AI evidence extraction

The AI Guilt Complex: Moral Emotions and Ethical Dilemmas in Academic Technology Adoption

Diane Vassallo · Journal of Academic Ethics · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

3/4

Quality (LMQS)

Evidence

Citations

30.06

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1007/s10805-026-09726-3

Methodology & findings

Study design

Cross-sectional mixed-methods survey design with exploratory investigation.

Sample

N = 109, 20 groups

Primary method

Statistical analyses conducted using R (version 4.3.1) with packages including psych, cluster, and factoextra. Methods included: (1) Descriptive statistics (means, standard deviations, frequency distributions); (2) Scale validation (Cronbach's α reliability analysis, item-total correlations, exploratory factor analysis using minimum residual extraction with oblimin rotation); (3) Group comparisons using independent samples t-tests and one-way ANOVA with effect sizes (Cohen's d and η²); (4) Exploratory k-means cluster analysis to identify moral response profiles with optimal cluster number determined through scree plot examination; (5) Pearson correlations with 95% confidence intervals. Statistical significance set at p<.05; marginal results (p<.10) noted. Post-hoc power analysis using G*Power 3.1 indicated adequate power (0.80) for detecting medium effect sizes (r=.30) in correlations and large effects (f²=0.35) in regression. Qualitative data underwent thematic analysis following Braun and Clarke's (2006) reflexive thematic analysis approach.

Main result

The study found that "the phrase 'I feel like I'm cheating when I use AI in my academic work', endorsed by over a quarter of respondents in this study, captures a moral dimension of AI integration that has received little empirical attention." Most significantly, "non-users report significantly higher guilt than users, particularly for the remorse item interpreted as anticipated remorse among nonusers," revealing a "guilt paradox" where "anticipatory guilt appears to function as a barrier to engagement, preventing the very experiences that might otherwise reduce ethical anxiety." Four distinct moral response profiles were identified: Comfortable Adopters (26.6%), Guilty Non-Users (29.4%), Cautious Users (28.4%), and Morally Distressed Avoiders (15.6%), with "anticipatory guilt substantially exceeds experienced guilt."

Reports effect sizes and confidence intervals.

Research paradigm

Mixed methods (positivist quantitative + constructivist qualitative)

Author conclusions

"This study addressed its three guiding questions at an exploratory level. First, the nature, prevalence, and structure of AI-related moral emotions were characterised through the AI Guilt Index, prevalence estimates, and cluster analysis, revealing anticipatory guilt as more salient than experienced remorse and identifying four distinct response profiles." The authors conclude that "the AI Guilt Complex represents a new form of moral emotion at the intersection of academic values and technological augmentation" and that "addressing these moral emotions, alongside technical training and policy, will be crucial for healthy adoption that preserves academic values while embracing innovation." They note that "by foregrounding moral emotions and ethical dilemmas in academic AI adoption, this paper offers an initial ethical lens that can inform future empirical research and institutional support strategies."

Risk of bias

Selection bias: 3.5% response rate (109/3,100 invited); self-selected sample willing to discuss AI ethics; Author positionality bias: Researcher is member of Faculty of Education (23.3% response rate vs. 3.5% overall, vs. Medicine 1.5%); Differential response rates across disciplines (Education 24.8% vs. Medicine 5.5%); Missing data from non-respondents (96.5% of population); reasons for non-response unknown; Lack of attention check items despite sensitive topic; Single institution, single country context limits generalizability; Small non-user subgroup (n=8) limits statistical power for key comparisons; Possible social desirability bias in self-reported AI use and guilt responses; Selection bias: Low response rate (3.5%) suggests self-selection of academics willing to discuss AI ethical concerns; Institutional bias: Single-institution study at University of Malta may not represent broader academic populations; Researcher positionality bias: Author is Faculty of Education member, which showed highest response rate (23.3% vs. 3.5% overall); Measurement bias: No attention check items included; survey completion time median ~15 minutes may indicate rushed responses; Disciplinary bias: Faculty of Education overrepresented (24.8%); Medicine underrepresented (5.5% vs. 396 staff); Non-response bias: 96.5% non-response rate may indicate moral disengagement or avoidance among broader population; Selection bias: 3.5% response rate (109/3,100) indicates substantial self-selection; those willing to engage with sensitive ethical topics likely differ from non-respondents; Response bias: Sensitive topic (AI ethics guilt) may discourage participation from those experiencing moral discomfort or disengagement; Institutional bias: Single-institution study with higher response rates from Faculty of Education (23.3%) than Medicine (1.5%), suggesting differential engagement by discipline; Researcher positionality bias: Author acknowledged as member of Faculty of Education, which showed notably higher participation rate (27/116 = 23.3% vs. overall 3.5%); Attrition: No attention check items included due to sensitivity concerns, potentially allowing inattentive responses; Social desirability bias: Sensitive topic may elicit socially desirable rather than authentic responses regarding moral discomfort; Sampling bias: Non-users constitute only small subgroup (n=8), limiting statistical power for user vs. non-user comparisons; Temporal bias: Single time-point measurement prevents assessment of whether guilt changes over time or with AI experience

Limitations

"The 3.5% response rate, while providing meaningful data about those willing to engage with AI ethical concerns, limits statistical generalisability." "The cross-sectional design prevents causal inference about relationships between variables." "The single-institution, single-country context limits transferability to other academic cultures." "The development of new measures, while necessary given the novel phenomenon, means psychometric properties require further validation." "The four moral response profiles identified through cluster analysis should be considered tentative heuristic categories requiring confirmation in larger, more diverse samples rather than stable typologies." "Given the exploratory nature of this study and sample constraints, all findings should be considered hypothesis-generating rather than hypothesis-testing."

Open questions raised

Authors identify several gaps: "First, empirical research on emotional responses to AI adoption is virtually absent. While opinion pieces proliferate, systematic data on how academics experience AI integration emotionally remains scarce. Second, the relationship between AI use and professional identity has received primarily theoretical rather than empirical treatment. Third, the behavioural consequences of AI-related moral emotions, particularly regarding disclosure, collaboration, and teaching practices, remain unexplored." Future research directions include: intervention studies testing approaches to address AI guilt, experimental research on AI framings' effects on moral emotions, international/cross-cultural research on the AI Guilt Complex, longitudinal research tracking emotional journeys through AI adoption, and research exploring protective factors against AI-related guilt.
Empirical research on emotional responses to AI adoption in academia is virtually absent
Relationship between AI use and professional identity has received primarily theoretical rather than empirical treatment
Behavioural consequences of AI-related moral emotions remain unexplored, particularly regarding disclosure, collaboration, and teaching practices
Intervention studies testing approaches to address AI guilt, particularly for early-career academics
Experimental research examining how different framings of AI influence moral emotions

Data: Not mentioned; interviews noted but not yet conducted: "email addresses requested only from those willing to participate in follow-up interviews (34 participants expressed interest but these interviews were not conducted within the study timeframe and will be reported separately)"Code: Not mentionedExtracted from: pdfAgreement 48%

Explore related topics

Responsible AI Use AI Governance Mixed Methods