prior research presents two competing hypotheses: (1) the confidence heuristic holds that more confidence increases credibility, and (2) the calibration hypothesis asserts that overconfidence will backfire when others find out. Study 1 reveals that, consistent with the calibration hypothesis, while accurate advisors benefit from displaying confidence, confident but inaccurate advisors received low credibility ratings. However, Study 2 shows that when feedback on advisor accuracy is unavailable or costly, confident advisors hold sway regardless of accuracy. People also made less effort to determine the accuracy of confident advisors; interest in buying advisor performance data decreased as the advisor’s confidence went up. These results add to our understanding of how advisor confidence, accuracy, and calibration influence others.
Recent research has shown that rapid judgments about the personality traits of political candidates, based solely on their
appearance, can predict their electoral success. This suggests that voters rely heavily on appearances when choosing which
candidate to elect. Here we review this literature and examine the determinants of the relationship between appearance-based
trait inferences and voting. We also reanalyze previous data to show that facial competence is a highly robust and specific
predictor of political preferences. Finally, we introduce a computer model of face-based competence judgments, which we use
to derive some of the facial features associated with these judgments.
KeywordsFirst impressions-Voting-Political decision making-Face perception-Social cognition
Social psychologists conceptualize trust in a manner that differs from conceptualizations used by marketing researchers to date. Building from the social psychology literature, we posit that interpersonal trust in consumer-level service relationships has cognitive and affective dimensions. We examine the relative impact of service provider expertise, product performance, firm reputation, satisfaction, and similarity in influencing customer's perception of these dimensions of trust in a service provider. Using survey data from 349 customers of a firm of financial advisers in the United Kingdom, our results show that cognitive and affective dimensions of trust can be empirically distinguished and have both common and unique antecedents. The results also provide further clarification concerning the relationship between trust and sales effectiveness.
People tend to believe, and take advice from, informants who are highly confident. However, people use more than a mere “confidence heuristic.” We believe that confidence is influential because—in the absence of other information—people assume it is a valid cue to an informant’s likelihood of being correct. However, when people get evidence about an informant’s calibration (i.e., her confidence–accuracy relationship) they override reliance on confidence or accuracy alone. Two experiments in which participants choose between two opposing witnesses to a car accident show that neither confidence nor accuracy alone explains judgments of credibility; rather, whether a person is seen as credible ultimately depends on whether the person demonstrates good calibration. Credibility depends on whether sources were justified in believing what they believed.
This paper presents design research conducted as part of a larger project on human-robot interaction. The primary goal of this study was to come to an initial understanding of what features and dimensions of a humanoid robot's face most dramatically contribute to people's perception of its humanness. To answer this question we analyzed 48 robots and conducted surveys to measure people's perception of each robot's humanness. Through our research we found that the presence of certain features, the dimensions of the head, and the total number of facial features heavily influence the perception of humanness in robot heads. This paper presents our findings and initial guidelines for the design of humanoid robot heads.
Can human beings relate to computer or television programs in the same way they relate to other human beings? Based on numerous psychological studies, this book concludes that people not only can but do treat computers, televisions, and new media as real people and places. Studies demonstrate that people are "polite" to computers; that they treat computers with female voices differently than "male" ones; that large faces on a screen can invade our personal space; and that on-screen and real-life motion can provoke the same physical responses. Using everyday language to engage readers interested in psychology, communication, and computer technology, Reeves and Nass detail how this knowledge can help in designing a wide range of media.
The process of making judgments and decisions requires a method for combining data. To compare the accuracy of clinical and mechanical (formal, statistical) data-combination techniques, we performed a meta-analysis on studies of human health and behavior. On average, mechanical-prediction techniques were about 10% more accurate than clinical predictions. Depending on the specific analysis, mechanical prediction substantially outperformed clinical prediction in 33%-47% of studies examined. Although clinical predictions were often as accurate as mechanical predictions, in only a few studies (6%-16%) were they substantially more accurate. Superiority for mechanical-prediction techniques was consistent, regardless of the judgment task, type of judges, judges' amounts of experience, or the types of data being combined. Clinical predictions performed relatively less well when predictors included clinical interview data. These data indicate that mechanical predictions of human behaviors are equal or superior to clinical prediction methods for a wide range of circumstances.
We show that inferences of competence based solely on facial appearance predicted the outcomes of U.S. congressional elections
better than chance (e.g., 68.8% of the Senate races in 2004) and also were linearly related to the margin of victory. These
inferences were specific to competence and occurred within a 1-second exposure to the faces of the candidates. The findings
suggest that rapid, unreflective trait inferences can contribute to voting choices, which are widely assumed to be based primarily
on rational and deliberative considerations.
Anthropomorphism describes the tendency to imbue the real or imagined behavior of nonhuman agents with humanlike characteristics, motivations, intentions, or emotions. Although surprisingly common, anthropomorphism is not invariant. This article describes a theory to explain when people are likely to anthropomorphize and when they are not, focused on three psychological determinants--the accessibility and applicability of anthropocentric knowledge (elicited agent knowledge), the motivation to explain and understand the behavior of other agents (effectance motivation), and the desire for social contact and affiliation (sociality motivation). This theory predicts that people are more likely to anthropomorphize when anthropocentric knowledge is accessible and applicable, when motivated to be effective social agents, and when lacking a sense of social connection to other humans. These factors help to explain why anthropomorphism is so variable; organize diverse research; and offer testable predictions about dispositional, situational, developmental, and cultural influences on anthropomorphism. Discussion addresses extensions of this theory into the specific psychological processes underlying anthropomorphism, applications of this theory into robotics and human-computer interaction, and the insights offered by this theory into the inverse process of dehumanization.
Two experiments are described that examined operators' perceptions of decision aids.
Research has suggested certain biases against automation that influence human interaction with automation. We differentiated preconceived biases from post hoc biases and examined their effects on advice acceptance.
In Study 1 we examined operators' trust in and perceived reliability of humans versus automation of varying pedigree (expert vs. novice), based on written descriptions of these advisers prior to operators' interacting with these advisers. In Study 2 we examined participants' post hoc trust in, perceived reliability of, and dependence on these advisers after their objective experience of advisers' reliability (90% vs. 70%) in a luggage-screening task.
In Study 1 measures of perceived reliability indicated that automation was perceived as more reliable than humans across pedigrees. Measures of trust indicated that automated "novices" were trusted more than human "novices"; human "experts" were trusted more than automated "experts." In Study 2, perceived reliability varied as a function of pedigree, whereas subjective trust was always higher for automation than for humans. Advice acceptance from novice automation was always higher than from novice humans. However, when advisers were 70% reliable, errors generated by expert automation led to a drop in compliance/reliance on expert automation relative to expert humans.
Preconceived expectations of automation influence the use of these aids in actual tasks.
The results provide a reference point for deriving indices of "optimal" user interaction with decision aids and for developing frameworks of trust in decision support systems.
We examine the effect of humanizing (naming) robo‐advisors on investor judgments, which has taken on increased importance as robo‐advisors have become increasingly common and there is currently little SEC regulation governing key aspects of their use. In our first experiment, we predict and find that investors are more likely to rely on the investment recommendation of an unnamed robo‐advisor, whereas they are more likely to rely on the investment recommendation of a named human advisor. Theory suggests one reason that naming a robo‐advisor may have drawbacks pertains to the complexity of the task the robo‐advisor performs. We explore the importance of task complexity in our second experiment. We predict and find that investors are less likely to rely on a named robo‐advisor when the advisor is perceived to be performing a relatively complex task, consistent with our first experiment, and more likely to rely on a named robo‐advisor when the advisor is perceived to be performing a relatively simple task, consistent with prior research on human‐computer interactions. Our findings contribute to the literature examining how technology influences the acquisition and use of financial information and the general literature on human‐computer interactions. Our study also addresses a call by the SEC to learn more about robo‐advisors. Lastly, our study has practical implications for wealth management firms by demonstrating the potentially negative effects of making robo‐advisors more humanlike in an attempt to engage and attract users.
This article is protected by copyright. All rights reserved.
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.
Research suggests that consumers are averse to relying on algorithms to perform tasks that are typically done by humans, despite the fact that algorithms often perform better. The authors explore when and why this is true in a wide variety of domains. They find that algorithms are trusted and relied on less for tasks that seem subjective (vs. objective) in nature. However, they show that perceived task objectivity is malleable and that increasing a task’s perceived objectivity increases trust in and use of algorithms for that task. Consumers mistakenly believe that algorithms lack the abilities required to perform subjective tasks. Increasing algorithms’ perceived affective human-likeness is therefore effective at increasing the use of algorithms for subjective tasks. These findings are supported by the results of four online lab studies with over 1,400 participants and two online field studies with over 56,000 participants. The results provide insights into when and why consumers are likely to use algorithms and how marketers can increase their use when they outperform humans.
Even though computational algorithms often outperform human judgment, received wisdom suggests that people may be skeptical of relying on them (Dawes, 1979). Counter to this notion, results from six experiments show that lay people adhere more to advice when they think it comes from an algorithm than from a person. People showed this effect, what we call algorithm appreciation, when making numeric estimates about a visual stimulus (Experiment 1A) and forecasts about the popularity of songs and romantic attraction (Experiments 1B and 1C). Yet, researchers predicted the opposite result (Experiment 1D). Algorithm appreciation persisted when advice appeared jointly or separately (Experiment 2). However, algorithm appreciation waned when: people chose between an algorithm’s estimate and their own (versus an external advisor’s; Experiment 3) and they had expertise in forecasting (Experiment 4). Paradoxically, experienced professionals, who make forecasts on a regular basis, relied less on algorithmic advice than lay people did, which hurt their accuracy. These results shed light on the important question of when people rely on algorithmic advice over advice from people and have implications for the use of “big data” and algorithmic advice it generates.
Computer algorithms are increasingly being used to predict people's preferences and make recommendations. Although people frequently encounter these algorithms because they are cheap to scale, we do not know how they compare to human judgment. Here, we compare computer recommender systems to human recommenders in a domain that affords humans many advantages: predicting which jokes people will find funny. We find that recommender systems outperform humans, whether strangers, friends, or family. Yet people are averse to relying on these recommender systems. This aversion partly stems from the fact that people believe the human recommendation process is easier to understand. It is not enough for recommender systems to be accurate, they must also be understood.
OBJECTIVE
We measured neurophysiologic responses and task performance while participants solved mazes after choosing whether to adopt an imperfect helper algorithm.
BACKGROUND
Every day we must decide whether to trust or distrust algorithms. Will an algorithm improve our performance on a task? What if we trust it too much?
METHOD
Participants had to pay to use the algorithm and were aware that it offered imperfect help. We varied the information about the algorithm to assess the factors that affected adoption while measuring participants’ peripheral neurophysiology.
RESULTS
We found that information about previous adoption by others had a larger effect on adoption and resulted in lower cognitive load than did information about algorithm accuracy. The neurophysiologic measurement showed that algorithm adoption without any information resulted in low cognitive engagement during the task and impaired task performance. Conversely, algorithm use after information about others’ use improved engagement and performance.
CONCLUSION
By objectively measuring cognitive load and task performance, we identified how to increase algorithm adoption while sustaining high performance by human operators.
APPLICATION
Algorithm adoption can be increased by sharing previous use information and performance improved by providing a reason to monitor the algorithm.
Precis
We collected neurophysiologic data while varying information about an algorithm that assisted participants in solving a timed and incentivized maze and found that information about prior use by others more effectively influenced adoption, reduced cognitive load, and improved performance compared to algorithm accuracy information.
Forecasting advice from human advisors is often utilized more than advice from automation. There is little understanding of why “algorithm aversion” occurs, or specific conditions that may exaggerate it. This paper first reviews literature from two fields—interpersonal advice and human–automation trust—that can inform our understanding of the underlying causes of the phenomenon. Then, an experiment is conducted to search for these underlying causes. We do not replicate the finding that human advice is generally utilized more than automated advice. However, after receiving bad advice, utilization of automated advice decreased significantly more than advice from humans. We also find that decision makers describe themselves as having much more in common with human than automated advisors despite there being no interpersonal relationship in our study. Results are discussed in relation to other findings from the forecasting and human–automation trust fields and provide a new perspective on what causes and exaggerates algorithm aversion.
Research shows that evidence-based algorithms more accurately predict the future than do human forecasters. Yet, when forecasters are deciding whether to use a human forecaster or a statistical algorithm, they often choose the human forecaster. This phenomenon, which we call algorithm aversion, is costly, and it is important to understand its causes. We show that people are especially averse to algorithmic forecasters after seeing them perform, even when they see them outperform a human forecaster. This is because people more quickly lose confidence in algorithmic than human forecasters after seeing them make the same mistake. In five studies, participants either saw an algorithm make forecasts, a human make forecasts, both, or neither. They then decided whether to tie their incentives to the future predictions of the algorithm or the human. Participants who saw the algorithm perform were less confident in it, and less likely to choose it over an inferior human forecaster. This was true even among those who saw the algorithm outperform the human.
Seasonalities and empirical regularities on financial markets have been well documented in the literature for three decades. While one should suppose that documenting an arbitrage opportunity makes it vanish there are several regularities that have persisted over the years. These include, for example, upward biases at the turn-of-the-month, during exchange holidays and the pre-FOMC announcement drift. Trading regularities is already in and of itself an interesting strategy. However, unfiltered trading leads to potential large drawdowns. In the paper we present a decision support algorithm which uses the powerful ideas of reinforcement learning in order to improve the economic benefits of the basic seasonality strategy. We document the performance on two major stock indices.
Virtually all discussions and applications of statistical mediation analysis have been based on the condition that the independent variable is dichotomous or continuous, even though investigators frequently are interested in testing mediation hypotheses involving a multicategorical independent variable (such as two or more experimental conditions relative to a control group). We provide a tutorial illustrating an approach to estimation of and inference about direct, indirect, and total effects in statistical mediation analysis with a multicategorical independent variable. The approach is mathematically equivalent to analysis of (co)variance and reproduces the observed and adjusted group means while also generating effects having simple interpretations. Supplementary material available online includes extensions to this approach and Mplus, SPSS, and SAS code that implements it.
Although democracies should ideally elect leaders based on their abilities, voters are often biased by seemingly unrelated factors, such as a candidate's appearance. Prior work examining the relations between election outcomes and appearance has primarily focused on a restricted range of the top candidates, examined in pairwise comparisons. In the present study, we tested whether the predictive ability of ratings based on facial appearance would extend to a wider range of candidates. Specifically, we examined whether individuals in the US could predict outcomes in the 2011 Bulgarian presidential elections by evaluating the facial appearance of 18 candidates. The large number of candidates naturally running for the high level office allowed us to accurately test the strength of the relationship between judgments of facial appearance and election outcomes across a broad range of faces. We found that a strong correlation between ratings of facial competence and election outcomes persisted across the full range of candidates, and that US participants' hypothetical choices paralleled actual Bulgarian election outcomes. We demonstrated that competence ratings were more effective at predicting election outcomes than judgments on a variety of other characteristics deemed important by Bulgarian voters as well as ratings of attractiveness. Furthermore, judgments of competence largely drove the correlation between hypothetical and actual votes.
The trust placed in diagnostic aids by the human operator is a critical psychological factor that influences operator reliance on automation. Studies examining the nature of human interaction with automation have revealed that users have a propensity to apply norms of human–human inter-personal interaction to their interaction with ‘intelligent machines’. Nevertheless, there exist subtle differences in the manner in which humans perceive and react to automated aids compared to human team-mates. In the present paper, the concept of trust in human–automation dyads is compared and contrasted with that of human–human dyads. A theoretical framework that synthesizes and describes the process of trust development in humans vs automated aids is proposed and implications for the design of decision aids are provided. Potential implications of this research include the improved design of decision support systems by incorporating features into automated aids that elicit operator responses mirroring responses in human–human inter-personal interaction. Such interventions will likely facilitate better quantification and prediction of human responses to automation, while improving the quality of human interaction with non-human team-mates.
Forecasting support systems (FSSs) have little value if users distrust the information and advice that they offer. Two experiments were used to investigate: (i) factors that influence the levels of users’ stated trust in advice provided by an FSS, when this advice is provided in the form of interval forecasts, (ii) the extent to which stated trust is associated with users’ modifications of the provided forecasts, and (iii) the consequences of these modifications for the calibration of the interval forecasts. Stated trust was influenced by the levels of noise in time series and whether a trend was present but was unaffected by the presence or absence of point forecasts. It was also higher when the intervals were framed as ‘best-case/worst-case’ forecasts and when the FSS provided explanations. Absence of trust was associated with a tendency to narrow the provided prediction intervals, which reduced their calibration.
Following Langer (1992), this article reviews a series of experimental studies that demonstrate that individuals mindlessly apply social rules and expectations to computers. The first set of studies illustrates how individuals overuse human social categories, applying gender stereotypes to computers and ethnically identifying with computer agents. The second set demonstrates thatpeople exhibit overlearned social behaviors such as politeness and reciprocity toward computers. In the third set of studies, premature cognitive commitments are demonstrated: A specialist television set is perceived as providing better content than a generalist television set. A final series of studies demonstrates the depth of social responses with respect to computer ‘personality.’ Alternative explanations for these findings, such asanthropomorphism and intentional social responses, cannot explain the results. We conclude with an agenda for future research.
Given a data set about an individual or a group (e.g., interviewer ratings, life history or demographic facts, test results, self-descriptions), there are two modes of data combination for a predictive or diagnostic purpose. The clinical method relies on human judgment that is based on informal contemplation and, sometimes, discussion with others (e.g., case conferences). The mechanical method involves a formal, algorithmic, objective procedure (e.g., equation) to reach the decision. Empirical comparisons of the accuracy of the two methods (136 studies over a wide range of predictands) show that the mechanical method is almost invariably equal to or superior to the clinical method: Common antiactuarial arguments are rebutted, possible causes of widespread resistance to the comparative research are offered, and policy implications of the statistical method's superiority are discussed.
This study provides a theoretical framework and experimental evidence on how managers' disclosure decisions affect their credibility with investors. I find that in the short-term, more forthcoming disclosure has a positive effect on management's reporting credibility, especially when management is forthcoming about negative news. However, these short-term credibility effects do not persist over time. In the long-term, managers who report positive earnings news are rated as having higher reporting credibility than managers who report negative earnings news, regardless of their previous disclosure decisions.
The face is our primary source of visual information for identifying people and reading their emotional and mental states. With the exception of prosopagnosics (who are unable to recognize faces) and those suffering from such disorders of social cognition as autism, people are extremely adept at these two tasks. However, our cognitive powers in this regard come at the price of reading too much into the human face. The face is often treated as a window into a person's true nature. Given the agreement in social perception of faces, this paper discusses that it should be possible to model this perception.
Testing the assumption that more anthropomorphic (human-like) computer representations elicit more social responses from people, a between-participants experiment (N = 168) manipulated 12 computer agents to represent four levels of anthropomorphism: low, medium, high, and real human images. Social responses were assessed with users’ social judgment and homophily perception of the agents, conformity in a choice dilemma task, and competency and trustworthiness ratings of the agents. Linear polynomial trend analyses revealed significant linear trends for almost all the measures. As the agent became more anthropomorphic to being human, it received more social responses from users.
Advancements in computer technology have allowed the development of human-appearing and -behaving virtual agents. This study examined if increased richness and anthropomorphism in interface design lead to computers being more influential during a decision-making task with a human partner. In addition, user experiences of the communication format, communication process, and the task partner were evaluated for their association with various features of virtual agents. Study participants completed the Desert Survival Problem (DSP) and were then randomly assigned to one of five different computer partners or to a human partner (who was a study confederate). Participants discussed each of the items in the DSP with their partners and were then asked to complete the DSP again. Results showed that computers were more influential than human partners but that the latter were rated more positively on social dimensions of communication than the former. Exploratory analysis of user assessments revealed that some features of human–computer interaction (e.g. utility and feeling understood) were associated with increases in anthropomorphic features of the interface. Discussion focuses on the relation between user perceptions, design features, and task outcomes.
The claim that computer personalities can be human personalities was tested by demonstrating that (1) computer personalities can be easily created using a minimal set of cues, and (2) that people will respond to these personalities in the same way they would respond to similar human personalities. The present study focused on the "similarity-attraction hypothesis," which predicts that people will prefer to interact with others who are similar in personality. In a 2 × 2, balanced, between-subjects experiment (n = 48), dominant and submissive subjects were randomly matched with a computer that was endowed with the properties associated with dominance or submissiveness. Subjects recognized the computer's personality type, distinct from friendliness and competence. In addition, subjects not only preferred the similar computer, but they were more satisfied with the interaction. The findings demonstrate that personality does not require richly defined agents, sophisticated pictorial representations, nautral language processing, or artificial intelligence. Rather, even the most superficial manipulations are sufficient to exhibit personality, with powerful effects.
Predictive models that incorporate a functional relationship of program error measures with software complexity metrics and metrics based on factor analysis of empirical data are developed. Specific techniques for assessing regression models are presented for analyzing these models. Within the framework of regression analysis, the authors examine two separate means of exploring the connection between complexity and errors. First, the regression models are formed from the raw complexity metrics. Essentially, these models confirm a known relationship between program lines of code and program errors. The second methodology involves the regression of complexity factor measures and measures of errors. These complexity factors are orthogonal measures of complexity from an underlying complexity domain model. From this more global perspective, it is believed that there is a relationship between program errors and complexity domains of program structure and size (volume). Further, the strength of this relationship suggests that predictive models are indeed possible for the determination of program errors from these orthogonal complexity domains
In theories and studies of persuasion, people's personal knowledge about persuasion agents' goals and tactics, and about how to skillfully cope with these, has been ignored. We present a model of how people develop and use persuasion knowledge to cope with persuasion attempts. We discuss what the model implies about how consumers use marketers' advertising and selling attempts to refine their product attitudes and attitudes toward the marketers themselves. We also explain how this model relates to prior research on consumer behavior and persuasion and what it suggests about the future conduct of consumer research. Copyright 1994 by the University of Chicago.
Review and reflection indicate that no more than 5% of what was written in the 1954 book entitled, Clinical Versus Statistical Prediction (Meehl, 1984), needs to be retracted 30 years later. If anything, these retractions would result in the book's being more actuarial than it was. Seven factors appear to account for the failure of mental health professionals to apply in practice the strong and clearly supported empirical generalizations demonstrating the superiority of actuarial over clinical prediction.
People often draw trait inferences from the facial appearance of other people. We investigated the minimal conditions under which people make such inferences. In five experiments, each focusing on a specific trait judgment, we manipulated the exposure time of unfamiliar faces. Judgments made after a 100-ms exposure correlated highly with judgments made in the absence of time constraints, suggesting that this exposure time was sufficient for participants to form an impression. In fact, for all judgments-attractiveness, likeability, trustworthiness, competence, and aggressiveness-increased exposure time did not significantly increase the correlations. When exposure time increased from 100 to 500 ms, participants' judgments became more negative, response times for judgments decreased, and confidence in judgments increased. When exposure time increased from 500 to 1,000 ms, trait judgments and response times did not change significantly (with one exception), but confidence increased for some of the judgments; this result suggests that additional time may simply boost confidence in judgments. However, increased exposure time led to more differentiated person impressions.