Algorithms have rapidly developed to exert an unprecedented effect on society worldwide and to transform how people live, for example, algorithms are helpful in predicting the weather and pandemics (Jia et al., 2020; Sawaitul et al., 2012). Yet they can also be problematic, as they have been found to inspire negative behavior, such as impulse shopping or social media addiction, by forecasting customers’ hidden needs and preferences (Anderson & Wood, 2021).

Business owners and researchers have also begun to use algorithms in person judgments. For example, consumer lending teams have adopted algorithms to decide whether to accept or reject loan applicants (Faggella, 2020). Further, according to a growing number of press articles, algorithms are reshaping employee recruitment by predicting candidates’ suitability for a job or their potential organizational loyalty (Koenig, 2018).

Studies have further shown that algorithms often outperform people in making predictive judgments (Highhouse, 2008; Kuncel et al., 2013). For instance, Grove et al. (2000) noted in a meta-analysis that algorithms can significantly increase accurate predictions of people’s health conditions and diagnoses. However, despite these advantages, people remain reluctant to rely on algorithms when making decisions (Diab et al., 2011; Vrieze & Grove, 2009). For example, employers tend to overestimate the usefulness of human judgment in worker selection (Highhouse, 2008). Similarly, doctors often prefer to make judgments based on their intuition and experience rather than on algorithms (Keeffe et al., 2005). These phenomena have been labeled algorithm aversion (Dietvorst et al., 2018).

Researchers have proposed several reasons for algorithm aversion, including that algorithms do not learn and update (Dawes, 1979), cannot address qualitative information (Grove & Meehl, 1996), and are unsuitable for subjective decision making (Castelo et al., 2019). In this study I focused on person-judgment tasks (e.g., student admission or consumer lending decisions) and explained the occurrence of algorithm aversion from the perspective of decision makers’ information processing style in candidate evaluation.

Literature Review and Development of Hypotheses

Information Processing in Person Judgment

There are two types of information on which people can depend when making person judgments (Yan & Sengupta, 2013; Zukier & Pepitone, 1984): Case-specific information, which refers to concrete and individuating characteristics of the target person, and general propositions, which are common laws and attributes that hold true for most cases in a certain category. Numerous researchers have found that when people make person judgments, they often rely on specific information about each person rather than on broader categorical information (Lyon & Slovic, 1976; Zuckerman, 1978). This tendency manifests because general propositions (vs. case-specific information) are perceived as less concrete and vivid; thus, they are thought to offer limited informational value when judgments are made (Nisbett & Borgida, 1975; Tversky & Kahneman, 1980). In a seminal experiment by Kahneman and Tversky (1973), participants judged the likelihood that a person was an engineer or lawyer based on case-specific information, which consisted of a set of detailed descriptions about the person’s behavior and other characteristics, and general propositions, which included the proportion of lawyers and engineers in the entire population. The study revealed that when participants judged a specific person’s category membership, they placed more weight on individuating and case-specific information, and ignored general propositions. Drawing on these findings, I further examined whether people’s tendency to overuse case-specific information conflicts with their lay beliefs about algorithms, leading to algorithm aversion in person-judgment tasks.

Lay Beliefs About Algorithms

Algorithmic judgment is based on mathematical models, whereas human judgment relies on personal expertise and experience (Dawes et al., 1989; Diab et al., 2011; Eastwood et al., 2012). I propose that this causes people to assume that algorithms perform well for determining general propositions but poorly for considering case-specific information. On one hand, people presume that algorithms can incorporate large sample sizes that people may have difficulty managing, that is, algorithms (vs. human experts) return more accurate results when judgments are made on an aggregate level, such as an examination of patterns among large datasets (Grove & Meehl, 1996; Meehl, 1954). On the other hand, people may believe that algorithms (vs. human experts) are preprogrammed and cannot account for all case-specific information that facilitates person judgments (e.g., situational factors and detailed descriptions about human behavior; Dawes, 1979). Therefore, I proposed the following hypothesis:
Hypothesis 1: Decision makers will be less likely to rely on algorithms (vs. human experts) when making person judgments.

Algorithms have been shown to outperform people in many person judgment tasks, so it is important to lessen aversion to algorithmic judgments and thereby maximize the potential benefits of this technological advance. People tend to underestimate the informational value of general propositions in person judgments (Kahneman & Tversky, 1973; Lyon & Slovic, 1976; Zuckerman, 1978). Therefore, I proposed the following hypothesis:
Hypothesis 2: Decision makers will be more likely to adopt algorithms in person judgment tasks when they attach greater importance to general propositions (vs. case-specific information).

I tested the hypotheses in one pilot study and three main studies. In the pilot study I aimed to show if algorithm aversion was significant in person-judgment tasks, and I explored via content analysis the mechanism underlying this phenomenon. To moderate people’s algorithm aversion in person-judgment tasks, I measured individual differences in decision makers’ information processing style in Study 1, decreased the perceived importance of case-specific information in Study 2A, and increased the perceived importance of general propositions in Study 2B. All studies were conducted in accordance with the Declaration of Helsinki for research involving human subjects, and informed consent was obtained from all participants.

Pilot Study

Method

Participants
A sample of 252 U.S. participants (106 women, 146 men; M_age = 37.60 years, SD = 11.26) recruited from Amazon’s Mechanical Turk (MTurk) completed a brief survey in exchange for USD 0.50.

Procedure
Participants were told that the survey was intended to identify people’s preferences when making decisions. They were informed that many organizations and institutions have begun using human experts or algorithms to make decisions related to tasks such as employee selection, student admissions, and consumer lending. Next, they read that human experts usually make decisions based on years of experience and expertise, whereas algorithms typically rely on big data and mathematical modeling. Participants then responded to the question “Which method do you think is better?” on a 7-point Likert scale (1 = human expert, 7 = algorithm). They were then asked to provide the rationale for their responses and to report their age and gender. Finally, they were debriefed, thanked, and compensated for their participation.

Results and Discussion

A one-sample t test comparison of participants’ ratings (M = 3.36, SD = 1.85) to the midpoint of the scale (i.e., 4 points) revealed that participants were generally averse to the use of algorithms in person-judgment tasks, t(251) = 5.51, p < .001, 95% confidence interval (CI) of the mean difference [−0.87, −0.41], d = 0.35. In particular, 139 participants (55.2%) had a score lower than 4 (i.e., they preferred human experts to algorithms), 72 (28.6%) had a score higher than 4 (i.e., they preferred algorithms to human experts), and 41 (16.3%) had a neutral attitude. These findings confirm the trend toward algorithm aversion in person-judgment tasks.

Next, I followed the recommended steps for conducting a content analysis to explore why participants experienced algorithm aversion (Postmes et al., 2001; Tinsley & Brett, 2001). Two independent coders who were blind to my hypotheses analyzed the justifications provided by the 139 participants who objected to algorithms (i.e., had a score lower than 4). The coders first independently judged whether each participant mentioned a concern about the limited capacity of algorithms to address case-specific information (e.g., “Algorithms cannot understand nuances or exceptions”), and then reconvened to resolve any inconsistencies. When providing their rationale, 67 participants (48.2%) who objected to algorithms expressed such concerns. These findings supported Hypothesis 1, suggesting that algorithm aversion in person-judgment tasks occurred largely because of the conflict between people’s propensities in these tasks and their lay beliefs about algorithms. On the basis of these results, I designed the following studies to moderate this effect by varying the importance of case-specific information and general propositions in person-judgment tasks.

Study 1

Method

Participants
Two surveys were completed by 301 participants (149 women, 152 men; M_age = 38.17 years, SD = 11.09) on MTurk in exchange for USD 0.50. I adopted a correlational research design and predicted that people would tend to prefer algorithms (vs. human experts) when placing more weight on general propositions (vs. case-specific information) in person-judgment tasks.

Procedure
Participants were told that the purpose of this study was to understand their judgment and decision making. In the first of two surveys I stated that I was interested in learning about their decision-making style and asked them to read descriptions adapted from Zukier and Pepitone (1984) of the two primary information-processing styles: Style A and Style B. Style A decision makers care about propositions that hold generally, whereas Style B decision makers devote more effort to understanding individual cases (see Appendix A). Participants then responded to the question “In most cases, which decision-making style fits you better?” on a 7-point Likert scale (1 = Style A, 7 = Style B).

Next, participants proceeded to the additional survey pages. They were told that in many organizations, people have begun relying on human experts or algorithms to make decisions. Participants also learned that algorithms make decisions based on large datasets and mathematical models, whereas human experts make decisions based on their years of experience and expertise. I presented participants with three questions about three types of person judgment tasks (i.e., hiring decisions, student admissions, and consumer lending) and asked them which decision-making method they would prefer in each scenario. Participants responded to the question “Which method do you think is better?” on a 7-point Likert scale (1 = human expert, 7 = algorithm). After providing their age and gender, participants were debriefed, thanked, and compensated.

Results and Discussion

I averaged participants’ relative preferences across the three person-judgment scenarios as the dependent variable (α = .73). Next, I regressed participants’ relative preferences on their information-processing style (reverse-coded). Results show there was a positive association between participants’ algorithm adoption (M = 3.27, SD = 1.42) and their tendency to use general propositions for judgments (M = 3.05, SD = 1.62), β = .19, t(299) = 3.28, p = .001, 95% CI [0.07, 0.30]. Correlations among the variables are shown in Table 1. When people tended to evaluate a person based more on general propositions than on case-specific information, they expressed a greater preference for algorithms in a series of person-judgment tasks. Thus, Hypothesis 2 was supported.

The Study 1 results show there was a correlation between participants’ individual differences in information-processing styles and their algorithm adoption in person-judgment tasks. In the subsequent studies I tested the moderating effect of information-processing style on algorithm adoption. In particular, I aimed to demonstrate that algorithm adoption would increase when the importance of reliance on case-specific information was downplayed, and the importance of considering general propositions was salient.

Study 2A

Method

Participants
This study was completed by 503 participants (252 women, 251 men; M_age = 37.61 years, SD = 12.29) on MTurk in exchange for USD 0.50. I adopted a one-factor (importance of case-specific information: downplayed vs. control) between-subjects design.

Procedure
Participants were told that the aim in this study was to explain people’s preferences when making decisions. They were instructed to imagine that they were the director of a postgraduate program and were considering admitting one student. Next, they learned that they and their colleagues had received résumés from 100 candidates. Participants were then presented with information depending on the condition assigned to them in the study. Respondents in the importance-downplayed condition were told they would admit the student with the best academic performance during undergraduate studies. Conversely, respondents in the control condition were asked to admit the student with the greatest potential to be successful during the postgraduate program (see Appendix B).

I chose these judgment criteria (Tormala et al., 2012) because achievement-based judgment relies on a student’s cumulative records, whereas potential-based judgment requires greater consideration of case-specific information (e.g., nuances that may undermine or facilitate a student’s future performance). To confirm this assumption, I ran an independent pretest in which 100 participants (51 women, 49 men; M_age = 39.52 years, SD = 10.79) on MTurk responded to the item “Relative to assessment of a person’s achievement in the past years, assessment of a person’s potential requires more case-by-case analysis” on a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree). A one-sample t test comparing participants’ ratings (M = 5.64, SD = 1.06) to the scale midpoint (i.e., 4 points) indicated that case-specific information was less important in the judgment based on achievement (vs. potential), t(99) = 15.49, p < .001, d = 1.55, 95% CI of mean difference [1.43, 1.85].

All participants in the main study were then informed that they could rely on either human experts or algorithms to make their decisions. The human expert and algorithm descriptions were identical to those in Study 1. Next, I measured participants’ relative preferences with the question “In this scenario, which method do you think is relatively better?” on a 7-point Likert scale (1 = human expert, 7 = algorithm). After providing their age and gender, participants were debriefed, thanked, and compensated.

Results and Discussion

Results of an independent-samples t test show that participants were more likely to adopt algorithms when making decisions based on achievement (M = 3.13, SD = 1.90) than on potential, M = 2.76, SD = 1.78, t(501) = 2.25, p = .025, d = 0.20, 95% CI of mean difference [0.05, 0.69]. This pattern lends additional support to Hypothesis 2. In particular, when the judgment task required less case-specific information, participants were more likely to adopt algorithms (see Figure 1).

Participants in each condition in this study were presented with distinct judgment tasks (i.e., based on potential vs. achievement). As I was curious about how to increase people’s adoption of algorithms when facing the same task, I aimed to increase algorithm adoption in Study 2B by manipulating people’s perceived importance of general propositions.

Study 2B

Method

Participants
This study was completed by 186 participants (89 women, 97 men; M_age = 35.20 years, SD = 11.60) on MTurk in exchange for USD 0.50. I adopted a one-factor (importance of general propositions: salient vs. control) between-subjects design.

Procedure
Participants were told that people’s decision-making preferences would be explored in this study. They read a scenario in which they imagined themselves as the director of a consumer lending team in a financial institution. Next, they learned that their team often received loan applications in which applicants provided information about themselves (e.g., income, education, and credit history) and their loan purposes. Finally, they were told that their task was to review these applications and make accept or reject decisions based on their estimation of applicants’ ability to repay the loan.

Next, I manipulated the importance of general propositions drawing on logic in previous studies (i.e., Ajzen, 1977; Zukier & Pepitone, 1984), which showed that highlighting the usefulness of general propositions to make an accurate judgment can enhance people’s reliance on them. In particular, in the importance-salience condition, participants were informed that although loan repayment behavior can vary situationally, they could still apply basic rules or laws for prediction purposes. I told participants that researchers have shown that applicants’ income, education, and credit history are the most important factors in loan repayment. In the control condition, participants did not receive this information (see Appendix C).

Participants were then told they could rely on either human experts or algorithms to decide whether to accept or reject applicants. Instructions about algorithms and human experts were identical to those in Studies 1 and 2A. I measured participants’ relative preferences for algorithms and human experts with the question “In this scenario, which method do you think is relatively better?” on a 7-point Likert scale (1 = human expert, 7 = algorithm). Participants were debriefed, thanked, and compensated, after reporting their age and gender.

Results and Discussion

Results of an independent-samples t test show that participants were more likely to adopt algorithms when the importance of general propositions was salient (M = 3.92, SD = 1.86) than when they were not (M = 3.29, SD = 1.82), t(184) = 2.33, p = .021, d = 0.34, 95% CI of mean difference [0.10, 1.16]. This trend again supported Hypothesis 2. In brief, when participants were informed of the importance of general propositions for accurate person judgments, they were more likely to adopt algorithms (see Figure 2). These findings provide further evidence for the mechanism underlying algorithm aversion and offer a way to boost people’s algorithm adoption in person-judgment tasks.

General Discussion

In four studies I demonstrated that decision makers tend to rely on human experts rather than on algorithms to make person judgments, supporting Hypothesis 1. The pilot study results show that people believed algorithms could not address case-specific information, which conflicted with their perception that such information was more important than general propositions in person-judgment tasks. Study 1 demonstrated through a correlational design that algorithm adoption increased if decision makers used general propositions in person-judgment tasks, supporting Hypothesis 2. I then experimentally varied the importance of case-specific information and general propositions, and the results of Study 2A and 2B further supported Hypothesis 2.

My results contribute to the literature by providing an information-based perspective to elucidate algorithm aversion, which is related to, but differs from, several alternative explanations. First, Castelo et al. (2019) found that people prefer algorithms for tasks requiring calculation (i.e., quantitative) versus subjective feelings (i.e., qualitative). However, as Yan and Sengupta (2013) noted, case-specific information/general propositions—the foci of my research—are not necessarily qualitative (subjective)/quantitative (objective). In addition, Longoni et al. (2019) pointed out that people mostly prefer a person (vs. a computer) for skin cancer diagnosis because they consider themselves unique. Although I concede that a degree of overlap exists between the unique and case-specific notions, I extended Longoni et al.’s (2019) work by adopting a different angle: Longoni et al. examined the role of perceived self-uniqueness (i.e., differences between oneself and one’s peers) in consumers’ attitudes towards algorithmic judgments about themselves. By contrast, I tested whether decision makers tend to rely on algorithms when making judgments about others and I built my hypotheses around the values of two types of information used in person judgments. Future researchers could further parse the differences in these explanations.

My results also offer practical implications by identifying ways to boost algorithm adoption. Researchers have sought to increase consumers’ algorithm adoption owing to algorithms’ superior performance over human judgment. In addition, researchers have found that people tend to adopt algorithms when their perceived control over algorithmic judgments is enhanced (Dietvorst et al., 2018), and judgment tasks rely on calculations rather than subjective feelings (Castelo et al., 2019). Differing from earlier results, I found that participants’ algorithm aversion could be attenuated when individuals were directed to attach relatively more (vs. less) importance to general propositions (vs. case-specific information) in person-judgment tasks. My findings provide a psychology-based solution to encourage algorithm adoption. Although algorithm adoption can be improved by continued technological advances, such development involves considerable investment in time and money. In contrast, the use of influencing psychological factors by enhancing the importance of general propositions (Case et al., 1999) should require a lower marketing budget, offering an economical yet efficient addition to artificial intelligence product managers’ marketing toolbox.

There are several limitations in this study. First, I used one-item scales to measure the dependent variable in each empirical study. Future researchers could include more items to allow for greater variability in dependent measures. Second, I used 7-point Likert scales to assess algorithm adoption in each study. Future studies could use a variety of measures (e.g., binary choices, willingness to pay, and actual behavior) to enhance the generalizability of my findings. Participants could also be asked to show their preferences separately for algorithms and people. Third, my scenarios included hypothetical decision tasks, which can appear artificial. Future researchers could explore related research questions further in real decision contexts. For example, field studies could be conducted to analyze hiring managers’ preference for algorithmic versus human guidance.

Studies could also explore other factors that may influence artificial intelligence adoption. For example, an increasing number of researchers are considering moral concerns behind algorithm aversion, such as threats to fairness (Verma & Rubin, 2018) or humanity (Nowak et al., 2018). Romantic matching is another noteworthy context in which algorithms’ utility may be debated (Joel et al., 2017). Why and when users’ acceptance of algorithms can be enhanced in these settings may be theoretically and practically useful. Researchers could consider how reliance on algorithms or human experts may affect people’s subsequent behavior. For example, whether heavy reliance on algorithms leads users to make decisions based on feelings or cognition (Chang & Hung, 2018), inspires them to focus on base rate rather than case-specific information (Yan & Sengupta, 2013), or amplifies their tendency to dehumanize targets (Haslam, 2006). Future studies could consider the possibilities I have outlined here, which may generate long-term and profound impacts on our society and everyday life.

Appendix A

Measurement of Decision Styles in Study 1

There are two primary decision-making styles, which involve different information use, thinking modes, and goals.

Style A: Decision makers care about propositions that hold generally (i.e., basic concepts or laws). For example, they have a strong interest in statements about the population and empirical evidence for those statements. They are interested in statements that hold true for many. They are concerned with the generality of their assertions and sacrifice specificity and comprehensiveness of descriptions to gain wider applicability of their principles.

Style B: Decision makers invest more effort in understanding individual cases and are relatively less oriented toward general propositions or principles. For example, they attempt to gain a fuller picture of an individual case and to capture its singularity and originality. Thus, they will pull together all available information about an individual case into a coherent narrative or case history. Their judgments trade extension for intension and general referents for singular comprehensiveness.

Appendix B

Manipulation of Information Processing Styles, Study 2A

Achievement Condition

Please imagine that you are the director of a postgraduate program and are considering admitting one student.

You and your colleagues have received résumés from 100 candidates and have decided to admit the one with the best academic achievement/performance during their undergraduate studies.

Nowadays, you can rely on either human experts or algorithms to make decisions.

Human experts usually make decisions based on their years of experience and expertise.

Algorithms usually make decisions based on historical data and mathematical models.

Potential Condition

Please imagine that you are the director of a postgraduate program and are considering admitting one student.

You and your colleagues have received résumés from 100 candidates and have decided to admit the one with the greatest potential to be successful during their postgraduate studies.

Nowadays, you can rely on either human experts or algorithms to make decisions.

Human experts usually make decisions based on their years of experience and expertise.

Algorithms usually make decisions based on historical data and mathematical models.

Appendix C

Manipulation of Information Processing Styles, Study 2B

Importance–Salience Condition

Please imagine that you are the director of a consumer lending team in a financial institution. Your team usually receives many loan applications in which applicants provide information about themselves and their loan purposes.

Your team needs to review these applications and then make accept or reject decisions based on your estimation of applicants’ ability to repay their loan.

Although a borrower’s actual loan repayment behavior may vary in different situations, you can still find basic rules or laws when predicting it.

It has been found in extensive research that people who fail to repay loans are similar in some respects. Specifically, applicants’ income, level of education, and credit history, which are reported on their application form, are the most important factors in loan repayment.

Nowadays, you can rely on either human experts or algorithms to decide whether to accept or reject applications.

Human experts usually make decisions based on their years of experience and expertise.

Algorithms usually make decisions based on historical data and mathematical models.

Control Condition

Please imagine that you are the director of a consumer lending team in a financial institution. Your team usually receives many loan applications in which applicants provide information about themselves (e.g., income, education, and credit history) and their loan purposes.

Your team needs to review these applications and then make accept or reject decisions based on your estimation of applicants’ ability to repay their loan.

Nowadays, you can rely on either human experts or algorithms to decide whether to accept or reject applications.

Human experts usually make decisions based on their years of experience and expertise.

Algorithms usually make decisions based on historical data and mathematical models.

Ajzen, I. (1977). Intuitive theories of events and the effects of base-rate information on prediction. Journal of Personality and Social Psychology, 35(5), 303–314.
https://doi.org/10.1037/0022-3514.35.5.303

Anderson, I. A., & Wood, W. (2021). Habits and the electronic herd: The psychology behind social media’s successes and failures. Consumer Psychology Review, 4(1), 83–99.
https://doi.org/10.1002/arcp.1063

Case, D. A., Fantino, E., & Goodie, A. S. (1999). Base-rate training without case cues reduces base-rate neglect. Psychonomic Bulletin & Review, 6(2), 319–327.
https://doi.org/10.3758/BF03212337

Castelo, N., Bos, M. W., & Lehmann, D. R. (2019). Task-dependent algorithm aversion. Journal of Marketing Research, 56(5), 809–825.
https://doi.org/10.1177/0022243719851788

Chang, H. H., & Hung, I. W. (2018). Mirror, mirror on the retail wall: Self-focused attention promotes reliance on feelings in consumer decisions. Journal of Marketing Research, 55(4), 586–599.
https://doi.org/10.1509/jmr.15.0080

Dawes, R. M. (1979). The robust beauty of improper linear models in decision making. American Psychologist, 34(7), 571–582.
https://doi.org/10.1037/0003-066X.34.7.571

Dawes, R. M., Faust, D., & Meehl, P. E. (1989). Clinical versus actuarial judgment. Science, 243(4899), 1668–1674.
https://doi.org/10.1126/science.2648573

Diab, D. L., Pui, S.-Y., Yankelevich, M., & Highhouse, S. (2011). Lay perceptions of selection decision aids in US and non-US samples. International Journal of Selection and Assessment, 19(2), 209–216.
https://doi.org/10.1111/j.1468-2389.2011.00548.x

Dietvorst, B. J., Simmons, J. P., & Massey, C. (2018). Overcoming algorithm aversion: People will use imperfect algorithms if they can (even slightly) modify them. Management Science, 64(3), 1155–1170.
https://doi.org/10.1287/mnsc.2016.2643

Eastwood, J., Snook, B., & Luther, K. (2012). What people want from their professionals: Attitudes toward decision-making strategies. Journal of Behavioral Decision Making, 25(5), 458–468.
https://doi.org/10.1002/bdm.741

Faggella, D. (2020, April 3). Artificial intelligence applications for lending and loan management. Emerj Artificial Intelligence Research. https://bit.ly/3tYEk9E

Grove, W. M., & Meehl, P. E. (1996). Comparative efficiency of informal (subjective, impressionistic) and formal (mechanical, algorithmic) prediction procedures: The clinical–statistical controversy. Psychology, Public Policy, and Law, 2(2), 293–323.
https://doi.org/10.1037/1076-8971.2.2.293

Grove, W. M., Zald, D. H., Lebow, B. S., Snitz, B. E., & Nelson, C. (2000). Clinical versus mechanical prediction: A meta-analysis. Psychological Assessment, 12(1), 19–30.
https://doi.org/10.1037/1040-3590.12.1.19

Haslam, N. (2006). Dehumanization: An integrative review. Personality and Social Psychology Review, 10(3), 252–264.
https://doi.org/10.1207/s15327957pspr1003_4

Highhouse, S. (2008). Stubborn reliance on intuition and subjectivity in employee selection. Industrial and Organizational Psychology, 1(3), 333–342.
https://doi.org/10.1111/j.1754-9434.2008.00058.x

Jia, J. S., Lu, X., Yuan, Y., Xu, G., Jia, J., & Christakis, N. A. (2020). Population flow drives spatio-temporal distribution of COVID-19 in China. Nature, 582(7812), 389–394.
https://doi.org/10.1038/s41586-020-2284-y

Joel, S., Eastwick, P. W., & Finkel, E. J. (2017). Is romantic desire predictable? Machine learning applied to initial romantic attraction. Psychological Science, 28(10), 1478–1489.
https://doi.org/10.1177/0956797617714580

Kahneman, D., & Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80(4), 237–251.
https://doi.org/10.1037/h0034747

Keeffe, B., Subramanian, U., Tierney, W. M., Udris, E., Willems, J., McDonell, M., & Fihn, S. D. (2005). Provider response to computer-based care suggestions for chronic heart failure. Medical Care, 43(5), 461–465.
https://doi.org/10.1097/01.mlr.0000160378.53326.f3

Koenig, R. (2018, August 22). Your next recruiter could be an algorithm. U.S. News & World Report. https://bit.ly/3v7GMMk

Kuncel, N. R., Klieger, D. M., Connelly, B. S., & Ones, D. S. (2013). Mechanical versus clinical data combination in selection and admissions decisions: A meta-analysis. Journal of Applied Psychology, 98(6), 1060–1072.
https://doi.org/10.1037/a0034156

Longoni, C., Bonezzi, A., & Morewedge, C. K. (2019). Resistance to medical artificial intelligence. Journal of Consumer Research, 46(4), 629–650.
https://doi.org/10.1093/jcr/ucz013

Lyon, D., & Slovic, P. (1976). Dominance of accuracy information and neglect of base rates in probability estimation. Acta Psychologica, 40(4), 287–298.
https://doi.org/10.1016/0001-6918(76)90032-9

Meehl, P. E. (1954). Clinical versus statistical prediction: A theoretical analysis and review of the evidence. University of Minnesota Press.
https://doi.org/10.1037/11281-000

Nisbett, R. E., & Borgida, E. (1975). Attribution and the psychology of prediction. Journal of Personality and Social Psychology, 32(5), 932–943.
https://doi.org/10.1037/0022-3514.32.5.932

Nowak, A., Lukowicz, P., & Horodecki, P. (2018). Assessing artificial intelligence for humanity: Will AI be the our biggest ever advance? Or the biggest threat [Opinion]. IEEE Technology and Society Magazine, 37(4), 26–34.
https://doi.org/10.1109/MTS.2018.2876105

Postmes, T., Spears, R., & Cihangir, S. (2001). Quality of decision making and group norms. Journal of Personality and Social Psychology, 80(6), 918–930.
https://doi.org/10.1037/0022-3514.80.6.918

Sawaitul, S. D., Wagh, K. P., & Chatur, P. N. (2012). Classification and prediction of future weather by using back propagation algorithm: An approach. International Journal of Emerging Technology and Advanced Engineering, 2(1), 110–113.

Tinsley, C. H., & Brett, J. M. (2001). Managing workplace conflict in the United States and Hong Kong. Organizational Behavior and Human Decision Processes, 85(2), 360–381.
https://doi.org/10.1006/obhd.2000.2944

Tormala, Z. L., Jia, J. S., & Norton, M. I. (2012). The preference for potential. Journal of Personality and Social Psychology, 103(4), 567–583.
https://doi.org/10.1037/a0029227

Tversky, A., & Kahneman, D. (1980). Causal schemas in judgments under uncertainty. In M. Fishbein (Ed.), Progress in social psychology (Vol. 1, pp. 49–72). Erlbaum.

Verma, S., & Rubin, J. (2018, May 29). Fairness definitions explained [Conference presentation]. 2018 IEEE/ACM International Workshop on Software Fairness, Association for Computing Machinery, New York, NY, USA.
https://doi.org/10.1145/3194770.3194776

Vrieze, S. I., & Grove, W. M. (2009). Survey on the use of clinical and mechanical prediction methods in clinical psychology. Professional Psychology: Research and Practice, 40(5), 525–531.
https://doi.org/10.1037/a0014693

Yan, D., & Sengupta, J. (2013). The influence of base rate and case information on health-risk perceptions: A unified model of self-positivity and self-negativity. Journal of Consumer Research, 39(5), 931–946.
https://doi.org/10.1086/666596

Zuckerman, M. (1978). Use of consensus information in prediction of behavior. Journal of Experimental Social Psychology, 14(2), 163–171.
https://doi.org/10.1016/0022-1031(78)90022-7

Zukier, H., & Pepitone, A. (1984). Social roles and strategies in prediction: Some determinants of the use of base-rate information. Journal of Personality and Social Psychology, 47(2), 349–360.
https://doi.org/10.1037/0022-3514.47.2.349

This work was supported by the National Natural Science Foundation of China (71802125).

Ke Zhang, SILC Business School, Office 306, Wenshang Building, Shanghai University, 20 Chengzhong Road, Jiading, Shanghai 201800, People’s Republic of China. Email: [email protected]