Researcher Artificial Intelligence Addiction Scale

1. Abstract

The contemporary proliferation of generative artificial intelligence (GenAI) systems and large language models (LLMs) has initiated a structural paradigm shift across academic scholarship, scientific inquiry, and biomedical research. While automated technologies offer marked efficiencies in literature synthesis, automated data processing, coding, and manuscript structuring, their unregulated assimilation into scholarly workflows introduces notable psychological and behavioral risks. Chief among these concerns is the transition from instrumental augmentation to maladaptive cognitive dependency and behavioral addiction. The Researcher Artificial Intelligence Addiction Scale (RAIAS) was formulated to operationalize, quantify, and psychometrically evaluate this emerging behavioral phenomenon among academic investigators and healthcare researchers. Developed by Ahmed Abdelwahab Ibrahim El-Sayed, Samira Ahmed Alsenany, Maha Gamal Ramadan Asal, and Ibrahim Alasqah (2025), the RAIAS is a dedicated 22-item self-report instrument grounded in established behavioral addiction frameworks, specifically adapting the classic components model of addiction and standard diagnostic criteria to academic labor.

The psychometric evaluation of the RAIAS was executed in a robust multi-stage validation study comprising a convenience sample of 718 nursing researchers, partitioned into independent subsamples for cross-validation. Exploratory Factor Analysis (EFA) identified a five-factor latent architecture explaining 73.66% of the cumulative variance, encompassing Compulsive Behavior, Overdependency, Functional Impairment, Withdrawal, and Tolerance. Confirmatory Factor Analysis (CFA) substantiated this multidimensional construct across first-order and second-order structural models, achieving exceptional goodness-of-fit indices (Comparative Fit Index [CFI] = 0.962, Root Mean Square Error of Approximation [RMSEA] = 0.060). The instrument exhibits high internal consistency, demonstrated by a global Cronbach’s alpha (α) of 0.924, a McDonald’s omega (ω) of 0.870, and a Spearman-Brown split-half reliability coefficient of 0.814. Interfactor correlations range moderately from 0.41 to 0.62, confirming discriminant distinctiveness alongside a unifying higher-order construct. Diagnostic stratification utilizes percentile-based risk cut-offs: low risk (< 25th percentile), moderate risk (25th to 75th percentile), and high risk (≥ 75th percentile). The RAIAS represents an important psychometric development for institutional oversight, research integrity maintenance, and scholarly self-regulation in the era of automated intelligence.

2. Keywords

Artificial Intelligence, Behavioral Addiction, Psychometrics, Research Ethics, Cognitive Dependency, Academic Integrity, Researcher Well-Being, Generative AI, Construct Validity, Nursing Research

3. Authors

The Researcher Artificial Intelligence Addiction Scale was conceptualized, developed, and empirically validated by an interdisciplinary team of behavioral science, nursing, and medical informatics researchers:

• Ahmed Abdelwahab Ibrahim El-Sayed — Lead investigator specializing in psychometric instrumentation, healthcare informatics, and clinical nursing education.
• Samira Ahmed Alsenany — Professor and senior academic researcher focusing on gerontological nursing, professional healthcare ethics, and digital health technology implementation.
• Maha Gamal Ramadan Asal — Scholar and clinical researcher investigating ethical dimensions of algorithmic tools, educational technology, and nursing administration.
• Ibrahim Alasqah (Corresponding Author) — Department of Medical Surgical Nursing / Healthcare Management, Qassim University, Saudi Arabia. Correspondence email: [email protected].

4. Purpose

The primary purpose of the Researcher Artificial Intelligence Addiction Scale (RAIAS) is to assess and quantify problematic, addictive, and cognitively debilitating patterns of artificial intelligence reliance among academic faculty, graduate researchers, clinical investigators, and scientific scholars. Over the past several decades, the psychometric canon has generated numerous measurement scales addressing general Internet addiction, social media overuse, and smartphone dependency (e.g., Young’s Internet Addiction Test, the Smartphone Addiction Scale). However, these legacy instruments are ill-equipped to evaluate academic workflows. Scientific research is an intrinsically goal-directed, computer-mediated, and cognitively intensive endeavor. Researchers routinely spend tens of hours each week immersed in digital environments conducting data mining, computational modeling, and manuscript editing.

Consequently, administering generalized digital addiction scales to academic populations frequently yields high rates of false positives, erroneously pathologizing productive, normative occupational involvement. Conversely, generic instruments completely fail to capture the subtle, insidious boundary where algorithmic assistance metastasizes into epistemic abdication—the uncritical outsourcing of hypothesis formulation, critical appraisal, literature synthesis, and interpretive reasoning to probabilistic text models. The RAIAS was purposefully engineered to address this theoretical and diagnostic void. It differentiates functional technological synergy from compulsive technological dependency, pinpointing when the researcher’s cognitive agency, analytical rigor, and professional self-efficacy become severely compromised.

In clinical, institutional, and organizational settings, the RAIAS provides academic deans, research integrity officers, institutional review boards (IRBs), doctoral supervisors, and occupational health psychologists with a standardized, empirically validated screening tool. The scale fulfills multiple strategic objectives:

• Institutional Screening and Integrity Assurance: Enabling research institutions and universities to monitor the ethical and behavioral impact of generative AI integration, ensuring that institutional adoption does not undermine original scholarship, rigorous scientific discovery, or peer-review fidelity.
• Early Psychological and Educational Intervention: Identifying postgraduate trainees and early-career researchers who exhibit acute psychological vulnerabilities, such as academic imposter syndrome or performance anxiety, which frequently manifest as excessive reliance on algorithmic generation.
• Workplace and Occupational Health Monitoring: Facilitating targeted counseling and educational workshops on algorithmic hygiene, critical AI literacy, and cognitive resilience before pathological dependency culminates in catastrophic research misconduct, automated hallucination dissemination, or career derailment.
• Cross-Disciplinary Empirical Research: Serving as an operationalized dependent or independent variable in large-scale social science, medical education, and bibliometric studies investigating the ongoing structural transformation of academic work under pervasive machine intelligence.

5. Psychological Construct

The Researcher Artificial Intelligence Addiction Scale operationalizes artificial intelligence addiction not as a monolithic clinical disease entity, but as a multidimensional, cognitive-behavioral constellation characterized by an impaired capacity to regulate AI tool utilization, an escalating psychological and intellectual dependence, and marked occupational impairment within scholarly environments. The instrument synthesizes cognitive psychology, occupational health, and behavioral addiction paradigms into five correlated, theoretically bounded subscales comprising 22 self-report items:

Compulsive Behavior

The Compulsive Behavior subscale measures the automatic, involuntary, and irresistible impulse to engage generative AI platforms during scientific workflows, frequently bypassing deliberate, conscious decision-making. Researchers exhibiting high levels of compulsivity report initiating queries, generating automated summaries, or employing algorithmic writing assistants as an automated default reaction rather than a considered strategic choice. This dimension mirrors the classic compulsive-impulsive spectrum of behavioral disorders, wherein the individual experiences an elevated internal drive or cognitive tension that is momentarily alleviated only upon engaging the computational interface, despite pre-existing intentions to formulate thoughts independently.

Overdependency

The Overdependency dimension evaluates the erosion of scholarly autonomy and intellectual self-efficacy, manifested as a profound psychological belief that one is incapable of conducting rigorous research, articulating coherent scientific prose, or conceptualizing novel hypotheses without synthetic intervention. In contrast to simple tool usage, overdependency captures a state of cognitive atrophy and learned helplessness. Scholars scoring high on this dimension exhibit an epistemic crisis: they second-guess their innate intellect, experience paralysis when confronted with a blank document, and perceive machine-generated outputs as inherently superior to their own cognitive faculties.

Functional Impairment

The Functional Impairment subscale appraises the deleterious downstream consequences of chronic, unmonitored AI over-reliance upon scientific production, methodological validity, and research integrity. This facet captures tangible occupational decrements, such as the uncritical transcription of AI hallucinations, superficial literature reviews that neglect foundational primary sources, degradation of advanced analytical problem-solving skills, and ethical breaches concerning unverified citations or compromised intellectual property. Furthermore, it measures the interpersonal and professional friction emerging when collaborators or peer reviewers detect synthetic artifacts, formulaic rhetoric, or compromised originality in the researcher’s scholarly output.

Withdrawal

The Withdrawal subscale captures the acute negative affective and psychological state experienced by researchers when access to AI tools is abruptly severed, restricted, or technically disrupted. Operationally defined within the academic domain, withdrawal symptoms do not manifest as autonomic physiological distress (e.g., diaphoresis, tremors), but rather as significant cognitive disorientation, intense situational anxiety, acute frustration, scholarly helplessness, and executive dysfunction. Researchers with high withdrawal vulnerability report experiencing existential dread, severe procrastination, or an inability to initiate basic scholarly writing tasks during server outages, institutional firewalls, or offline working conditions.

Tolerance

The Tolerance subscale quantifies the progressive, escalating requirement for deeper, more pervasive, and cognitively foundational AI involvement to achieve the same subjective sense of productivity, scholarly confidence, or intellectual gratification. While an investigator may initially employ conversational AI merely for peripheral copyediting or reference formatting, tolerance is demonstrated when the technological scope systematically broadens to encompass core intellectual processes: framing research questions, constructing theoretical frameworks, synthesizing discussion sections, and drawing experimental conclusions. The individual progressively yields higher levels of intellectual sovereignty to the algorithm over time.

6. Theoretical Framework

The conceptual architecture of the RAIAS is anchored within established theoretical traditions from clinical psychology, behavioral addiction, and human-computer interaction (HCI). Most prominently, the instrument adapts Mark Griffiths’ (2005) Components Model of Addiction, which posits that all behavioral addictions—irrespective of whether they involve gambling, gaming, the internet, or specific digital utilities—share six core biopsychosocial components: salience, mood modification, tolerance, withdrawal symptoms, conflict, and relapse.

In the RAIAS, Griffiths’ foundational taxonomy is rigorously adapted to the intellectual and occupational idiosyncrasies of academic scholarship:

• Cognitive and Behavioral Salience: Reflected across the Compulsive Behavior and Overdependency dimensions, where AI platforms dominate the researcher’s cognitive space, cognitive schema, and task-initiation strategies.
• Conflict and Harm: Directly operationalized through the Functional Impairment subscale, representing both intrapsychic conflict (cognitive dissonance over diminished authenticity) and interpersonal/occupational conflict (academic scrutiny, loss of originality, journal retractions).
• Tolerance and Escalation: Operationalized through the progressive cognitive outsourcing construct, capturing the shift from peripheral syntactic support to central epistemological delegation.
• Withdrawal: Captured through the acute situational affective distress, academic panic, and executive stalling that accompany computational unavailability.

Beyond Griffiths’ addiction model, the RAIAS integrates principles from Albert Bandura’s Social Cognitive Theory, specifically the dynamics of perceived self-efficacy. Bandura highlighted that human agency operates via reciprocal determinism among cognitive factors, behavioral patterns, and environmental influences. When scholars consistently attribute successful academic outcomes (e.g., published articles, secured grants, streamlined coding) to an external computational agent rather than their own internal intellect, their academic self-efficacy deteriorates. This process fosters an external locus of control and accelerates compulsive reliance. Furthermore, the instrument draws from cognitive offloading theory in cognitive science, which explains how cognitive agents naturally minimize biological mental effort by distributing computational demands onto reliable external artifacts. When institutional hyper-competition, rapid publication timelines, and performance pressures intersect with cognitive offloading affordances, the transition from functional cognitive scaffolding to maladaptive psychological addiction becomes markedly amplified.

7. Validity

The psychometric validation of the Researcher Artificial Intelligence Addiction Scale was conducted using an extensive, multi-phase methodological protocol conforming to the Standards for Educational and Psychological Testing. The initial pool of candidate items underwent comprehensive content and face validation via an expert panel comprising nursing scholars, biostatisticians, clinical psychologists, and medical ethicists. Experts systematically evaluated item clarity, conceptual relevance, readability, and content coverage, leading to iterative linguistic refinements and the elimination of ambiguous or overly broad prompts to ensure high content validity.

Construct validity was empirically substantiated through structural equation modeling (SEM) and factor analytic procedures conducted on a broad sample of 718 nursing researchers. The empirical data corroborated the five-dimensional theoretical model, demonstrating that the five subscales represent structurally distinct yet cohesive psychological facets. The interfactor correlations among the five dimensions ranged from 0.41 to 0.62. This moderate correlation range is psychometrically optimal: it indicates sufficient shared variance to warrant their aggregation under a unified overarching construct of AI addiction, while confirming that no two subscales suffer from multicollinearity or excessive redundancy (which would be indicated by correlations exceeding 0.85).

Convergent validity was further demonstrated through high, statistically significant standardized factor loadings across all 22 items on their respective latent variables, with all standardized loadings exceeding the conservative psychometric threshold of 0.50 (with the majority surpassing 0.70). Discriminant validity was affirmed through average variance extracted (AVE) analyses and factor correlation matrices, which confirmed that the shared variance between any two factors was consistently lower than the variance extracted by each individual factor construct.

8. Reliability

The RAIAS displays exceptional internal consistency and psychometric reliability across diverse statistical metrics, confirming that its items consistently capture the underlying latent constructs without excessive measurement error:

• Cronbach’s Alpha (α): The total 22-item scale demonstrated an overall Cronbach’s alpha of 0.924 (frequently reported in scorecards as 0.92). In psychometric theory, an alpha coefficient exceeding 0.90 signifies excellent internal consistency suitable for both group-level academic research and individual-level psychodiagnostic screening. Individual subscales also exhibited robust internal consistency values ranging well above accepted academic thresholds (α > 0.80).
• McDonald’s Omega (ω): Recognizing that Cronbach’s alpha assumes tau-equivalence (equal factor loadings across all items) and can consequently under- or overestimate true reliability, the scale developers calculated McDonald’s omega total. The RAIAS yielded a high McDonald’s ω of 0.870, verifying superior composite reliability under congeneric structural assumptions.
• Split-Half Reliability: To verify measurement precision across item halves, the Spearman-Brown split-half reliability coefficient was determined, yielding a robust coefficient of 0.814. This confirmed that score stability remains resilient against item order effects and internal test fatigue.

Collectively, these empirical indices confirm that the RAIAS functions with high measurement precision, exhibiting minimal random error variance across academic research cohorts.

9. Factor Analysis

The structural dimensionality of the RAIAS was elucidated through a rigorous split-sample cross-validation methodology utilizing the total sample of 718 nursing researchers. The dataset was randomly partitioned into two independent, equal subsamples to ensure methodological independence: one designated for Exploratory Factor Analysis (EFA, n = 359) and the second reserved for Confirmatory Factor Analysis (CFA, n = 359).

Exploratory Factor Analysis (EFA)

Prior to extraction, the sampling adequacy and matrix factorability were verified. The Kaiser-Meyer-Olkin (KMO) measure demonstrated high sampling adequacy, and Bartlett’s Test of Sphericity attained statistical significance (p < 0.001), indicating that the correlation matrix was appropriate for factor extraction. Principal Component Analysis followed by oblique (Promax) rotation was conducted to accommodate anticipated theoretical correlations among behavioral dimensions. The EFA extracted five distinct factors with eigenvalues greater than 1.0 (Kaiser criterion), which collectively accounted for 73.66% of the cumulative variance. The extracted five-factor pattern cleanly mapped onto the theoretical dimensions:

1. Compulsive Behavior
2. Overdependency
3. Functional Impairment
4. Withdrawal
5. Tolerance

Confirmatory Factor Analysis (CFA)

To cross-validate the empirical structure identified in the EFA, Confirmatory Factor Analysis was executed on the independent second subsample using maximum likelihood estimation. Both first-order (five correlated latent factors) and hierarchical second-order (five first-order latent factors loading onto a single higher-order “Researcher AI Addiction” construct) structural models were evaluated.

The first-order structural model demonstrated an exceptional fit to the empirical data:

• Comparative Fit Index (CFI): 0.962 (exceeding the standard benchmark of ≥ 0.95 for superior model fit).
• Root Mean Square Error of Approximation (RMSEA): 0.060 (with a 90% confidence interval falling entirely within acceptable limits of ≤ 0.08).
• Tucker-Lewis Index (TLI): Maintained values aligning with excellent goodness-of-fit benchmarks (> 0.95).

The higher-order structural model demonstrated comparable, robust fit statistics, confirming that while the five subscales represent clinically distinct behavioral phenomena, they legitimately converge onto a global, overarching construct of researcher AI addiction.

10. Instrument / Measurement Tool

The Researcher Artificial Intelligence Addiction Scale is a structured psychometric instrument designed for self-report administration. Its formal administrative profile is outlined below:

• Test Type: Standardized self-report rating scale / psychodiagnostic screening questionnaire.
• Target Population: Academic researchers, clinical investigators, university faculty, doctoral candidates, graduate students, and healthcare professionals (validated primarily in nursing researchers).
• Number of Items: 22 items distributed across five latent subscales.
• Latent Subscales:
  • Compulsive Behavior
  • Overdependency
  • Functional Impairment
  • Withdrawal
  • Tolerance
• Response Format: 22 items evaluated along a graded Likert-type frequency continuum.
• Administration Mode: Self-administered (paper-and-pencil or secure web-based survey systems); typical completion duration is approximately 6 to 10 minutes.
• Scoring and Risk Categorization: Global and subscale scores are computed by aggregating individual item responses. Higher total scores denote an elevated frequency and severity of maladaptive AI addiction behaviors. Diagnostic stratification is operationalized using percentile-based distribution thresholds:
  • Low Risk: Overall score falling below the 25th percentile (< 25th percentile). Reflects functional, controlled, and adaptive AI tool utilization.
  • Moderate Risk: Overall score falling between the 25th and 75th percentiles (25th–75th percentile). Indicates emerging over-reliance, periodic compulsivity, and early vulnerability to cognitive dependency.
  • High Risk: Overall score falling at or above the 75th percentile (≥ 75th percentile). Denotes substantial behavioral addiction, functional impairment, ethical compromise, and profound cognitive dependency requiring immediate self-reflection or institutional mentoring intervention.

11. Permissions & Fee and Test Year

The Researcher Artificial Intelligence Addiction Scale was published in 2025 in the Journal of Nursing Management. The scientific article detailing the psychometric formulation, factor structure, and reliability coefficients was authored by Ahmed Abdelwahab Ibrahim El-Sayed, Samira Ahmed Alsenany, Maha Gamal Ramadan Asal, and Ibrahim Alasqah. The scale was established as a standardized research tool for academic evaluation.

Regarding licensing and operational accessibility: while the overarching psychometric findings, statistical indices, and theoretical dimensions are published within the peer-reviewed scholarly literature, the full, verbatim item inventory is retained under copyright protection. Investigators, universities, and mental health clinicians wishing to employ, adapt, translate, or digitally administer the RAIAS for non-commercial academic research, institutional audits, or educational investigations must direct formal permission requests to the corresponding author, Dr. Ibrahim Alasqah, at [email protected]. Commercial utilization, automated digital platform integration, and proprietary testing require explicit contractual authorization from the copyright holders.

12. References

The theoretical foundations, methodological antecedents, and validation literature associated with the RAIAS are documented below in APA 7th edition format:

• Abdelhafiz, A. S., Ali, A., Maaly, A. M., Ziady, H. H., Sultan, E. A., & Mahgoub, M. (2024). Knowledge, perceptions and attitude of researchers towards using ChatGPT in research. Journal of Medical Systems, 48(1), Article 24. https://doi.org/10.1007/s10916-024-02044-4
• Asal, M. G. R., Alsenany, S. A., Badoman, T. O., & El‐Sayed, A. A. I. (2025). Ethical awareness in the use of large language models: Development and validation of a scale for healthcare professionals. Journal of Evaluation in Clinical Practice, 31(5), Article 70241. https://doi.org/10.1111/jep.70241
• Billieux, J., Maurage, P., Lopez-Fernandez, O., Kuss, D. J., & Griffiths, M. D. (2015). Can disordered mobile phone use be considered a behavioral addiction? An update on current evidence and a comprehensive model for future research. Current Addiction Reports, 2(2), 156–162. https://doi.org/10.1007/s40429-015-0054-y
• Coaley, K. (2014). An introduction to psychological assessment and psychometrics (2nd ed.). SAGE Publications.
• DeVellis, R. F. (2021). Scale development: Theory and applications (5th ed.). SAGE Publications.
• Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L., Jeyaraj, A., Kar, A. K., … & Wright, R. (2023). Opinion Paper: “So what if ChatGPT wrote it?” Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy. International Journal of Information Management, 71, Article 102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642
• El-Sayed, A. A. I., Alsenany, S. A., Asal, M. G. R., & Alasqah, I. (2025). Researcher Artificial Intelligence Addiction Scale. Journal of Nursing Management, 2025, Article 8458533. https://doi.org/10.1155/jonm/8458533
• El‐Sayed, A. A. I., Alsenany, S. A., Badoman, T. O., & Asal, M. G. R. (2025). Development and validation of a scale for nurses’ ethical awareness in the use of artificial intelligence: A methodological study. Nursing & Health Sciences, 27(2), Article 70140. https://doi.org/10.1111/nhs.70140
• Filetti, S., Fenza, G., & Gallo, A. (2024). Research design and writing of scholarly articles: New artificial intelligence tools available for researchers. Endocrine, 85(3), 1104–1116. https://doi.org/10.1007/s12020-024-03977-z
• Fournier, L., Schimmenti, A., Musetti, A., Boursier, V., Flayelle, M., Cataldo, I., … & Billieux, J. (2023). Deconstructing the components model of addiction: An illustration through “addictive” use of social media. Addictive Behaviors, 143, Article 107694. https://doi.org/10.1016/j.addbeh.2023.107694
• Griffiths, M. (2005). A ‘components’ model of addiction within a biopsychosocial framework. Journal of Substance Use, 10(4), 191–197. https://doi.org/10.1080/14659890500114359
• Khan, N., Osmonaliev, K., & Sarwar, M. (2023). Pushing the boundaries of scientific research with the use of artificial intelligence tools: Navigating risks and unleashing possibilities. Nepal Journal of Epidemiology, 13(1), 1258–1263. https://doi.org/10.3126/nje.v13i1.53721
• Kline, R. B. (2023). Principles and practice of structural equation modeling (5th ed.). Guilford Press.
• Lim, B. K. H., Seth, I., & Rozen, W. M. (2024). The role of artificial intelligence tools on advancing scientific research. Aesthetic Plastic Surgery, 48(15), 3036–3038. https://doi.org/10.1007/s00266-023-03526-5
• Lin, C. Y., & Chien, Y. C. (2024). ChatGPT addiction: A proposed phenomenon of dual parasocial interaction. Taiwanese Journal of Psychiatry, 38(3), 153–155. https://doi.org/10.4103/TPSY.TPSY_28_24
• Morales-García, W. C., Sairitupa-Sanchez, L. Z., Morales-García, S. B., & Morales-García, M. (2024). Development and validation of a scale for dependence on artificial intelligence in university students. Frontiers in Education, 9, Article 1323898. https://doi.org/10.3389/feduc.2024.1323898
• Polit, D. F. (2020). Essentials of nursing research: Appraising evidence for nursing practice (10th ed.). Wolters Kluwer.
• Satchell, L. P., Fido, D., Harper, C. A., Shaw, H., Davidson, B. I., Ellis, D. A., … & Pavetich, M. (2021). Development of an Offline-Friend Addiction Questionnaire (O-FAQ): Are most people really social addicts? Behavior Research Methods, 53(3), 1097–1106. https://doi.org/10.3758/s13428-020-01462-9
• Schumacker, R. E., & Lomax, R. G. (2010). A beginner’s guide to structural equation modeling (3rd ed.). Routledge.
• Tamrin, S., Omar, N., Kamaruzaman, K., Zaghlol, A., & Abdul Aziz, M. (2024). Evaluating the impact of AI dependency on cognitive ability among Generation Z in higher educational institutions: A conceptual framework. Information Management and Business Review, 16(3S(I)a), 1027–1033. https://doi.org/10.22610/imbr.v16i3S(I)a.4191
• Ullman, J. B. (2012). Structural equation modeling. In I. B. Weiner (Ed.), Handbook of Psychology (2nd ed., Vol. 2, pp. 679–707). John Wiley & Sons.
• Zhang, S., Zhao, X., Zhou, T., & Kim, J. (2024). Do you have AI dependency? The roles of academic self-efficacy, academic stress, and performance expectations on problematic AI usage behavior. International Journal of Educational Technology in Higher Education, 21(1), Article 40. https://doi.org/10.1186/s41239-024-00467-0

13. Items of the Scale