Che-Wei Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Che-Wei Liu has a total of 33 co-authors. Here are the top co-authors by number of shared publications:
Jiang Duan
3 shared publications- Sunil Mithas
University of South Florida
3 shared publications - Akshat Lakhiwal
C. Herman and Mary Virginia Terry College of Business, University of Georgia, Athens, Georgia 30602
2 shared publications - Guodong (Gordon) Gao
Johns Hopkins University
2 shared publications - Hung-Yue Suen
Technology Application and Human Resource Development, National Taiwan Normal University, Taipei, Taiwan
2 shared publications - J. J. Po-An Hsieh
Robinson College of Business, Georgia State University, Atlanta, Georgia 30303
2 shared publications
Latest Publications
- 20 October 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
Seeing Less, Engaging More: Rethinking Early User Experience on Generative AI Cocreation Platforms—Findings from a Field Experiment
Generative artificial intelligence (GenAI) content generation platforms such as DALL·E or Midjourney produce impressive content in seconds. But this very immediacy often flattens creation into a single moment, stripping away the evolving experience shown to sustain engagement. In response, platforms have increasingly been experimenting with how much output to reveal and how to prompt registration, restoring some of that lost experience. How, then, should early user experience be structured to bolster engagement? This research studies two user experience levers. The first is fulfillment, that is, the extent to which GenAI content is revealed before registration. The second is the framing of the registration message itself. We partnered with a GenAI content generation platform, running a randomized field experiment with over 20,000 first-time users. Full fulfillment satisfied users immediately, but partial fulfillment worked better, activating two motivational pathways: value-in-use as the output reflected the user’s prompt and curiosity about what remained unseen. Counterintuitively, loss-framed registration messages substituted for rather than reinforced this effect, increasing registration under partial disclosure but losing impact once the full output was visible. These effects extended beyond registration, as users returned more often and engaged more substantively. Revealing less, structured well, is what turns a fleeting encounter into lasting engagement.
- 20 October 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
Seeing Less, Engaging More: Rethinking Early User Experience on Generative AI Cocreation Platforms—Findings from a Field Experiment
Generative artificial intelligence (GenAI) content generation platforms such as DALL·E or Midjourney produce impressive content in seconds. But this very immediacy often flattens creation into a single moment, stripping away the evolving experience shown to sustain engagement. In response, platforms have increasingly been experimenting with how much output to reveal and how to prompt registration, restoring some of that lost experience. How, then, should early user experience be structured to bolster engagement? This research studies two user experience levers. The first is fulfillment, that is, the extent to which GenAI content is revealed before registration. The second is the framing of the registration message itself. We partnered with a GenAI content generation platform, running a randomized field experiment with over 20,000 first-time users. Full fulfillment satisfied users immediately, but partial fulfillment worked better, activating two motivational pathways: value-in-use as the output reflected the user’s prompt and curiosity about what remained unseen. Counterintuitively, loss-framed registration messages substituted for rather than reinforced this effect, increasing registration under partial disclosure but losing impact once the full output was visible. These effects extended beyond registration, as users returned more often and engaged more substantively. Revealing less, structured well, is what turns a fleeting encounter into lasting engagement.
- Open Access
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Free First vs. Trial First: Onboarding Design and Premium Discovery in Freemium Services
Freemium firms often choose between a free first approach, where new users begin in the basic version, and a trial first approach, where onboarding guides users into a free premium trial. We argue that this choice determines who bears the cost of discovering premium value. Because premium features are experience goods, users must reach and experience premium before they can judge its worth, yet users left in a free first flow rarely seek it out on their own. We study a natural experiment at a large image-editing platform, where an unexpected glitch shifted desktop onboarding from the normal trial first flow to a free first flow for 48 hours while the free trial remained available elsewhere. Using a difference-indifferences design on 174,159 new users, we estimate that free first lowers subscription by 42%. Clickstream evidence further shows that free first works by keeping users from reaching premium at all, not by changing how they evaluate it afterward. The effect persists in lower yearly-plan adoption and two-year spending, and refund patterns suggest that trial first expands conversion without diluting subscriber quality. The disadvantage of free first is larger where users face greater uncertainty or exploration frictions and disappears for high-intent organic-search users, a gradient consistent with a discovery-cost mechanism rather than a mechanical loss of a visible entry point.
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFree First vs. Trial First: Onboarding Design and Premium Discovery in Freemium Services
Freemium firms often choose between a free first approach, where new users begin in the basic version, and a trial first approach, where onboarding guides users into a free premium trial. We argue that this choice determines who bears the cost of discovering premium value. Because premium features are experience goods, users must reach and experience premium before they can judge its worth, yet users left in a free first flow rarely seek it out on their own. We study a natural experiment at a large image-editing platform, where an unexpected glitch shifted desktop onboarding from the normal trial first flow to a free first flow for 48 hours while the free trial remained available elsewhere. Using a difference-indifferences design on 174,159 new users, we estimate that free first lowers subscription by 42%. Clickstream evidence further shows that free first works by keeping users from reaching premium at all, not by changing how they evaluate it afterward. The effect persists in lower yearly-plan adoption and two-year spending, and refund patterns suggest that trial first expands conversion without diluting subscriber quality. The disadvantage of free first is larger where users face greater uncertainty or exploration frictions and disappears for high-intent organic-search users, a gradient consistent with a discovery-cost mechanism rather than a mechanical loss of a visible entry point.
- Open Access
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Governing the Social Vacuum: How Vulnerability and Social Norms Shape Prosocial Behavior in Unsupervised Service Environments
As autonomous machines replace human service providers, the social accountability that ordinarily restrains misconduct vanishes. We propose that AI systems can curb antisocial behavior by explicitly revealing their functional limitations and soliciting human help, rather than projecting technological perfection. We test this AI vulnerability signaling strategy against the established benchmark of social norm appeals in a large-scale randomized field experiment encompassing 90,428 rides on a commercial autonomous robotaxi platform. Both strategies reduced littering by approximately 18-21% relative to the baseline rate. However, AI vulnerability proved superior in two respects. First, social norm appeals lost nearly all effectiveness in the presence of environmental stressors, in our case rainy days; AI vulnerability remained fully robust. Second, social norm appeals were associated with a marginally significant reduction in riders’ willingness to take a subsequent ride while AI vulnerability carried no such cost. A controlled laboratory experiment (1,317 participants; 2,144 task-level observations) revealed why: these strategies engage fundamentally different psychological pathways. Social norm appeals operate through descriptive norms as an externally oriented compliance mechanism that demands cognitive resources. AI vulnerability operates through perceived prosocial impact as an internal empowerment mechanism that activates users’ sense of making a difference. These pathways function as substitutes: combining both strategies yielded no additional benefit. Results establish AI vulnerability signaling as a scalable governance tool for unsupervised human–machine interaction.
- 29 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessGoverning the Social Vacuum: How Vulnerability and Social Norms Shape Prosocial Behavior in Unsupervised Service Environments
As autonomous machines replace human service providers, the social accountability that ordinarily restrains misconduct vanishes. We propose that AI systems can curb antisocial behavior by explicitly revealing their functional limitations and soliciting human help, rather than projecting technological perfection. We test this AI vulnerability signaling strategy against the established benchmark of social norm appeals in a large-scale randomized field experiment encompassing 90,428 rides on a commercial autonomous robotaxi platform. Both strategies reduced littering by approximately 18-21% relative to the baseline rate. However, AI vulnerability proved superior in two respects. First, social norm appeals lost nearly all effectiveness in the presence of environmental stressors, in our case rainy days; AI vulnerability remained fully robust. Second, social norm appeals were associated with a marginally significant reduction in riders’ willingness to take a subsequent ride while AI vulnerability carried no such cost. A controlled laboratory experiment (1,317 participants; 2,144 task-level observations) revealed why: these strategies engage fundamentally different psychological pathways. Social norm appeals operate through descriptive norms as an externally oriented compliance mechanism that demands cognitive resources. AI vulnerability operates through perceived prosocial impact as an internal empowerment mechanism that activates users’ sense of making a difference. These pathways function as substitutes: combining both strategies yielded no additional benefit. Results establish AI vulnerability signaling as a scalable governance tool for unsupervised human–machine interaction.
- 15 June 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
From Opacity to Transparency: User Behavior and Downstream Effects in Algorithmic Evaluation
Organizations are increasingly integrating artificial intelligence (AI) into traditionally human-driven evaluation processes, such as recruitment and performance evaluation. Yet the complexity of the underlying AI algorithms often renders these processes opaque, limiting users’ understanding and potentially inducing stress and other behavioral changes. In response to these challenges, transparency is often advocated as a mitigative approach, but its effects remain somewhat ambiguous. While transparency in algorithmic evaluation may mitigate interviewees’ stress in the recruitment context, it may also incentivize opportunistic impression management (IM), thereby engendering a transparency paradox. Focusing on this tension, we theorize that while algorithmic evaluation, relative to human evaluation, may heighten stress and deceptive IM in the recruitment process, transparency has the potential to mitigate these disruptions. We test this across eight studies, including two main experiments and a quasi-field replication. In Experiment I, we found that while algorithmic evaluation increased interviewees’ stress and deceptive IM relative to human evaluation, transparency counteracted these effects, aligning them closely with human evaluation. In Experiment II, we study how these effects caused deviations in downstream interview performance when such performance is assessed in human-only, AI-augmented (wherein human decision-makers are assisted by AI scores), or fully automated decision-making configurations. Although transparency consistently reduced deviations in stress and IM from human evaluation for interviewees, its corrective effect on interview performance was constrained on the evaluative side: when AI scores were present, evaluators anchored on them, discounting their own judgment, even as the AI scores failed to distinguish honest from deceptive behaviors. Together, these findings disentangle the dual-edged effects of transparency in algorithmic evaluation, showing that its benefits depend not only on how users respond, but also on how evaluation is conducted. This work advances research on algorithmic decision-making and socio-technical systems, and offers implications for organizations, policymakers, and AI developers.
- 15 June 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
From Opacity to Transparency: User Behavior and Downstream Effects in Algorithmic Evaluation
Organizations are increasingly integrating artificial intelligence (AI) into traditionally human-driven evaluation processes, such as recruitment and performance evaluation. Yet the complexity of the underlying AI algorithms often renders these processes opaque, limiting users’ understanding and potentially inducing stress and other behavioral changes. In response to these challenges, transparency is often advocated as a mitigative approach, but its effects remain somewhat ambiguous. While transparency in algorithmic evaluation may mitigate interviewees’ stress in the recruitment context, it may also incentivize opportunistic impression management (IM), thereby engendering a transparency paradox. Focusing on this tension, we theorize that while algorithmic evaluation, relative to human evaluation, may heighten stress and deceptive IM in the recruitment process, transparency has the potential to mitigate these disruptions. We test this across eight studies, including two main experiments and a quasi-field replication. In Experiment I, we found that while algorithmic evaluation increased interviewees’ stress and deceptive IM relative to human evaluation, transparency counteracted these effects, aligning them closely with human evaluation. In Experiment II, we study how these effects caused deviations in downstream interview performance when such performance is assessed in human-only, AI-augmented (wherein human decision-makers are assisted by AI scores), or fully automated decision-making configurations. Although transparency consistently reduced deviations in stress and IM from human evaluation for interviewees, its corrective effect on interview performance was constrained on the evaluative side: when AI scores were present, evaluators anchored on them, discounting their own judgment, even as the AI scores failed to distinguish honest from deceptive behaviors. Together, these findings disentangle the dual-edged effects of transparency in algorithmic evaluation, showing that its benefits depend not only on how users respond, but also on how evaluation is conducted. This work advances research on algorithmic decision-making and socio-technical systems, and offers implications for organizations, policymakers, and AI developers.
- 3 June 2026
- Published by SAGE Publications in Production and Operations Management
EXPRESS: How Information Technology and R&D Investments Influence Geographic Dispersion and Production Costs: Evidence from US Multinational Manufacturers
How do information technology (IT) and R&D investments jointly shape the global dispersion of multinational manufacturers’ operations and their production costs? We argue that IT enhances a firm’s ability to leverage R&D-induced technological competence across different regions, facilitating the geographic dispersion of internal supply chain units and subsidiaries. In addition, IT helps firms extract greater value from geographic dispersion in terms of reduced production costs via enhanced coordination, control, and connectivity. Using rare archival firm-level data on US-based multinational manufacturers for the 1994–2009 period, we test these arguments and document two major findings. First, IT intensity positively moderates the relationship between R&D intensity and geographic dispersion. Second, IT intensity negatively moderates the relationship between geographic dispersion and production costs. Together, these findings highlight the critical role of IT in amplifying the effect of R&D on geographic dispersion, and in turn, the effect of geographic dispersion on production cost advantages. Overall, the findings suggest that investments in IT systems and resources help mitigate coordination challenges in leveraging technological competence for global dispersion and managing dispersed global supply chains for superior firm performance.
- 3 June 2026
- Published by SAGE Publications in Production and Operations Management
EXPRESS: How Information Technology and R&D Investments Influence Geographic Dispersion and Production Costs: Evidence from US Multinational Manufacturers
How do information technology (IT) and R&D investments jointly shape the global dispersion of multinational manufacturers’ operations and their production costs? We argue that IT enhances a firm’s ability to leverage R&D-induced technological competence across different regions, facilitating the geographic dispersion of internal supply chain units and subsidiaries. In addition, IT helps firms extract greater value from geographic dispersion in terms of reduced production costs via enhanced coordination, control, and connectivity. Using rare archival firm-level data on US-based multinational manufacturers for the 1994–2009 period, we test these arguments and document two major findings. First, IT intensity positively moderates the relationship between R&D intensity and geographic dispersion. Second, IT intensity negatively moderates the relationship between geographic dispersion and production costs. Together, these findings highlight the critical role of IT in amplifying the effect of R&D on geographic dispersion, and in turn, the effect of geographic dispersion on production cost advantages. Overall, the findings suggest that investments in IT systems and resources help mitigate coordination challenges in leveraging technological competence for global dispersion and managing dispersed global supply chains for superior firm performance.
- Open Access
- 20 April 2026
- Published by Elsevier in SSRN Electronic Journal
Free First vs. Trial First: The Impact of Trial Approach on User Conversion
The freemium model, which provides basic services at no cost and charges for advanced features, has seen widespread adoption in the software, gaming, and web service sectors. Firms typically adopt either a free first approach, where users access basic features with some limitations, or a trial first approach, which offers time-limited access to premium features. Despite the prevalence of the freemium approach, there is limited empirical evidence on the relative effectiveness of these strategies. We address this gap using a natural experiment caused by an unexpected system glitch, in which the trial prompt page was unintentionally removed for 48 hours. This change redirected users to the basic service from trial first to free first and led to a 42% decline in subscription rates. Our further analysis reveals two mechanisms behind this effect. The free first limits early exposure to premium features, reflected in lower trial entry and reduced use of premium functionalities. This constrained exposure weakens users' learning about product value and leads to less efficient task execution, reducing their willingness to subscribe. Moreover, the negative impact persists over time, leading to fewer yearly subscriptions after the free trial and lower cumulative subscriptions and total spending over a two-year period. Using a Generalized Random Forest approach, we further document substantial heterogeneity. The negative effect of free first is stronger in countries with higher uncertainty avoidance and lower internet efficiency, and weaker for users arriving via organic search. Our findings provide novel evidence on how early user experience shapes learning, efficiency, and long-term revenue, offering actionable guidance for firms seeking to design more effective freemium strategies.
- 20 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFree First vs. Trial First: The Impact of Trial Approach on User Conversion
The freemium model, which provides basic services at no cost and charges for advanced features, has seen widespread adoption in the software, gaming, and web service sectors. Firms typically adopt either a free first approach, where users access basic features with some limitations, or a trial first approach, which offers time-limited access to premium features. Despite the prevalence of the freemium approach, there is limited empirical evidence on the relative effectiveness of these strategies. We address this gap using a natural experiment caused by an unexpected system glitch, in which the trial prompt page was unintentionally removed for 48 hours. This change redirected users to the basic service from trial first to free first and led to a 42% decline in subscription rates. Our further analysis reveals two mechanisms behind this effect. The free first limits early exposure to premium features, reflected in lower trial entry and reduced use of premium functionalities. This constrained exposure weakens users' learning about product value and leads to less efficient task execution, reducing their willingness to subscribe. Moreover, the negative impact persists over time, leading to fewer yearly subscriptions after the free trial and lower cumulative subscriptions and total spending over a two-year period. Using a Generalized Random Forest approach, we further document substantial heterogeneity. The negative effect of free first is stronger in countries with higher uncertainty avoidance and lower internet efficiency, and weaker for users arriving via organic search. Our findings provide novel evidence on how early user experience shapes learning, efficiency, and long-term revenue, offering actionable guidance for firms seeking to design more effective freemium strategies.
- Open Access
- 17 April 2026
- Published by Elsevier in SSRN Electronic Journal
Signals of Efforts or Evidence of Automation? The Countersignaling Effect of Prompt Disclosure
Transparency has often been regarded as a valuable practice in user-generated content (UGC), where revealing the creative process is thought to enhance perceived authenticity, effort, and credibility. However, in the context of human-AI co-creation, this logic may no longer hold. When generative AI (GenAI) plays a co-creating role in producing content, process-level disclosure-such as revealing the textual prompt used to generate an image-may instead expose the minimal nature of human input, undermining perceptions of creators' efforts and creativity. Drawing on signaling theory and the concept of countersignaling, we examine how prompt visibility affects audience engagement on a major GenAI platform. Analyzing a largescale dataset of user posts, we find that prompt display significantly reduces engagement relative to concealment. Rather than reflecting a uniform transparency effect, this engagement penalty arises because prompt disclosure shifts audience evaluation toward a proportionality-based assessment of creative contribution, in which viewers implicitly compare observable human input with the sophistication of the generated output. Consistent with this logic, transparency is less costly when disclosed prompts make substantial human input salient, but more damaging when highly polished outputs render the revealed input seemingly insufficient. In sum, we find that in AI-mediated creativity, transparency is not uniformly rewarded; its effects depend on how disclosed process information reshapes audience inferences about human contribution. Our findings challenge conventional assumptions about transparency in digital creativity and offer new insights into how audiences evaluate human contribution in AI-mediated content. We discuss implications for creators, platform designers, and researchers navigating attribution, engagement, and disclosure in the GenAI age.
- 17 April 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSignals of Efforts or Evidence of Automation? The Countersignaling Effect of Prompt Disclosure
Transparency has often been regarded as a valuable practice in user-generated content (UGC), where revealing the creative process is thought to enhance perceived authenticity, effort, and credibility. However, in the context of human-AI co-creation, this logic may no longer hold. When generative AI (GenAI) plays a co-creating role in producing content, process-level disclosure-such as revealing the textual prompt used to generate an image-may instead expose the minimal nature of human input, undermining perceptions of creators' efforts and creativity. Drawing on signaling theory and the concept of countersignaling, we examine how prompt visibility affects audience engagement on a major GenAI platform. Analyzing a largescale dataset of user posts, we find that prompt display significantly reduces engagement relative to concealment. Rather than reflecting a uniform transparency effect, this engagement penalty arises because prompt disclosure shifts audience evaluation toward a proportionality-based assessment of creative contribution, in which viewers implicitly compare observable human input with the sophistication of the generated output. Consistent with this logic, transparency is less costly when disclosed prompts make substantial human input salient, but more damaging when highly polished outputs render the revealed input seemingly insufficient. In sum, we find that in AI-mediated creativity, transparency is not uniformly rewarded; its effects depend on how disclosed process information reshapes audience inferences about human contribution. Our findings challenge conventional assumptions about transparency in digital creativity and offer new insights into how audiences evaluate human contribution in AI-mediated content. We discuss implications for creators, platform designers, and researchers navigating attribution, engagement, and disclosure in the GenAI age.
- 2 October 2025
- Published by Taylor & Francis in Journal of Management Information Systems
- Vol. 42 (4) , 1243-1278
When Good Intentions Backfire: The Asymmetric Effects of Minority-Ownership Markers for Businesses on Online Platforms
- 2 October 2025
- Published by Taylor & Francis in Journal of Management Information Systems
- Vol. 42 (4) , 1243-1278
When Good Intentions Backfire: The Asymmetric Effects of Minority-Ownership Markers for Businesses on Online Platforms
- 1 June 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 36 (2) , 828-846
Mobile Apps, Trading Behaviors, and Portfolio Performance: Evidence from a Quasi-Experiment in China
Mobile Apps, Trading Behaviors, and Portfolio Performance: Evidence from a Quasi-Experiment in China How do mobile apps influence individual investors’ financial decisions and performance? This study answers this timely and important question by using rare archival data from a large securities company in China using a sample of 20,665 investors. Authors find that mobile app adoption does not affect investors’ portfolio performance when one examines aggregate impacts using a binary indicator of mobile app use. Their additional analyses suggest that adopting mobile apps results in a noticeable decrease in time constraints, a proxy for transaction friction, and a modest increase in trend-chasing bias, reflecting tendencies toward myopic decision making. Because the reduction in time constraints can benefit investors’ performance, the increase in trend chasing can be detrimental to investors’ performance, these findings explain why mobile app adoption has no overall effect on portfolio performance. Further analyses of adopters’ postadoption behaviors provide interesting insights and show that the mobile app usage intensity has an inverted U–shaped relationship with portfolio performance. The results are robust to using different samples or excluding high market volatility periods and by using a variety of methods, such as propensity score matching, dynamic matching, stacked difference-in-differences, or an instrumental variable approach.
- 1 June 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Information Systems Research
- Vol. 36 (2) , 828-846
Mobile Apps, Trading Behaviors, and Portfolio Performance: Evidence from a Quasi-Experiment in China
Mobile Apps, Trading Behaviors, and Portfolio Performance: Evidence from a Quasi-Experiment in China How do mobile apps influence individual investors’ financial decisions and performance? This study answers this timely and important question by using rare archival data from a large securities company in China using a sample of 20,665 investors. Authors find that mobile app adoption does not affect investors’ portfolio performance when one examines aggregate impacts using a binary indicator of mobile app use. Their additional analyses suggest that adopting mobile apps results in a noticeable decrease in time constraints, a proxy for transaction friction, and a modest increase in trend-chasing bias, reflecting tendencies toward myopic decision making. Because the reduction in time constraints can benefit investors’ performance, the increase in trend chasing can be detrimental to investors’ performance, these findings explain why mobile app adoption has no overall effect on portfolio performance. Further analyses of adopters’ postadoption behaviors provide interesting insights and show that the mobile app usage intensity has an inverted U–shaped relationship with portfolio performance. The results are robust to using different samples or excluding high market volatility periods and by using a variety of methods, such as propensity score matching, dynamic matching, stacked difference-in-differences, or an instrumental variable approach.
- 8 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (5) , 5949-5960
Artificial Intelligence Can Recognize Whether a Job Applicant Is Selling and/or Lying According to Facial Expressions and Head Movements Much More Correctly Than Human Interviewers
Whether an interviewee’s honest and deceptive responses can be detected by the signals of facial expressions in videos has been debated and called to be researched. We developed deep learningmodels enabled by computer vision to extract the temporal patterns of job applicants’ facial expressions and head movements to identify self-reported honest and deceptive impression management (IM) tactics from video frames in real asynchronous video interviews. A 12- to 15-min video was recorded for each of the N = 121 job applicants as they answered five structured behavioral interview questions. Each applicant completed a survey to self-evaluate their trustworthiness on four IM measures. Additionally, a field experiment was conducted to compare the concurrent validity associated with self-reported IMs between our modeling and human interviewers. Human interviewers’ performance in predicting these IMmeasures from another subset of 30 videos was obtained by having N = 30 human interviewers evaluate three recordings. Our models explained 91% and 84% of the variance in honest and deceptive IMs, respectively, and showed a stronger correlation with self-reported IMscores compared to human interviewers.
- 8 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. 11 (5) , 5949-5960
Artificial Intelligence Can Recognize Whether a Job Applicant Is Selling and/or Lying According to Facial Expressions and Head Movements Much More Correctly Than Human Interviewers
Whether an interviewee’s honest and deceptive responses can be detected by the signals of facial expressions in videos has been debated and called to be researched. We developed deep learningmodels enabled by computer vision to extract the temporal patterns of job applicants’ facial expressions and head movements to identify self-reported honest and deceptive impression management (IM) tactics from video frames in real asynchronous video interviews. A 12- to 15-min video was recorded for each of the N = 121 job applicants as they answered five structured behavioral interview questions. Each applicant completed a survey to self-evaluate their trustworthiness on four IM measures. Additionally, a field experiment was conducted to compare the concurrent validity associated with self-reported IMs between our modeling and human interviewers. Human interviewers’ performance in predicting these IMmeasures from another subset of 30 videos was obtained by having N = 30 human interviewers evaluate three recordings. Our models explained 91% and 84% of the variance in honest and deceptive IMs, respectively, and showed a stronger correlation with self-reported IMscores compared to human interviewers.