Paper deep dive
Rethinking AI Literacy Education in Higher Education: Bridging Risk Perception and Responsible Adoption
Shasha Yu, Fiona Carroll, Barry L. Bentley
Intelligence
Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 96%
Last extracted: 4/1/2026, 1:39:09 AM
Summary
This study investigates AI risk perception and adoption willingness among 139 technology students. It identifies a 'risk underappreciation' phenomenon where students in AI-related fields show high explicit risk awareness but lower recognition of risks in applied scenarios. The findings highlight an inverse relationship between perceived risk and adoption willingness, and suggest that technical education influences gender differences in risk awareness and adoption behavior.
Entities (6)
Relation Signals (4)
Shasha Yu â authored â Rethinking AI Literacy Education in Higher Education: Bridging Risk Perception and Responsible Adoption
confidence 100% · Title page authorship
Fiona Carroll â authored â Rethinking AI Literacy Education in Higher Education: Bridging Risk Perception and Responsible Adoption
confidence 100% · Title page authorship
Barry L. Bentley â authored â Rethinking AI Literacy Education in Higher Education: Bridging Risk Perception and Responsible Adoption
confidence 100% · Title page authorship
Technology Students â exhibited â Risk Underappreciation
confidence 90% · Study findings section
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:As AI becomes increasingly embedded across societal domains, understanding how future AI practitioners, particularly technology students, perceive its risks is essential for responsible development and adoption. This study analyzed responses from 139 students in Computer Science, Data Science/Data Analytics, and other disciplines using both explicit AI risk ratings and scenario-based assessments of risk and adoption willingness. Four key findings emerged: (1) Students expressed substantially higher concern for concrete, explicitly stated risks than for abstract or scenario-embedded risks; (2) Perceived risk and willingness to adopt AI demonstrated a clear inverse relationship; (3) Although technical education narrowed gender differences in risk awareness, male students reported higher adoption willingness; and (4) A form of "risk underappreciation" was observed, wherein students in AI-related specializations showed both elevated explicit risk awareness and higher willingness to adopt AI, despite lower recognition of risks in applied scenarios. These findings underscore the need for differentiated AI literacy strategies that bridge the gap between awareness and responsible adoption and offer valuable insights for educators, policymakers, industry leaders, and academic institutions aiming to cultivate ethically informed and socially responsible AI practitioners.
Tags
Links
- Source: https://arxiv.org/abs/2603.29935v1
- Canonical: https://arxiv.org/abs/2603.29935v1
Trouble viewing inline? Open PDF directly â
Full Text
67,455 characters extracted from source content.
Expand or collapse full text
Rethinking AI Literacy Education in Higher Education: Bridging Risk Perception and Responsible Adoption Shasha Yu Fiona Carroll Barry L. Bentley Abstract As AI becomes increasingly embedded across societal domains, understanding how future AI practitionersâparticularly technology studentsâperceive its risks is essential for responsible development and adoption. This study analyzed responses from 139 students in Computer Science, Data Science/Data Analytics, and other disciplines using both explicit AI risk ratings and scenario-based assessments of risk and adoption willingness. Four key findings emerged: (1) Students expressed substantially higher concern for concrete, explicitly stated risks than for abstract or scenario-embedded risks; (2) Perceived risk and willingness to adopt AI demonstrated a clear inverse relationship; (3) Although technical education narrowed gender differences in risk awareness, male students reported higher adoption willingness; and (4) A form of ârisk underappreciationâ was observed, wherein students in AI-related specializations showed both elevated explicit risk awareness and higher willingness to adopt AI, despite lower recognition of risks in applied scenarios. These findings underscore the need for differentiated AI literacy strategies that bridge the gap between awareness and responsible adoption and offer valuable insights for educators, policymakers, industry leaders, and academic institutions aiming to cultivate ethically informed and socially responsible AI practitioners. keywords: Artificial Intelligence; Risk Awareness; Risk Perception; AI Education; Adoption; Scenario-Based Assessment tn1tn1footnotetext: This is the accepted manuscript of the article accepted for publication in Social Sciences & Humanities Open. The final authenticated version will be available via the journal website. [inst1]organization=School of Professional Studies, Clark University,addressline=950 Main Street, city=Worcester, postcode=01610, state=MA, country=USA [inst2]organization=Cardiff School of Technologies, Cardiff Metropolitan University,addressline=Llandaff Campus, Western Avenue, city=Cardiff, postcode=CF5 2YB, state=Wales, country=UK [inst3]organization=Harvard Medical School,addressline=25 Shattuck Street, city=Boston, postcode=02115, state=MA, country=USA 1 Introduction Artificial Intelligence (AI) is reshaping modern society across sectors such as healthcare, education, transportation, and entertainment (Shaheen, 2021; Zhai et al., 2021). Its integration delivers substantial benefitsâincluding efficiency gains, new discoveries, and improved convenience (Lee et al., 2023)âwhile simultaneously introducing well-documented risks such as privacy breaches, algorithmic bias, job displacement, and misuse (Yu and Carroll, 2023). Despite growing public exposure to AI, awareness of these risks remains uneven (Yu et al., 2024a; Brauner et al., 2023), at times leading to uninformed adoption and inadequate scrutiny (Yu et al., 2024c). The rapid expansion of generative AI in both educational and professional settings has intensified the need to understand how emerging practitioners assess AI-related risks. Within computing and engineering education, scholars emphasize that future AI practitioners require explicit preparation for ethical reasoning and responsible design practices (Kasinidou et al., 2021b; Brown et al., 2024; Archambault et al., 2024). Technology studentsâwho will become the developers, analysts, and engineers responsible for building and governing AI systemsârepresent a particularly consequential population. Their perceptions of AI risks shape not only their own adoption behaviors but also the decisions they will make as future professionals. However, although prior studies have examined student attitudes toward AI (SĂĄez-Velasco et al., 2025; Stöhr et al., 2024) and perceptions of AIâs benefits and risks in educational contexts (Li et al., 2025; Tierney et al., 2025), comparatively little work has jointly examined (a) explicit risk awareness, (b) scenario-based recognition of risks in applied contexts, and (c) how these relate to willingness to adopt AI technologies. Key questions therefore persist: 1. What types of AI risks are tech students most concerned with? 2. Are there patterns or gaps in their understanding of AI risks that need to be addressed? 3. How does their awareness of risks shape their willingness to adopt, develop, or advocate for AI technologies? 4. Are there differences in the understanding of AI risks based on technical specialization? Existing studies examine studentsâ attitudes toward AI adoption (SĂĄez-Velasco et al., 2025) and their perceptions of AIâs benefits and risks in educational contexts (Li et al., 2025; Tierney et al., 2025). Some works investigate general student populations (Delello et al., 2023) or center on adoption determinants rather than risk awareness (Wang et al., 2021), leaving technology studentsâ AI risk understanding comparatively underexamined. Research in computing education and FATE further documents uneven recognition of fairness and safety issues in real-world systems (Kasinidou et al., 2021b, a; Brown et al., 2024; Archambault et al., 2024; Pierson, 2017), underscoring the need for empirical work that situates risk awareness as a core construct in understanding responsible AI adoption. This gap limits the development of targeted AI literacy programs that prepare students to navigate and mitigate risks responsibly. To address this need, the present study investigates the current state of AI risk awareness among technology students, examines patterns in explicit and scenario-based assessments, and analyzes how these perceptions shape willingness to adopt AI technologies. By integrating both explicit and scenario-based assessments, the study also provides a more comprehensive view of how students articulate risks in the abstract and how they recognize them in applied contexts. By uncovering these relationships, this work seeks to inform curriculum design and policy development aimed at fostering responsible AI use and development. Ultimately, the study contributes to bridging the gap between rapid AI innovation and informed, ethical practice. 2 Background 2.1 AI Risks and Challenges A growing body of research highlights the multifaceted risks associated with AI systems across technical, ethical, and societal dimensions. Concerns about privacy and data security are among the most frequently discussed, particularly as AI technologies increasingly rely on large-scale personal data and often enable secondary uses beyond individualsâ awareness (Yu et al., 2023; Manikonda et al., 2018; NaudĂ©, 2020). Alongside privacy issues, scholars have extensively documented how algorithmic systems reproduce or even amplify social biases, leading to unfair outcomes in high-stakes settings such as hiring, lending, or policing (Roselli et al., 2019; Raghavan et al., 2020; Albaroudi et al., 2024). Technical vulnerabilities also represent an important class of risks: adversarial attacks and other forms of system manipulation can compromise the reliability of AI models deployed in critical domains (Chen et al., 2019; Manikonda et al., 2018). Beyond these technical concerns, AIâs broader societal implications have also drawn increasing scholarly attention. Automation-related job displacement continues to raise questions about economic inequality and the need for large-scale workforce adaptation (Rawashdeh, 2025; Yu and Carroll, 2023). The rapid spread of misinformationâincluding deepfakes and synthetic mediaâillustrates how AI can be leveraged to distort public discourse and manipulate public opinion (AĂŻmeur et al., 2023; Al-Asadi and Tasdemir, 2021; Agarwal et al., 2020). In parallel, the expansion of AI-enabled autonomous weapons has intensified ethical and accountability debates, especially regarding the delegation of lethal decision-making to non-human agents (Ăngel GĂłmez de Ăgreda, 2020; Yu and Carroll, 2022a). Lack of transparency further complicates responsible AI deployment. âBlack-boxâ systems limit usersâ and regulatorsâ ability to understand or contest automated decisions, raising particular concerns in areas such as criminal justice and healthcare (Larsson and Heintz, 2020; Von Eschenbach, 2021; Bernal and Mazo, 2022). These transparency challenges intersect with a broader set of ethical issuesâincluding autonomy, consent, and civil libertiesâthat scholars have argued are central to assessing the real-world impact of AI technologies (Stahl and Stahl, 2021; Benzinger et al., 2023; Saheb, 2023). AI systems also pose psychological and behavioral risks, influencing creativity, critical thinking, social interaction, and exposure to diverse perspectives through personalization and algorithmic filtering (Torres et al., 2024; Zhai et al., 2024; Yu et al., 2024b; Park and Park, 2024). Finally, researchers have emphasized that AI has the potential to exacerbate existing social inequalities by embedding structural disadvantages present in training data or institutional contexts (Zajko, 2022; Yu and Carroll, 2022b; Capraro et al., 2024). These concerns intersect with the broader risk of AI misuseâranging from targeted disinformation campaigns to surveillance and political manipulationâenabled by the scale and efficiency of contemporary AI systems (Anderljung et al., 2024; Yu and Carroll, 2022a). Together, these strands of scholarship identify a comprehensive set of AI-related risks that inform the conceptual framing of this study and underpin the explicit AI risk awareness items used in the survey instrument. 2.2 AI Risk Perception and Adoption Risk perception often diverges between abstract judgments and applied, context-dependent decisions. Foundational work shows that framing, salience, and bounded rationality shape how individuals weigh risks and benefits (Slovic, 2016; Kahneman and Tversky, 2013). Technology-adoption research similarly models how perceived risk moderates intention (Venkatesh et al., 2003; Ajzen, 1991). In education settings, studies report mixed patterns in studentsâ AI risk judgments and adoption tendencies, with gaps between perceived usefulness and concerns about accuracy, bias, and learning effects (Crockett et al., 2020; Oc et al., 2024; Schei et al., 2024). These literatures motivate our use of both explicit and scenario-based measures: the former elicits articulated concerns about named risks; the latter probes applied recognition when risks are not made salient a priori, which prior research suggests can surface different judgments (Slovic, 2016; Kahneman and Tversky, 2013). 2.3 AI Literacy Education in Higher Education Universities increasingly incorporate ethical and societal dimensions into computing curricula, yet implementations vary and often emphasize principles over applied risk judgment (Brown et al., 2024). Empirical studies show that while students can articulate aspects of fairness, accountability, and transparency, recognition of risks in realistic systems is uneven and sensitive to disciplinary identity and context (Kasinidou et al., 2021b, a; Pierson, 2017; Archambault et al., 2024). Policy frameworks likewise foreground responsible AI and critical AI literacy (UNESCO, 2021; European Union, 2024), but evidence-based guidance on integrating scenario-based risk recognition into technical programs remains limited. This background motivates curricula that combine explicit instruction with applied, scenario-centered activities to strengthen transfer from concepts to practice (Kasinidou et al., 2021b; Archambault et al., 2024). 2.4 AI Risk: The Missing Piece of a Tech Curriculum Although AI perceptions and adoption have been widely studied, most existing work examines the general public or broad student populations rather than technology students (Crockett et al., 2020; SĂĄez-Velasco et al., 2025; Tierney et al., 2025). Prior studies often highlight general attitudes toward AI and concerns such as rapid technological change or unemployment, but they rarely assess concrete risk domains such as bias, privacy, or misuse in technically trained groups (Jeffrey, 2020; Ghotbi and Ho, 2021). Research on ChatGPT adoption focuses on usability and credibility while overlooking studentsâ awareness of potential risks (Masaâdeh et al., 2024). As a result, relatively less is known about how different dimensions of AI risk awareness relate to studentsâ willingness to adopt or develop AI systems. This gap is important because technology students are the future developers, engineers, and decision-makers who will influence the ethical direction of AI (Yu et al., 2024a). Understanding their risk awareness is essential for clarifying how such awareness shapes responsible adoption and for determining whether students are prepared to incorporate ethical considerations into system design. Addressing this gap requires empirical study and educational models that explicitly connect risk perception, adoption behavior, and ethical development practices (Mittelstadt, 2019; Floridi, 2010). This study responds to this need by examining technology studentsâ awareness of a range of AI risks and their willingness to adopt AI technologies. The findings are intended to support AI ethics education and curriculum design that better prepare students for responsible AI development. 3 Methodology 3.1 Research Design This study utilized a cross-sectional, mixed-methods survey to investigate technology studentsâ awareness of AI risks and their willingness to adopt AI technologies. The instrument contained four sections: demographic characteristics, explicit AI risk awareness (quantitative), scenario-based evaluation (quantitative), and a short set of open-ended questions (qualitative) designed to elicit studentsâ broader views on AI, perceived risks, and expectations for AI risk education. The full survey instrument is provided in the Supplementary Material. The present manuscript analyzes only the quantitative components, which provide structured measures of explicit and scenario-based risk awareness and adoption willingness. The qualitative responses will be examined and reported in a separate manuscript, where they will offer complementary insight into studentsâ interpretations of AI risks and their perspectives on educational needs. 3.2 Sample and Participants This study targeted students and recent graduates (within one year) enrolled in AI/ML-related graduate programs (e.g., Computer Science, Data Science, Data Analytics) at a private university in the United States. These programs include structured coursework in artificial intelligence and machine learning and therefore provide access to students with substantial exposure to AI technologies. Participants were recruited through multiple channels: (1) distribution of the anonymous Qualtrics survey link during class sessions with instructor permission, (2) the universityâs SONA Research Participation System, which allowed eligible volunteers to complete the study for course credit, (3) direct email invitations sent to students in the target programs, and (4) voluntary snowball sampling in which participants shared the anonymous survey link with peers. All data were collected online, and no referral incentives were offered. A total of 189 responses were recorded between November 19, 2024, and February 19, 2025. After removing 50 incomplete or inconsistent submissions, and with no exclusions based on demographic characteristics, 139 valid responses remained. Of these, 116 participants were enrolled in AI-focused programs (Computer Science: n=48n=48; Data Science/Data Analytics: n=68n=68). An additional 23 respondents were obtained through snowball sampling from programs with varying degrees of AI exposure, including Information Technology, Human-Computer Interaction, Interactive Media, Game Design, Psychology, and Biology. These comparative groups were included to examine how differences in formal AI training relate to risk perception and adoption patterns. Throughout this manuscript, âCSâ refers to Computer Science, âDS/DAâ to Data Science or Data Analytics, and âOtherâ to interdisciplinary or non-technical fields represented in the sample. Demographic profile. The sample was predominantly young, with 63.3% aged 18â24 and 34.5% aged 25â34, and 2.2% aged 35â44. Gender distribution was nearly balanced, with 51.1% identifying as female and 48.9% as male. The majority identified as Asian (77.7%), followed by White or Caucasian (10.8%), Hispanic or Latino (4.3%), Black or African American (2.2%), Mixed or Multiple Ethnicities (2.9%), and 2.2% who preferred not to disclose. Most participants were current graduate students (Masterâs or PhD, 75.5%), with undergraduates comprising 21.6%; recent graduates and working professionals formed a small minority. By specialization, 48.9% reported Data Science/Data Analytics, 34.5% Computer Science, and 16.5% other fields. Regarding experience, 47.5% reported 1â3 years in the technology field, 28.1% less than one year, 21.6% 4â6 years, and 2.9% more than 7 years. Self-rated AI knowledge was generally high: 46.0% moderate, 19.4% proficient, 6.5% expert, 24.5% basic, and 3.6% no understanding. See Table 2 for full details. 3.3 Materials and Measures The survey instrument consisted of four sections: (a) seven demographic items; (b) one matrix-style item assessing explicit awareness across twelve AI risk domains; (c) ten scenario-based items, each measuring both perceived risk and willingness to adopt the technology; and (d) six open-ended qualitative questions, which are analyzed in a separate manuscript. One attention-check item was included to ensure response quality. Explicit AI Risk Awareness. Participants rated their concern for 12 explicitly named AI risk domains (1 = Not concerned at all, 5 = Very concerned). These domains reflect widely documented categories of AI-related risks across AI ethics, governance, and HCI scholarship. Scenario-Based Risk Awareness and Adoption Willingness. Ten realistic AI applications were presented without naming the underlying risks. For each application, participants rated (a) perceived risk and (b) willingness to adopt the system (1 = Not at all, 5 = Very). This design draws on research showing that risk judgments differ when individuals evaluate concrete contexts rather than abstract descriptions (Slovic, 2016; Kahneman and Tversky, 2013). Applications included hiring platforms, healthcare diagnostics, financial advisory tools, public surveillance, manufacturing automation, personalized learning, and smart-home systems. Table 1 provides an overview of how the twelve explicit AI risk domains correspond to the ten scenario-based applications. This mapping clarifies the conceptual alignment between explicit risk categories and their applied manifestations. Table 1: Mapping of Explicit AI Risk Domains to Scenario-Based Survey Items Risk domain Scenario example (Survey item) General AI risks No direct scenario counterpart Privacy / Data security Smart-home assistant learning user habits (Q13) Bias / Fairness Job-matching platform ranking applicants (Q14) Security vulnerabilities AI financial advisory system analyzing spending patterns (Q15) Job displacement AI-driven automation in manufacturing (Q16) Misinformation / Manipulation AI-curated personalized news feeds (Q17) Autonomous weapons Autonomous drones used for automated threat response (Q18) Transparency issues AI diagnostic tool predicting medical conditions (Q19) Ethical implications AI-enabled public surveillance systems (Q20) Psychological / Cognitive impacts AI assistant managing tasks and routines (Q21) Social inequalities Adaptive learning platform personalizing content (Q22) Misuse / Abuse No direct scenario counterpart Note. Scenario descriptions correspond to Survey Items Q13âQ22 in Supplementary Material. Supporting literature for each explicit risk domain: privacy/data security (Yu et al., 2023; Manikonda et al., 2018; NaudĂ©, 2020); bias/fairness (Raghavan et al., 2020; Roselli et al., 2019; Albaroudi et al., 2024); security vulnerabilities (Chen et al., 2019; Manikonda et al., 2018); job displacement (Rawashdeh, 2025; Yu and Carroll, 2023); misinformation (AĂŻmeur et al., 2023; Agarwal et al., 2020; Al-Asadi and Tasdemir, 2021); autonomous weapons (Ăngel GĂłmez de Ăgreda, 2020; Yu and Carroll, 2022a); transparency (Larsson and Heintz, 2020; Von Eschenbach, 2021); ethical implications (Stahl and Stahl, 2021; Benzinger et al., 2023; Saheb, 2023); psychological/cognitive impacts (Zhai et al., 2024; Torres et al., 2024; Yu et al., 2024b; Park and Park, 2024); social inequalities (Zajko, 2022; Yu and Carroll, 2022b; Capraro et al., 2024). General AI risks and Misuse/Abuse appear only as explicit items because they represent broad, cross-cutting concerns that do not map cleanly to a single application context. 3.4 Procedure The survey was administered online via Qualtrics, ensuring anonymity and data security. Participants provided informed consent through an electronic page detailing the studyâs purpose, voluntary nature, and confidentiality. Those declining consent were redirected to a closing page. In-class participants could opt for an alternative task, and SONA participants received course credits. A quality check question identified inattentive responses. This study was approved by the Institutional Review Board at Clark University (Protocol No. 701). All participants provided informed consent prior to participation. No personally identifiable information was collected. 3.5 Validation The instrument underwent a two-stage validation process prior to deployment. First, three faculty members with expertise in AI ethics and human-computer interaction reviewed all items for clarity, relevance, and completeness. Their feedback informed revisions to item wording and alignment across risk domains. Second, a small cognitive pilot with 12 technology students was conducted to evaluate item comprehension, survey flow, and response burden. Minor adjustments were made based on participantsâ comments before releasing the final instrument. 3.6 Data Analysis Quantitative data were analyzed using SPSS (v29). Data preparation included screening for completeness and removal of patterned or low-quality responses. Normality was assessed using Shapiro-Wilk tests. Reliability of the survey scales was evaluated using Cronbachâs alpha. Descriptive statistics were calculated for all key variables. Group comparisons by gender and specialization were conducted using independent-samples t-tests and one-way ANOVA for normally distributed variables, and Mann-Whitney U and Kruskal-Wallis tests for variables violating normality assumptions. This selection ensures that the statistical tests align appropriately with the distributional characteristics of each variable. The relationship between risk awareness and adoption willingness was assessed using Pearson or Spearman correlation coefficients, depending on normality. Detailed results of reliability analysis, descriptive statistics, and inferential tests are reported in the Results section. 4 Results 4.1 Demographic Overview Table 2 summarizes the demographic characteristics of the final sample (N=139N=139), including age, gender, ethnicity, student status, specialization, years of experience, and self-rated AI knowledge. Descriptive information is further detailed in Section 3.2. These distributions provide important contextual information for interpreting subsequent analyses. Table 2: Demographic Information of Participants (N=139N=139) Count % Age 18â24 years 88 63.3 25â34 years 48 34.5 35â44 years 3 2.2 Gender Male 68 48.9 Female 71 51.1 Ethnicity Asian 108 77.7 Black or African American 3 2.2 Hispanic or Latino 6 4.3 White or Caucasian 15 10.8 Mixed or Multiple Ethnicities 4 2.9 Prefer not to say 3 2.2 Status Current Undergraduate Student 30 21.6 Current Graduate Student 105 75.5 Recent Graduate 3 2.2 Working Professional 1 0.7 Specialization Computer Science 48 34.5 Data Science / Data Analytics 68 48.9 Other 23 16.5 Experience Less than 1 year 39 28.1 1â3 years 66 47.5 4â6 years 30 21.6 7+ years 4 2.9 AI Knowledge Not understanding 5 3.6 Basic understanding 34 24.5 Moderate understanding 64 46.0 Proficient understanding 27 19.4 Expert 9 6.5 1. Note: Percentages may not sum to 100 due to rounding. 4.2 Reliability and Descriptive Statistics Table 3 reports reliability and descriptive statistics for the three composite scales. All scales demonstrated acceptable to excellent reliability (Explicit: α=.93α=.93; Scenario-based: α=.88α=.88; Adoption willingness: α=.86α=.86). Composite mean scores were used for all subsequent analyses. Effect sizes (Cohenâs d for t-tests and r for non-parametric tests) were computed to complement p-values and provide estimates of practical significance. Table 3: Reliability and Descriptive Statistics for Survey Scales Scale Explicit Risk Scenario Risk Adoption Willingness Items 12 10 10 Cronbachâs α 0.93 0.88 0.86 Mean (SD) 3.70 (0.29) 3.10 (0.35) 3.40 (0.28) Item mean range 2.99â4.01 2.39â3.46 2.95â3.86 Inter-item correlations 0.30â0.76 0.18â0.66 0.09â0.64 Item-total correlations 0.50â0.77 0.48â0.70 0.39â0.68 4.3 Comparative Analysis Comparisons were conducted across gender and academic specialization for explicit awareness, scenario-based awareness, and adoption willingness. 4.3.1 Explicit AI Risk Awareness Descriptive patterns across risk categories by gender and specialization are presented in Table 4. Visual patterns of specialization differences are further illustrated in Figure 1, which highlights higher explicit risk awareness among CS and DS/DA students relative to the Other group. Explicit AI risk awareness showed no gender difference, U=2315.50U=2315.50, p=.678p=.678, d=0.14d=0.14, 95% CI [â0.198, 0.468][-0.198,\,0.468]. Specialization differences were significant, KruskalâWallis Ï2â(2)=13.91Ï^2(2)=13.91, p<.001p<.001. Post-hoc comparisons indicated: CS >> Other (p<.001p<.001, r=.41r=.41, 95% CI [.018, .191][.018,\,.191]); DS/DA >> Other (p=.014p=.014, r=.26r=.26, 95% CI [.004, .179][.004,\,.179]). Table 4: Mean (SD) of Explicit AI Risk Awareness by Gender and Specialization (N=139N=139) Risk Category Male Female CS DS/DA Other Total General AI risks 3.04 (1.20) 2.93 (1.19) 3.02 (1.30) 3.09 (1.09) 2.61 (1.23) 2.99 (1.19) Privacy/data security 4.19 (1.03) 3.77 (1.31) 4.25 (1.18) 3.97 (1.04) 3.43 (1.50) 3.98 (1.20) Bias/fairness 3.35 (1.09) 3.39 (1.29) 3.56 (1.40) 3.38 (1.05) 2.96 (1.07) 3.37 (1.19) Security vulnerabilities 3.87 (1.11) 3.80 (1.20) 4.10 (1.13) 3.84 (1.09) 3.26 (1.21) 3.83 (1.15) Job displacement 3.76 (1.28) 3.62 (1.35) 4.08 (1.15) 3.62 (1.27) 3.09 (1.54) 3.69 (1.31) Misinformation 3.91 (1.12) 3.99 (1.17) 4.08 (1.13) 3.87 (1.18) 3.91 (1.04) 3.95 (1.14) Autonomous weapons 4.06 (1.05) 3.58 (1.33) 4.08 (1.20) 3.99 (1.07) 2.74 (1.14) 3.81 (1.22) Transparency issues 3.90 (1.05) 3.72 (1.20) 4.10 (1.04) 3.79 (1.11) 3.22 (1.17) 3.81 (1.13) Ethical implications 3.84 (1.03) 3.73 (1.16) 4.10 (1.04) 3.76 (1.02) 3.17 (1.19) 3.78 (1.10) Psychological impacts 3.65 (1.28) 3.61 (1.35) 4.04 (1.22) 3.51 (1.26) 3.09 (1.41) 3.63 (1.31) Social inequalities 3.53 (1.13) 3.56 (1.30) 3.96 (1.18) 3.46 (1.13) 2.96 (1.26) 3.55 (1.21) Misuse/abuse 4.04 (1.23) 3.99 (1.25) 4.33 (1.14) 4.03 (1.17) 3.30 (1.36) 4.01 (1.23) Figure 1: Explicit AI Risk Awareness by Academic Specialization. CS and DS/DA students showed significantly higher scores than the Other group. 4.3.2 Scenario-Based AI Risk Awareness Scenario-based risk ratings for each application context are shown in Table 5. These results indicate substantial variability across scenarios but limited differences across demographic groups. Gender difference: tâ(137)=â0.23t(137)=-0.23, p=.819p=.819, d=â0.06d=-0.06, 95% CI [â0.390, 0.275][-0.390,\,0.275]. Specialization: non-significant, KruskalâWallis Ï2â(2)=3.83Ï^2(2)=3.83, p=.148p=.148, η2=.028η^2=.028, 95% CI [â0.015, 0.078][-0.015,\,0.078]. Table 5: Mean (SD) of Scenario-Based AI Risk Awareness by Gender and Specialization (N=139N=139) Risk Category Male Female CS DS/DA Other Total Personal privacy 2.96 (1.43) 3.32 (1.20) 3.13 (1.48) 3.34 (1.30) 2.22 (1.09) 3.14 (1.33) Bias 3.12 (1.47) 3.08 (1.42) 3.25 (1.60) 3.07 (1.34) 2.87 (1.39) 3.10 (1.44) Security vulnerability 3.09 (1.32) 3.06 (1.34) 3.35 (1.36) 2.81 (1.34) 3.26 (1.10) 3.07 (1.33) Job displacement 2.79 (1.40) 2.85 (1.25) 3.02 (1.42) 2.76 (1.33) 2.57 (1.04) 2.82 (1.32) Misinformation 3.41 (1.40) 3.18 (1.41) 3.75 (1.28) 3.07 (1.40) 3.00 (1.48) 3.29 (1.40) Autonomous weapons 3.40 (1.39) 3.49 (1.42) 3.54 (1.44) 3.32 (1.32) 3.61 (1.41) 3.45 (1.37) Transparency 3.04 (1.40) 3.28 (1.39) 3.38 (1.38) 2.85 (1.39) 3.65 (1.27) 3.17 (1.39) Misuse/abuse 2.76 (1.39) 3.03 (1.43) 2.88 (1.44) 2.63 (1.38) 3.74 (1.36) 2.90 (1.41) Psychological impacts 2.44 (1.30) 2.32 (1.19) 2.38 (1.33) 2.29 (1.23) 2.65 (1.07) 2.38 (1.24) Social inequalities 2.47 (1.38) 2.39 (1.22) 2.69 (1.52) 2.21 (1.17) 2.57 (1.08) 2.43 (1.30) Composite Score 2.96 (0.99) 3.00 (0.91) 3.14 (1.05) 2.83 (0.90) 3.11 (0.83) 2.98 (0.95) 4.3.3 Scenario-Based Adoption Willingness Figure 2 provides a comparison of scenario-specific risk awareness and adoption willingness across all ten applications, illustrating the inverse trend between perceived risk and willingness to adopt. Figure 2: Scenario-Based Risk Awareness and Adoption Willingness Across Application Contexts. A clear inverse relationship appears: scenarios rated as higher risk consistently show lower willingness to adopt, whereas lower-risk scenarios show higher willingness. Scenario-level willingness-to-adopt scores are reported in Table 6. Figure 3 provides a visualization of group differences, demonstrating that CS and DS/DA students exhibit notably higher adoption willingness than students from other fields. Gender difference present: tâ(137)=2.02t(137)=2.02, p=.046p=.046, d=0.34d=0.34, 95% CI [0.007, 0.677][0.007,\,0.677]. Specialization differences significant: ANOVA Fâ(2,136)=9.85F(2,136)=9.85, p<.001p<.001, η2=.127η^2=.127, 95% CI [.035, .226][.035,\,.226]. Table 6: Scenario-Based AI Risk Awareness and Adoption Willingness (N=139N=139) Risk Type Risk Awareness Adoption Willingness Mean SD Mean SD Personal Privacy 3.14 1.33 3.44 1.19 Bias 3.10 1.44 3.31 1.23 Security Vulnerability 3.07 1.33 3.22 1.25 Job displacement 2.82 1.32 3.55 1.22 Misinformation 3.29 1.40 3.17 1.32 Autonomous weapons 3.45 1.37 2.95 1.41 Transparency 3.17 1.39 3.28 1.30 Misuse/abuse 2.90 1.44 3.42 1.31 Psychological impacts 2.38 1.24 3.83 1.15 Social inequalities 2.43 1.30 3.86 1.21 Figure 3: Scenario-Based Adoption Willingness by Academic Specialization. CS and DS/DA students reported significantly higher willingness than students in Other disciplines. 4.3.4 Explicit vs. Scenario-Based AI Risk Awareness Results of the Wilcoxon signed-rank tests comparing explicit and scenario-based risk ratings are presented in Table 7. These findings confirm consistent and statistically significant differences between explicit and contextualized evaluations across nearly all matched domains. Explicit awareness (M=3.70M=3.70) significantly exceeded scenario-based awareness (M=2.98M=2.98) for nearly all risks, with largest effects for psychological impacts and social inequalities. Table 7: Wilcoxon Signed-Rank Test: Explicit vs. Scenario-Based AI Risk Awareness Risk/Scenario Z p Privacy in automation -5.409 <.001<.001 Bias in hiring -1.822 .068 Security in banking -5.074 <.001<.001 Job displacement in manufacturing -5.858 <.001<.001 Manipulation in social media -4.616 <.001<.001 Autonomous weapons in drones -2.211 .027 Transparency in healthcare -3.980 <.001<.001 Misuse in mass surveillance -6.061 <.001<.001 Psychological impacts of assistants -7.200 <.001<.001 Social inequalities in education -6.730 <.001<.001 5 Discussion Before interpreting the findings, it is important to note that the studyâs cross-sectional design permits only correlational inferences; causal relationships between AI risk perception and adoption willingness cannot be established. 5.1 Key Findings 5.1.1 Explicit Risks Elicit Stronger Concern than Scenario-Based Risks The results reveal a critical divergence in technology studentsâ AI risk perception: explicitly stated risks elicit substantially stronger concern than risks embedded within specific application scenarios. A clear example appears in the domain of privacy: students rated explicit privacy/data security risks highly (M=3.98M=3.98), yet their concern declined markedly (M=3.14M=3.14) when the same risk was embedded in a realistic smart-home scenario. Across the full sample, the overall mean of explicit AI risk ratings was 3.703.70, notably higher than the 3.103.10 observed for scenario-based AI risk ratings. These findings suggest that students recognize and articulate AI risks more strongly when risks are directly named and conceptually framed, a pattern consistent with research showing that abstract descriptions can activate higher-level evaluative reasoning (Trope and Liberman, 2010). However, this awareness weakens when those same risks are presented in concrete, operational contexts. In short, explicit risk knowledge does not fully transfer to contextualized risk recognition, highlighting a gap between abstract understanding and applied judgment. 5.1.2 Perceived Risk and Adoption Willingness Are Inversely Related The results provide strong evidence for an inverse relationship between perceived contextual risk and willingness to adopt AI technologies. Across all scenarios, students showed the lowest adoption willingness in domains they perceived as high-riskâmost notably autonomous weapons (Mrisk=3.45M_risk=3.45, Madopt=2.95M_adopt=2.95) and misinformation/manipulation (Mrisk=3.29M_risk=3.29, Madopt=3.17M_adopt=3.17). In contrast, the highest adoption willingness occurred in domains perceived as lower-risk, such as psychological impacts (Mrisk=2.38M_risk=2.38, Madopt=3.83M_adopt=3.83) and social inequalities (Mrisk=2.43M_risk=2.43, Madopt=3.86M_adopt=3.86). This consistent pattern demonstrates that adoption judgments are shaped less by general attitudes toward AI and more by scenario-specific risk cues. Taken together, the findings show that lower perceived risk reliably predicts greater willingness to adopt AI, even within the same student population and using parallelized scenario formats. This inverse association highlights the need for AI literacy and ethics education to incorporate contextualized risk assessments rather than relying solely on general or abstract discussions of AI safety. 5.1.3 Education Equalizes Risk Awareness but Gender Differences Persist in Adoption The results show that formal technical education effectively equalizes gender differences in AI risk perception. Male and female students demonstrated statistically equivalent levels of explicit risk awareness (p=.678p=.678) and scenario-based risk awareness (p=.819p=.819), indicating that domain-specific instruction provides a shared framework for evaluating AI-related harms. Despite this convergence in risk perception, a clear behavioral divergence emerged. Male students reported significantly higher willingness to adopt AI technologies than female students (Mmale=3.55M_male=3.55 vs. Mfemale=3.26M_female=3.26, p=.046p=.046, d=0.34d=0.34). This suggests that equalizing cognitive understanding of risks does not necessarily translate into equal adoption behavior. This dissociation between awareness and behavior aligns with prior research showing that gender differences in uncertainty tolerance, trust, and perceived behavioral control can influence technology-related decisions (Venkatesh et al., 2003; Byrnes et al., 1999; Gefen and Straub, 1997). Taken together, these findings highlight that while technical education may narrow gender gaps in evaluative judgments, it does not eliminate underlying psychological or sociocultural factors that shape willingness to adopt AI. As such, AI literacy initiatives should complement technical instruction with strategies that address these behavioral dimensions to support equitable engagement with AI technologies. 5.1.4 Specialization Differences Reveal Evidence of Applied Risk Underappreciation A clear pattern emerged when comparing AI-related specializations with students from non-technical fields. Computer Science (CS) and Data Science/Data Analytics (DS/DA) students demonstrated higher explicit AI risk awareness than students in the âOtherâ group, consistent with their stronger curricular exposure to AI systems. However, these same CS and DS/DA students did not show correspondingly higher recognition of risks when evaluating scenario-based applications, where their average scenario-awareness scores were comparable to or only marginally higher than those of non-technical peers. Despite this parallel in scenario-based risk perception, CS and DS/DA students expressed substantially greater willingness to adopt AI technologies across nearly all application contexts. Students from non-technical fields, by contrast, showed markedly lower adoption willingness, even when their scenario-based risk assessments were similar. This divergence suggests that technical specialization strengthens confidence and perceived control over AI systems, which in turn elevates willingness to adopt AI regardless of contextualized risk cues. Taken together, these results point to a form of risk underappreciation among students in AI-related fields: they exhibit high explicit awareness of AI risks yet demonstrate muted sensitivity to those same risks when embedded in applied scenarios, while simultaneously showing strong enthusiasm for adoption. A plausible mechanism underlying this pattern is that greater domain familiarity can increase perceived control and reduce sensitivity to potential harms, a tendency documented in research showing that expertise often promotes overconfidence and higher risk-taking (Hilary and Menzly, 2006; Said et al., 2023). The present findings extend this work by providing quantitative evidence of risk underappreciation within the context of university-level AI education. 5.2 Implications and Practical Recommendations The four key findings of this study collectively reveal several important implications for AI education, curriculum design, and responsible AI practice. These implications extend beyond the empirical patterns reported earlier and point to concrete opportunities for strengthening studentsâ ability to evaluate and engage with AI systems responsibly. In addition to these empirical findings, the study also contributes to broader work on AI risk perception and AI literacy. First, it offers a combined analysis of explicit and scenario-based AI risk awareness within the same population, enabling a direct comparison of articulated versus contextualized judgmentsâan approach that remains relatively uncommon in existing research. Second, it provides empirical evidence of how perceived contextual risk relates to adoption willingness across different application domains. Third, it highlights how gender and disciplinary identity shape adoption behavior even when risk awareness itself is comparable. Together, these contributions help refine theoretical understanding of AI risk perception and guide the development of more effective AI risk education strategies. 1. Strengthening Transfer Between Abstract Risk Knowledge and Applied Contexts Section 5.1.1 showed that students recognize AI risks more strongly when they are explicitly labeled than when the same risks are embedded in realistic scenarios. This gap suggests that existing AI curricula may succeed in conveying conceptual terminology but fall short in preparing students to detect risks in operational settings. To address this disconnect, educational programs should integrate scenario-based reasoning throughout instructionâfor example, by embedding case analyses, incident walkthroughs, and structured reflection exercises that require students to identify risks without being prompted by explicit labels. These strategies can support the transfer of abstract risk knowledge into applied judgment, a capacity essential for responsible AI development. 2. Embedding Contextualized Risk Evaluation Into AI Literacy Section 5.1.2 demonstrated that adoption willingness closely follows contextual risk perception, highlighting the importance of teaching students how to evaluate risks not just at the conceptual level but within specific application domains. To cultivate more calibrated decision-making, AI ethics modules should incorporate domain-specific exemplars across health, security, education, entertainment, and workplace contexts. Asking students to articulate both the potential harms and justifications for adoption in each context can strengthen their ability to weigh domain-relevant trade-offs and avoid over-reliance on general attitudes toward AI. 3. Addressing Behavioral Disparities Through Psychological and Sociocultural Support Although section 5.1.3 showed that technical education equalized gender differences in risk perception, it also revealed that gender gaps persist in adoption behaviors. This indicates that cognitive understanding alone is insufficient for equitable engagement. AI curricula should therefore integrate interventions that address psychological and sociocultural factors shaping adoption decisions. Activities that build self-efficacy, reduce uncertainty aversion, and strengthen calibrated trust in AI systems may help reduce gender-linked behavioral disparities. Importantly, such interventions should be framed not as remediation but as part of a broader commitment to supporting diverse forms of engagement with AI. 4. Expanding AI Literacy Beyond Technical Majors to Counteract Overconfidence and Vulnerability Section 5.1.4 highlighted a dual challenge: students in AI-related majors demonstrate strong explicit awareness yet show muted sensitivity to contextualized risks and elevated willingness to adopt AI technologies. Meanwhile, students in non-technical fields show lower adoption willingness despite comparable scenario-based awareness. These patterns indicate that both groups would benefit from tailored educational strategies. For technical majors, instruction should emphasize critical self-reflection and highlight the limits of expertise to reduce overconfidence and mitigate forms of risk underappreciation. For non-technical students, AI literacy programs should focus on building foundational understanding and confidence, ensuring they are not marginalized in technology discourse or decision-making. 5. Building Institution-Level Frameworks for Responsible AI Education Across section 5.1.1â 5.1.4, a consistent theme emerges: students require structured opportunities to integrate conceptual understanding, contextual reasoning, and reflective judgment. Institutions can support this integration by adopting program-wide frameworks that map AI risk competencies across curricula, ensuring that explicit instruction, scenario-based reasoning, and reflective activities appear throughout a studentâs academic trajectory. Periodic needs assessments, informed by student feedback and evolving AI technologies, can help institutions adapt AI risk education to diverse learner profiles and emerging societal challenges. 5.3 Limitations of the Study While this study provides meaningful insights into technology studentsâ AI risk awareness and adoption behavior, several limitations should be acknowledged to contextualize the findings. First, the data were collected from a single private university in the United States, which limits generalizability beyond similar institutional and disciplinary contexts. Participation was voluntary, which may have attracted students with higher interest in or familiarity with AI. As a result, the levels of risk awareness and adoption willingness observed in this sample may not fully reflect those of the broader student population. Second, although the sample included students from Computer Science, Data Science/Data Analytics, and a range of other majors, the âOtherâ group was comparatively small. This limits the ability to draw strong conclusions about non-technical fields and constrains the interpretation of disciplinary differences. Future research with more balanced representation across academic disciplines would strengthen the robustness of these comparisons. Third, the study relied on self-reported measures of risk perception, self-assessed AI knowledge, and adoption intentions. Such measures are susceptible to social desirability bias and may not fully correspond to actual behavior or objective proficiency. Incorporating performance-based assessments, behavioral tasks, or multimodal measures would help validate and extend the present findings. Fourth, the cross-sectional design restricts interpretation to correlational relationships. Because all variables were measured at a single time point, it is not possible to infer causal pathways between AI risk perception and adoption willingness, nor to examine how these perceptions evolve as students progress through their academic programs or gain professional experience. Longitudinal or repeated-measures studies would allow for analysis of developmental trajectories and provide stronger evidence on the impact of educational interventions. These limitations highlight the need for cautious interpretation of the findings and indicate clear opportunities for future research to build on and expand the present work. 6 Conclusion As AI technologies become increasingly integrated into society, understanding how future professionals evaluate AI-related risks is essential. This study examined technology studentsâ explicit and scenario-based AI risk awareness, their adoption willingness across application domains, and the influence of gender and disciplinary background on these judgments. Four consistent patterns emerged. First, students expressed substantially stronger concern for explicitly stated AI risks than for the same risks embedded in realistic scenarios, indicating that abstract knowledge does not readily translate into contextual risk recognition. Second, adoption willingness declined systematically as perceived contextual risk increased, demonstrating that studentsâ acceptance of AI systems is closely tied to their appraisal of domain-specific risks. Third, although technical training appeared to equalize gender differences in risk awareness, male students consistently reported higher willingness to adopt AI technologies, revealing a behavioral gap not explained by differences in evaluative judgment. Fourth, students in AI-intensive majors such as Computer Science and Data Science showed both higher explicit risk awareness and higher adoption willingness than their non-technical peersâa pattern consistent with the form of ârisk underappreciationâ in which technical familiarity coexists with reduced caution in applied contexts. Together, these findings show that technology studentsâ engagement with AI is shaped by both how risks are framed and who is evaluating them. Strengthening responsible AI practice will therefore require educational approaches that pair explicit instruction with scenario-based reasoning, address behavioral and psychological dimensions of adoption, and support more calibrated decision-making across disciplinary groups. Such efforts will be critical as AI continues to expand across academic and professional environments. Funding Statement This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Declaration of AI Assisted Writing During the preparation of this work the authors used ChatGPT (OpenAI) in order to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication. Data Availability The datasets generated and analyzed during the current study are not publicly available due to participant confidentiality but are available from the corresponding author on reasonable request. Ethics Approval and Consent to Participate This study was approved by the Institutional Review Board at Clark University (Protocol No. 701). All participants provided informed consent prior to participation. References Agarwal et al. (2020) Agarwal, S., Farid, H., El-Gaaly, T., Lim, S.N., 2020. Detecting deep-fake videos from appearance and behavior, in: 2020 IEEE International Workshop on Information Forensics and Security (WIFS), p. 1â6. doi:10.1109/WIFS49906.2020.9360904. AĂŻmeur et al. (2023) AĂŻmeur, E., Amri, S., Brassard, G., 2023. Fake news, disinformation and misinformation in social media: a review. Social Network Analysis and Mining 13, 30. URL: https://doi.org/10.1007/s13278-023-01028-5, doi:10.1007/s13278-023-01028-5. Ajzen (1991) Ajzen, I., 1991. The theory of planned behavior. Organizational Behavior and Human Decision Processes 50, 179â211. URL: https://w.sciencedirect.com/science/article/pii/074959789190020T, doi:10.1016/0749-5978(91)90020-T. theories of Cognitive Self-Regulation. Al-Asadi and Tasdemir (2021) Al-Asadi, M.A., Tasdemir, S., 2021. Using artificial intelligence against the phenomenon of fake news: A systematic literature review. Combating Fake News with Computational Intelligence Techniques , 39â54URL: https://doi.org/10.1007/978-3-030-90087-8_2, doi:10.1007/978-3-030-90087-8_2. Albaroudi et al. (2024) Albaroudi, E., Mansouri, T., Alameer, A., 2024. A comprehensive review of ai techniques for addressing algorithmic bias in job hiring. AI 5, 383â404. URL: https://w.mdpi.com/2673-2688/5/1/19, doi:10.3390/ai5010019. Anderljung et al. (2024) Anderljung, M., Hazell, J., Von Knebel, M., 2024. Protecting society from AI misuse: when are restrictions on capabilities warranted? AI & Society , 1â17URL: https://doi.org/10.1007/s00146-024-02130-8, doi:10.1007/s00146-024-02130-8. Archambault et al. (2024) Archambault, S.G., Ramachandran, S., Acosta, E., Fu, S., 2024. Ethical dimensions of algorithmic literacy for college students: Case studies and cross-disciplinary connections. The Journal of Academic Librarianship 50, 102865. URL: https://w.sciencedirect.com/science/article/pii/S0099133324000260, doi:10.1016/j.acalib.2024.102865. Benzinger et al. (2023) Benzinger, L., Ursin, F., Balke, W.T., Kacprowski, T., Salloch, S., 2023. Should artificial intelligence be used to support clinical ethical decision-making? A systematic review of reasons. BMC Medical Ethics 24, 48. URL: https://doi.org/10.1186/s12910-023-00929-6, doi:10.1186/s12910-023-00929-6. Bernal and Mazo (2022) Bernal, J., Mazo, C., 2022. Transparency of artificial intelligence in healthcare: Insights from professionals in computing and healthcare worldwide. Applied Sciences 12. URL: https://w.mdpi.com/2076-3417/12/20/10228, doi:10.3390/app122010228. Brauner et al. (2023) Brauner, P., Hick, A., Philipsen, R., Ziefle, M., 2023. What does the public think about artificial intelligence?âa criticality map to understand bias in the public perception of ai. Frontiers in Computer Science Volume 5 - 2023. URL: https://w.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2023.1113903, doi:10.3389/fcomp.2023.1113903. Brown et al. (2024) Brown, N., Xie, B., Sarder, E., Fiesler, C., Wiese, E.S., 2024. Teaching ethics in computing: A systematic literature review of acm computer science education publications. ACM Trans. Comput. Educ. 24. URL: https://doi.org/10.1145/3634685, doi:10.1145/3634685. Byrnes et al. (1999) Byrnes, J.P., Miller, D.C., Schafer, W.D., 1999. Gender differences in risk taking: A meta-analysis. Psychological Bulletin 125, 367â383. URL: https://psycnet.apa.org/fulltext/1999-13573-004.html. Capraro et al. (2024) Capraro, V., Lentsch, A., Acemoglu, D., Akgun, S., Akhmedova, A., Bilancini, E., Bonnefon, J.F., Brañas-Garza, P., Butera, L., Douglas, K.M., Everett, J.A.C., Gigerenzer, G., Greenhow, C., Hashimoto, D.A., Holt-Lunstad, J., Jetten, J., Johnson, S., Kunz, W.H., Longoni, C., Lunn, P., Natale, S., Paluch, S., Rahwan, I., Selwyn, N., Singh, V., Suri, S., Sutcliffe, J., Tomlinson, J., van der Linden, S., Van Lange, P.A.M., Wall, F., Van Bavel, J.J., Viale, R., 2024. The impact of generative artificial intelligence on socioeconomic inequalities and policy making. PNAS Nexus 3, pgae191. doi:10.1093/pnasnexus/pgae191. Chen et al. (2019) Chen, T., Liu, J., Xiang, Y., Niu, W., Tong, E., Han, Z., 2019. Adversarial attack and defense in reinforcement learning-from AI security view. Cybersecurity 2, 1â22. URL: https://doi.org/10.1186/s42400-019-0027-x, doi:10.1186/s42400-019-0027-x. Crockett et al. (2020) Crockett, K., Garratt, M., Latham, A., Colyer, E., Goltz, S., 2020. Risk and trust perceptions of the public of artifical intelligence applications, in: 2020 International Joint Conference on Neural Networks (IJCNN), p. 1â8. doi:10.1109/IJCNN48605.2020.9207654. Delello et al. (2023) Delello, J.A., Sung, W., Mokhtari, K., De Giuseppe, T., 2023. Exploring college studentsâ awareness of AI and ChatGPT: Unveiling perceived benefits and risks. Journal of Inclusive Methodology and Technology in Learning and Teaching 3, 1â25. URL: https://w.inclusiveteaching.it/index.php/inclusiveteaching/article/view/132, doi:10.32043/jimtlt.v3i4.132. European Union (2024) European Union, 2024. Artificial intelligence act. https://artificialintelligenceact.eu/ai-act-explorer/. Floridi (2010) Floridi, L., 2010. Ethics after the information revolution, in: Floridi, L. (Ed.), The Cambridge Handbook of Information and Computer Ethics. Cambridge University Press, Cambridge, p. 3â19. URL: https://doi.org/10.1017/CBO9780511845239.002, doi:10.1017/CBO9780511845239.002. Gefen and Straub (1997) Gefen, D., Straub, D.W., 1997. Gender differences in the perception and use of e-mail: An extension to the technology acceptance model. MIS Quarterly , 389â400. Ghotbi and Ho (2021) Ghotbi, N., Ho, M.T., 2021. Moral awareness of college students regarding artificial intelligence. Asian Bioethics Review 13, 421â433. URL: https://doi.org/10.1007/s41649-021-00182-2, doi:10.1007/s41649-021-00182-2. Ăngel GĂłmez de Ăgreda (2020) Ăngel GĂłmez de Ăgreda, 2020. Ethics of autonomous weapons systems and its applicability to any ai systems. Telecommunications Policy 44, 101953. URL: https://w.sciencedirect.com/science/article/pii/S0308596120300458, doi:10.1016/j.telpol.2020.101953. artificial intelligence, economy and society. Hilary and Menzly (2006) Hilary, G., Menzly, L., 2006. Does past success lead analysts to become overconfident? Review of Financial Studies 19, 1â31. URL: https://doi.org/10.1287/mnsc.1050.0485, doi:10.1287/mnsc.1050.0485. Jeffrey (2020) Jeffrey, T., 2020. Understanding college student perceptions of artificial intelligence. Systemics, Cybernetics and Informatics 18, 8â13. URL: https://w.iiisci.org/Journal/PDV/sci/pdfs/HB785N20.pdf. Kahneman and Tversky (2013) Kahneman, D., Tversky, A., 2013. Prospect theory: An analysis of decision under risk, in: Handbook of the fundamentals of financial decision making: Part I. World Scientific, p. 99â127. URL: https://w.worldscientific.com/doi/abs/10.1142/9789814417358_0006, doi:10.1142/9789814417358_0006. Kasinidou et al. (2021a) Kasinidou, M., Kleanthous, S., Barlas, P., Otterbacher, J., 2021a. I agree with the decision, but they didnât deserve this: Future developersâ perception of fairness in algorithmic decisions, in: Proceedings of the 2021 acm conference on fairness, accountability, and transparency, p. 690â700. URL: https://doi.org/10.1145/3442188.3445931, doi:10.1145/3442188.3445931. Kasinidou et al. (2021b) Kasinidou, M., Kleanthous, S., Orphanou, K., Otterbacher, J., 2021b. Educating computer science students about algorithmic fairness, accountability, transparency and ethics, in: Proceedings of the 26th ACM Conference on Innovation and Technology in Computer Science Education V. 1, p. 484â490. URL: https://doi.org/10.1145/3430665.3456311, doi:10.1145/3430665.3456311. Larsson and Heintz (2020) Larsson, S., Heintz, F., 2020. Transparency in artificial intelligence. Internet Policy Review 9, 1â16. URL: https://lup.lub.lu.se/search/files/79208055/Larsson_Heintz_2020_Transparency_in_artificial_intelligence_2020_05_05.pdf, doi:10.14763/2020.2.1469. Lee et al. (2023) Lee, P., Bubeck, S., Petro, J., 2023. Benefits, limits, and risks of GPT-4 as an AI chatbot for medicine. New England Journal of Medicine 388, 1233â1239. URL: https://w.nejm.org/doi/full/10.1056/NEJMsr2214184, doi:10.1056/NEJMsr2214184, arXiv:https://w.nejm.org/doi/pdf/10.1056/NEJMsr2214184. Li et al. (2025) Li, Y., Castulo, N.J., Xu, X., 2025. Embracing or rejecting AI? A mixed-method study on undergraduate studentsâ perceptions of artificial intelligence at a private university in China, in: Frontiers in Education, Frontiers Media SA. p. 1505856. URL: https://w.frontiersin.org/journals/education/articles/10.3389/feduc.2025.1505856, doi:10.3389/feduc.2025.1505856. Manikonda et al. (2018) Manikonda, L., Deotale, A., Kambhampati, S., 2018. Whatâs up with privacy? User preferences and privacy concerns in intelligent personal assistants, in: Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, p. 229â235. URL: https://dl.acm.org/doi/abs/10.1145/3278721.3278773, doi:10.1145/3278721.3278773. Masaâdeh et al. (2024) Masaâdeh, R., Majali, S.A., Alkhaffaf, M., Thurasamy, R., Almajali, D., Altarawneh, K., Al-Sherideh, A., Altarawni, I., 2024. Antecedents of adoption and usage of ChatGPT among Jordanian university students: Empirical study. International Journal of Data and Network Science 8, 1099â1110. URL: https://w.growingscience.com/ijds/Vol8/ijdns_2023_224.pdf, doi:10.5267/j.ijdns.2023.11.024. Mittelstadt (2019) Mittelstadt, B., 2019. Principles alone cannot guarantee ethical AI. Nature Machine Intelligence 1, 501â507. URL: https://w.nature.com/articles/s42256-019-0114-4, doi:10.1038/s42256-019-0114-4. NaudĂ© (2020) NaudĂ©, W., 2020. Artificial intelligence vs COVID-19: Limitations, constraints and pitfalls. AI & Society 35, 761â765. URL: https://link.springer.com/article/10.1007/S00146-020-00978-0, doi:10.1007/S00146-020-00978-0. Oc et al. (2024) Oc, Y., Gonsalves, C., Quamina, L.T., 2024. Generative AI in higher education assessments: Examining risk and tech-savviness on studentâs adoption. Journal of Marketing Education , 1â18URL: https://journals.sagepub.com/doi/10.1177/02734753241302459, doi:10.1177/02734753241302459. Park and Park (2024) Park, H.W., Park, S., 2024. The filter bubble generated by artificial intelligence algorithms and the network dynamics of collective polarization on YouTube: The case of South Korea. Asian Journal of Communication 34, 195â212. URL: https://doi.org/10.1080/01292986.2024.2315584, doi:10.1080/01292986.2024.2315584. Pierson (2017) Pierson, E., 2017. Demographics and discussion influence views on algorithmic fairness. arXiv preprint arXiv:1712.09124 URL: https://doi.org/10.48550/arXiv.1712.09124, doi:10.48550/arXiv.1712.09124. Raghavan et al. (2020) Raghavan, M., Barocas, S., Kleinberg, J., Levy, K., 2020. Mitigating bias in algorithmic hiring: Evaluating claims and practices, in: Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, p. 469â481. URL: https://dl.acm.org/doi/abs/10.1145/3351095.3372828, doi:10.1145/3351095.3372828. Rawashdeh (2025) Rawashdeh, A., 2025. The consequences of artificial intelligence: An investigation into the impact of AI on job displacement in accounting. Journal of Science and Technology Policy Management 16, 506â535. URL: https://doi.org/10.1108/JSTPM-02-2023-0030, doi:10.1108/JSTPM-02-2023-0030, arXiv:https://w.emerald.com/jstpm/article-pdf/16/3/506/9659776/jstpm-02-2023-0030.pdf. Roselli et al. (2019) Roselli, D., Matthews, J., Talagala, N., 2019. Managing bias in AI, in: Companion Proceedings of the 2019 World Wide Web Conference, p. 539â544. URL: https://dl.acm.org/doi/abs/10.1145/3308560.3317590, doi:10.1145/3308560.3317590. SĂĄez-Velasco et al. (2025) SĂĄez-Velasco, S., Alaguero-RodrĂguez, M., RodrĂguez-Cano, S., Delgado-Benito, V., 2025. Studentsâ attitudes towards AI and how they perceive the effectiveness of AI in designing video games. Sustainability 17, 3096. URL: https://w.mdpi.com/2071-1050/17/7/3096, doi:10.3390/su17073096. Saheb (2023) Saheb, T., 2023. Ethically contentious aspects of artificial intelligence surveillance: A social science perspective. AI and Ethics 3, 369â379. URL: https://link.springer.com/article/10.1007/s43681-022-00196-y, doi:10.1007/s43681-022-00196-y. Said et al. (2023) Said, N., Potinteu, A.E., Brich, I., Buder, J., Schumm, H., Huff, M., 2023. An artificial intelligence perspective: How knowledge and confidence shape risk and benefit perception. Computers in Human Behavior 149, 107855. URL: https://doi.org/10.1016/j.chb.2023.107855, doi:10.1016/j.chb.2023.107855. Schei et al. (2024) Schei, O.M., MĂžgelvang, A., Ludvigsen, K., 2024. Perceptions and use of AI chatbots among students in higher education: A scoping review of empirical studies. Education Sciences 14, 922. URL: https://w.mdpi.com/2227-7102/14/8/922, doi:10.3390/educsci14080922. Shaheen (2021) Shaheen, M.Y., 2021. Applications of artificial intelligence (AI) in healthcare: A review. ScienceOpen Preprints. Preprint. https://w.scienceopen.com/document?vid=b9349a6e-e2f0-4cf3-ae4e-efb4cda01690. Slovic (2016) Slovic, P., 2016. Perception of risk, in: The perception of risk. Routledge, p. 220â231. URL: https://w.taylorfrancis.com/chapters/edit/10.4324/9781315661773-13/perception-risk-paul-slovic. Stahl and Stahl (2021) Stahl, B.C., Stahl, B.C., 2021. Ethical issues of AI. Artificial Intelligence for a Better Future: An ecosystem Perspective on the Ethics of AI and Emerging Digital Technologies , 35â53URL: https://link.springer.com/chapter/10.1007/978-3-030-69978-9_4, doi:10.1007/978-3-030-69978-9_4. Stöhr et al. (2024) Stöhr, C., Ou, A.W., Malmström, H., 2024. Perceptions and usage of ai chatbots among students in higher education across genders, academic levels and fields of study. Computers and Education: Artificial Intelligence 7, 100259. URL: https://w.sciencedirect.com/science/article/pii/S2666920X24000626, doi:10.1016/j.caeai.2024.100259. Tierney et al. (2025) Tierney, A., Peasey, P., Gould, J., 2025. Student perceptions on the impact of AI on their teaching and learning experiences in higher education. Research & Practice in Technology Enhanced Learning 20, 1â25. doi:10.58459/rptel.2025.20005. Torres et al. (2024) Torres, A., Wenke, M., Lieneck, C., Ramamonjiarivelo, Z., Ari, A., 2024. A systematic review of artificial intelligence used to predict loneliness, social isolation, and drug use during the COVID-19 pandemic. Journal of Multidisciplinary Healthcare , 3403â3425URL: https://w.tandfonline.com/doi/full/10.2147/JMDH.S466099, doi:10.2147/JMDH.S466099. Trope and Liberman (2010) Trope, Y., Liberman, N., 2010. Construal-level theory of psychological distance. Psychological review 117, 440. URL: https://psycnet.apa.org/fulltext/2010-06891-005.html, doi:10.1037/a0018963. UNESCO (2021) UNESCO, 2021. Recommendation on the ethics of artificial intelligence. https://unesdoc.unesco.org/ark:/48223/pf0000381137. Venkatesh et al. (2003) Venkatesh, V., Morris, M.G., Davis, G.B., Davis, F.D., 2003. User acceptance of information technology: Toward a unified view. MIS quarterly , 425â478URL: https://misq.umn.edu/misq/article-abstract/27/3/425/1340/User-Acceptance-of-Information-Technology-Toward-A, doi:10.2307/30036540. Von Eschenbach (2021) Von Eschenbach, W.J., 2021. Transparency and the black box problem: Why we do not trust AI. Philosophy & Technology 34, 1607â1622. URL: https://link.springer.com/article/10.1007/s13347-021-00477-0, doi:10.1007/s13347-021-00477-0. Wang et al. (2021) Wang, Y., Liu, C., Tu, Y.F., 2021. Factors affecting the adoption of ai-based applications in higher education. Educational technology & society 24, 116â129. URL: https://w.jstor.org/stable/27032860. Yu et al. (2023) Yu, S., Bentley, B.L., Carroll, F., 2023. Enhancing smart home security: A privacy risk analysis framework, in: International Conference on Cyber Security, Privacy in Communication Networks, Springer. p. 295â308. URL: https://doi.org/10.1007/978-981-97-3973-8_18, doi:10.1007/978-981-97-3973-8_18. Yu and Carroll (2022a) Yu, S., Carroll, F., 2022a. Implications of AI in national security: understanding the security issues and ethical challenges, in: Artificial Intelligence in Cyber Security: Impact and Implications: Security Challenges, Technical and Ethical Issues, Forensic Investigative Challenges. Springer, p. 157â175. URL: https://doi.org/10.1007/978-3-030-88040-8_6, doi:10.1007/978-3-030-88040-8_6. Yu and Carroll (2022b) Yu, S., Carroll, F., 2022b. Insights into the next generation of policing: understanding the impact of technology on the police force in the digital age, in: Artificial Intelligence and National Security. Springer, p. 169â191. URL: https://doi.org/10.1007/978-3-031-06709-9_9, doi:10.1007/978-3-031-06709-9_9. Yu and Carroll (2023) Yu, S., Carroll, F., 2023. A balance of power: Exploring the opportunities and challenges of AI for a nation, in: Applications for Artificial Intelligence and Digital Forensics in National Security. Springer, p. 15â37. URL: https://doi.org/10.1007/978-3-031-40118-3_2, doi:10.1007/978-3-031-40118-3_2. Yu et al. (2024a) Yu, S., Carroll, F., Bentley, B.L., 2024a. Insights into privacy protection research in AI. IEEE Access 12, 41704â41726. URL: https://doi.org/10.1109/ACCESS.2024.3378126, doi:10.1109/ACCESS.2024.3378126. Yu et al. (2024b) Yu, S., Carroll, F., Bentley, B.L., 2024b. Trust and risk: Psybersecurity in the AI era, in: Psybersecurity. CRC Press, p. 130â155. URL: https://doi.org/10.1201/9781032664859-6, doi:10.1201/9781032664859-6. Yu et al. (2024c) Yu, S., Carroll, F., Bentley, B.L., 2024c. Trust and trustworthiness: Privacy protection in the ChatGPT era, in: Data Protection: The Wake of AI and Machine Learning. Springer, p. 103â127. URL: https://doi.org/10.1007/978-3-031-76473-8_6, doi:10.1007/978-3-031-76473-8_6. Zajko (2022) Zajko, M., 2022. Artificial intelligence, algorithms, and social inequality: Sociological contributions to contemporary debates. Sociology Compass 16, e12962. URL: https://compass.onlinelibrary.wiley.com/doi/full/10.1111/soc4.12962, doi:10.1111/soc4.12962. Zhai et al. (2024) Zhai, C., Wibowo, S., Li, L.D., 2024. The effects of over-reliance on AI dialogue systems on studentsâ cognitive abilities: A systematic review. Smart Learning Environments 11, 28. URL: https://link.springer.com/article/10.1186/s40561-024-00316-7, doi:10.1186/s40561-024-00316-7. Zhai et al. (2021) Zhai, X., Chu, X., Chai, C.S., Jong, M.S.Y., Istenic, A., Spector, M., Liu, J.B., Yuan, J., Li, Y., 2021. A review of artificial intelligence (AI) in education from 2010 to 2020. Complexity 2021, 1â18. URL: https://onlinelibrary.wiley.com/doi/full/10.1155/2021/8812542, doi:10.1155/2021/8812542.