Attitudes Towards Artificial Intelligence at Work Scale (AAAW)

Abstract

The Attitudes Towards Artificial Intelligence at Work Scale (AAAW) is a psychometric instrument developed by Jiyoung Park, Sang Eun Woo, and JeongJin Kim (2024) to evaluate the multidimensional psychological responses of employees toward the implementation and integration of artificial intelligence (AI) systems within contemporary organizational settings. Grounded in interdisciplinary paradigms spanning human-computer interaction, organizational psychology, and technology adoption literature, the AAAW captures cognitive appraisals, affective responses, and job-related evaluations across six distinct dimensions: Perceived Humanlikeness of AI, Perceived Adaptability of AI, Perceived Quality of AI, AI Use Anxiety, Job Insecurity, and Personal Utility of AI.

Comprising 25 items evaluated on an authentic 5-point Likert response scale ranging from 1 (strongly disagree) to 5 (strongly agree), the AAAW addresses significant limitations in pre-existing technology acceptance measures that fail to capture the autonomous, relational, and disruptive capabilities unique to modern machine learning and autonomous agents. The instrument was rigorously developed and validated across three independent samples of working adults using both exploratory factor analysis (EFA) and confirmatory factor analysis (CFA). Psychometric evaluations established robust internal consistency, with Cronbach’s alpha coefficients exceeding .80 for all six dimensions and composite reliability (CR) estimates ranging from .70 to .94. Construct validity was corroborated by average variance extracted (AVE) values ranging from .52 to .77, supporting convergent validity, while factor intercorrelations satisfied the Fornell-Larcker criterion for discriminant validity. Furthermore, structural equation modeling demonstrated superior goodness-of-fit for the first-order six-factor model (χ² = 673.60, df = 260, CFI = 0.98, NFI = 0.96, TLI = 0.97, RMSEA = 0.04, SRMR = 0.04), establishing the AAAW as a methodologically rigorous assessment tool for organizational research, talent management, and strategic technological change initiatives.

Keywords

Attitudes Towards Artificial Intelligence at Work Scale, AAAW, workplace artificial intelligence, technology acceptance, human-computer interaction, AI use anxiety, job insecurity, perceived adaptability, psychometric validation, occupational health psychology

Authors

The Attitudes Towards Artificial Intelligence at Work Scale was conceptualized, developed, and empirically validated by an interdisciplinary team of researchers in organizational psychology and management:

• Jiyoung Park, Ph.D. — Department of Psychological Sciences, Purdue University, West Lafayette, Indiana, United States. (ORCID: 0000-0002-3162-8700).
• Sang Eun Woo, Ph.D. — Professor of Psychological Sciences, Department of Psychological Sciences, Purdue University, West Lafayette, Indiana, United States. Dr. Woo specializes in personality psychology, workplace assessment, psychometrics, and quantitative methodologies in organizational research.
• JeongJin Kim, Ph.D. — Independent Researcher and Consultant in Industrial-Organizational Psychology and Human Resource Management. (ORCID: 0009-0005-2358-0677).

Correspondence regarding the instrument and underlying psychometric data can be directed to the corresponding author, Sang Eun Woo, via the Department of Psychological Sciences at Purdue University.

Purpose

The transition toward the fourth industrial revolution has accelerated the deployment of artificial intelligence algorithms, generative machine learning systems, automated decision-making platforms, and physical-digital collaborative robotics within workplaces worldwide. Traditional instruments assessing technology adoption—such as the original Technology Acceptance Model (TAM) formulated by Davis (1989) or the Unified Theory of Acceptance and Use of Technology (UTAUT; Venkatesh et al., 2003)—were primarily engineered to measure static, non-autonomous tools like spreadsheet software, enterprise resource planning databases, and electronic communication systems. These classic frameworks predominantly focus on instrumental utility (e.g., perceived usefulness and perceived ease of use) while neglecting the anthropomorphic, self-learning, autonomous, and potentially existential workplace threats introduced by modern AI agents.

The primary purpose of the Attitudes Towards Artificial Intelligence at Work Scale (AAAW) is to provide an empirically grounded, nuanced, and psychometrically robust measurement instrument capable of assessing the complex cognitive, affective, and behavioral dispositions of the workforce toward AI. Rather than treating attitudes as a monolithic positive-versus-negative continuum, the AAAW conceptualizes workplace AI attitudes as a multidimensional matrix reflecting both technological features (such as perceived quality, adaptability, and humanlikeness) and employee psychological vulnerability (such as technological anxiety, fears of obsolescence, and occupational displacement).

In applied organizational environments, the scale serves critical diagnostic and strategic functions. Human resource practitioners and organizational development leaders can employ the AAAW to assess readiness for technological change, identify specific psychological barriers to AI adoption, pinpoint training deficiencies, and gauge employee distress prior to large-scale system rollouts. For instance, diagnosing whether resistance stems from technical distrust (low perceived quality), user-interface intimidation (high AI use anxiety), or survival threats (high job insecurity) allows leadership to deploy targeted behavioral interventions, such as psychological safety workshops, algorithmic transparency initiatives, or upskilling programs.

From an academic and empirical perspective, the AAAW provides industrial-organizational psychologists, sociologists of work, and organizational behavior researchers with a standardized tool to model structural relationships between AI attitudes and critical occupational criteria. The scale has demonstrated explicit utility in predicting recruiting and personnel selection outcomes, employee well-being, counterproductive work behaviors, organizational citizenship behaviors directed toward automated workflows, and voluntary turnover intentions in technologically disrupted industries.

Psychological Construct

The AAAW operationalizes attitudes toward workplace artificial intelligence as a multifaceted psychological construct encompassing six interrelated yet functionally distinct dimensions. These dimensions bridge cognitive evaluations of machine agency and capability with deeply rooted affective and existential employee experiences:

1. Perceived Humanlikeness of AI

This subscale captures the extent to which an individual worker attributes anthropomorphic characteristics, human-like consciousness, social presence, natural conversational fluency, and socio-emotional communicative abilities to artificial intelligence systems operating within their professional sphere. Theoretical models of anthropomorphism suggest that humans intuitively map biological and psychological attributes onto non-human agents to make their actions predictable. In the workplace, perceived humanlikeness influences whether employees interact with an AI as a cold computational tool, a collaborative social teammate, or a deceptive entity potentially triggering the uncanny valley phenomenon.

2. Perceived Adaptability of AI

Perceived adaptability reflects an employee’s appraisal of an AI system’s capability to learn dynamically, adjust its operational parameters in real-time, integrate contextual modifications, and successfully function across diverse, unpredictable workplace environments without requiring constant manual reprogramming. In contrast to deterministic software that performs static, repetitive rules, an adaptable AI exhibits flexible problem-solving, cognitive plasticity, and iterative learning based on changing inputs, which influences whether employees trust the system in high-stakes, volatile task domains.

3. Perceived Quality of AI

This dimension evaluates the cognitive assessment of an AI system’s performance efficacy, precision, functional reliability, output accuracy, and operational consistency. It encompasses whether the algorithm produces dependable recommendations, minimizes computational error, processes complex information faster than manual alternatives, and meets or exceeds professional industry standards. Perceived quality forms the technical bedrock of systemic credibility and cognitive trust in machine-generated guidance.

4. AI Use Anxiety

AI use anxiety is an affective state characterized by apprehension, physiological tension, cognitive unease, intimidation, and fear experienced by a worker when confronted with the actual or anticipated necessity of interacting with, commanding, or depending upon an AI application. Drawing from established literature on computer anxiety and technostress, this subscale measures the dread of making irreversible operational errors, feelings of self-efficacy deficit, and emotional distress provoked by algorithmic complexity.

5. Job Insecurity

Distinct from generalized operational anxiety, this dimension measures the specific, threat-oriented appraisal that the deployment of artificial intelligence will lead to total job loss, involuntary occupational displacement, demotion, skill obsolescence, or substantial erosion of career advancement opportunities. Employees scoring high on this dimension view AI not as an assistive cognitive prosthetic, but as a direct computational rival poised to substitute human labor and disrupt career longevity.

6. Personal Utility of AI

Personal utility reflects an employee’s subjective evaluation of the instrumental value, productivity enhancement, and personal benefit derived from integrating AI into their daily vocational tasks. It encompasses appraisals of whether AI saves cognitive effort, automates mundane administrative burdens, optimizes personal time management, and elevates overall occupational competence and output quality. This subscale directly anchors the positive, performance-enhancing appraisal of emerging technological collaboration.

Theoretical Framework

The Attitudes Towards Artificial Intelligence at Work Scale is synthesized from several foundational theoretical frameworks in psychology, communication science, and management studies:

The Technology Acceptance Model (TAM) and UTAUT

The foundational architecture of the AAAW draws upon the classical Technology Acceptance Model proposed by Fred Davis (1989), which posits that perceived usefulness and perceived ease of use mediate the relationship between external system characteristics and behavioral intentions to use a technology. Park et al. (2024) significantly extended this paradigm by incorporating core tenets of the Unified Theory of Acceptance and Use of Technology (UTAUT; Venkatesh et al., 2003). While TAM and UTAUT conceptualize technology primarily through an instrumental lens, the AAAW recognizes that contemporary AI necessitates expanding “perceived usefulness” into granular constructs such as Personal Utility, Perceived Quality, and Perceived Adaptability to account for algorithmic autonomy.

The Computers Are Social Actors (CASA) Paradigm

Pioneered by Byron Reeves and Clifford Nass (1996), the CASA paradigm demonstrates that humans automatically apply social rules, relational heuristics, and interpersonal expectations to computers, algorithms, and automated interfaces, even when consciously aware that machines lack genuine consciousness. The AAAW leverages CASA to substantiate the Perceived Humanlikeness dimension. Because modern AI engages in natural language processing, contextual sentiment detection, and expressive dialogue, workers instinctively judge AI systems through socio-cognitive lenses similar to those used for human coworkers and supervisors.

Cognitive Appraisal Theory of Stress and Coping

Formulated by Richard Lazarus and Susan Folkman (1984), Cognitive Appraisal Theory states that emotional and behavioral outcomes result from a two-stage evaluation: primary appraisal (evaluating an environmental encounter as benign-positive, irrelevant, or stressful/threatening) and secondary appraisal (evaluating one’s coping resources and control options). In the AAAW framework, the introduction of workplace AI represents a critical environmental stressor. AI Use Anxiety and Job Insecurity constitute threat appraisals wherein the demands of working alongside AI are judged to exceed an individual’s perceived technological self-efficacy or jeopardize resource security.

Conservation of Resources (COR) Theory and JD-R Model

Stevan Hobfoll’s (1989) Conservation of Resources (COR) Theory posits that individuals are fundamentally driven to acquire, preserve, and protect valued resources, including employment, professional autonomy, and psychological well-being. Threat of resource loss is a primary driver of acute strain. Concurrently, the Job Demands-Resources (JD-R) model (Demerouti et al., 2001) categorizes working conditions into job demands (aspects requiring sustained physical or psychological effort) and job resources (aspects functional in achieving work goals and stimulating personal development). Within the AAAW architecture, AI functions ambivalently: it can act as a potent job resource (elevating personal utility and productivity) or as an overwhelming job demand that generates cognitive strain, anxiety, and perceived resource depletion via job insecurity.

Validity

The construct validation of the AAAW was conducted across three distinct, methodologically rigorous phases involving independent samples of working adults from diversified industrial sectors, ensuring wide generalizability:

Construct and Structural Validity

Structural validity was verified using both exploratory and confirmatory factor analytic approaches. The underlying six-dimensional structure emerged robustly across samples, demonstrating that attitudes toward AI at work are inherently multifactorial. Confirmatory factor analyses (CFA) demonstrated superior fit indices for the hypothesized oblique six-factor solution compared to competing unidimensional, two-factor (positive vs. negative), and second-order hierarchical configurations. The distinctiveness of each latent factor confirms that workers hold nuanced, simultaneous positive appraisals and negative concerns regarding workplace AI.

Convergent Validity

Convergent validity was empirically supported by examining the Average Variance Extracted (AVE) across all six dimensions. Following standard psychometric criteria established by Fornell and Larcker (1981), convergent validity is substantiated when the AVE for each latent construct equals or exceeds the recommended threshold of .50. In the AAAW validation studies, the empirical AVE values for all six subscales ranged from .52 to .77. Furthermore, standardized factor loadings across all 25 finalized items were statistically significant (p < .001) and uniformly high (predominantly > .70), demonstrating that each item accounts for substantial shared variance within its designated psychological domain.

Discriminant Validity

Discriminant validity was established through multiple rigorous statistical evaluations. First, the square root of the AVE for every individual dimension was confirmed to be substantially larger than any bivariate correlation between that dimension and any other latent factor in the model (satisfying the classical Fornell-Larcker criterion). Second, the heterotrait-monotrait ratio of correlations (HTMT) remained well below the conservative cutoff value of .85, verifying that constructs such as AI Use Anxiety and Job Insecurity—while positively correlated—represent statistically and conceptually distinct negative psychological states rather than redundant measurements of generalized occupational distress.

Criterion-Related and Predictive Validity

Criterion-related validity was established by testing the AAAW’s ability to account for unique variance in objective and subjective organizational outcomes. Specifically, in simulated and real-world personnel recruitment environments, the subscales demonstrated robust predictive power regarding applicant reactions, perceived fairness of AI-mediated evaluations, and intentions to accept job offers. Employees exhibiting higher scores on Personal Utility and Perceived Quality demonstrated significantly higher voluntary adoption rates of enterprise AI tools, whereas elevated scores on Job Insecurity and AI Use Anxiety predicted technological avoidance, lower psychological safety, and heightened burnout.

Reliability

The reliability of the Attitudes Towards Artificial Intelligence at Work Scale was confirmed across multiple independent empirical samples using classical test theory metrics and contemporary structural equation modeling parameters:

Internal Consistency

Internal consistency estimates were computed across all six subscales in three separate validation cohorts. Cronbach’s alpha (α) coefficients for all six dimensions consistently exceeded the standard .80 benchmark, demonstrating strong internal consistency across samples. The observed alpha values were consistently distributed as follows:

• Perceived Humanlikeness of AI: α > .83
• Perceived Adaptability of AI: α > .82
• Perceived Quality of AI: α > .86
• AI Use Anxiety: α > .88
• Job Insecurity: α > .89
• Personal Utility of AI: α > .85

Composite Reliability

Because Cronbach’s alpha can be vulnerable to violations of tau-equivalence, the authors calculated Composite Reliability (CR) indices derived from the confirmatory factor analytic models. The CR estimates across the six latent factors ranged from .70 to .94, safely surpassing the accepted psychometric adequacy benchmark of .70. These high CR values confirm that the designated indicator items reliably measure their corresponding latent constructs with minimal random measurement error.

Measurement Invariance

The scale exhibited cross-sample stability and measurement invariance across diverse occupational categories, tenure levels, and demographic profiles. Multigroup CFA testing established configural, metric, and scalar invariance across gender and age cohorts, confirming that observed variance across participant groups reflects genuine differences in attitudes rather than measurement bias or variable item interpretations.

Factor Analysis

The structural composition of the AAAW was rigorously developed through a two-stage analytic process combining Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA):

Exploratory Factor Analysis (EFA)

Initial item generation yielded a wide pool of candidate items extracted from established human-computer interaction, technology adoption, and organizational behavior literature. During the preliminary exploratory phase, an EFA utilizing maximum likelihood extraction with oblique (promax) rotation was conducted to allow latent constructs to correlate naturally. Scree plot inspection, parallel analysis, and eigenvalue criteria (λ > 1.0) supported a six-factor solution. Items exhibiting cross-loadings greater than .30 on secondary factors, low primary loadings (< .50), or conceptual redundancy were systematically eliminated. This process refined the initial pool to an interpretable and parsimonious 43-item solution, which was subsequently streamlined into the final 25-item psychometric scale.

Confirmatory Factor Analysis (CFA)

The finalized 25-item structure was subsequently validated in an independent holdout sample of working adults using Confirmatory Factor Analysis. The hypothesized oblique first-order six-factor model yielded exceptional fit indices, confirming that the theoretical architecture adhered precisely to the empirical covariance structure of the data:

• Chi-Square Goodness-of-Fit: χ² = 673.60
• Degrees of Freedom: df = 260
• Comparative Fit Index (CFI): 0.98 (exceeding the ≥ .95 standard for superior fit)
• Tucker-Lewis Index (TLI): 0.97 (exceeding the ≥ .95 benchmark)
• Normed Fit Index (NFI): 0.96 (exceeding the ≥ .95 benchmark)
• Root Mean Square Error of Approximation (RMSEA): 0.04 (90% CI [0.036, 0.045], well below the ≤ .06 threshold for excellent fit)
• Standardized Root Mean Square Residual (SRMR): 0.04 (well below the ≤ .08 threshold for close fit)

Model Comparisons

To confirm that the first-order six-factor solution represented the optimal structural configuration, the researchers tested competing nested models against the baseline. A unidimensional model (where all 25 items loaded on a single general AI attitude factor) exhibited unacceptable fit (χ² = 3841.12, df = 275, CFI = 0.62, RMSEA = 0.14). A two-factor model differentiating merely positive attitudes from negative attitudes also proved psychometrically inadequate. Although a second-order hierarchical model (with a general higher-order AI attitude construct driving the six primary factors) showed acceptable fit parameters, the first-order six-factor model exhibited significantly superior statistical fit and greater diagnostic utility, demonstrating that each dimension possesses unique explanatory power.

Instrument / Measurement Tool

The practical administration characteristics, structural format, and scoring instructions for the Attitudes Towards Artificial Intelligence at Work Scale are summarized below:

• Instrument Name: Attitudes Towards Artificial Intelligence at Work Scale (AAAW)
• Primary Authors: Jiyoung Park, Sang Eun Woo, and JeongJin Kim (2024)
• Instrument Type: Psychometric Self-Report Questionnaire / Multi-Dimensional Rating Inventory
• Target Population: Working adults, corporate employees, professionals, and job candidates across all organizational sectors where artificial intelligence tools are active or anticipated
• Total Item Count: 25 items distributed across 6 subscales
• Subscale Breakdown:
  • Perceived Humanlikeness of AI (measures anthropomorphic attribution)
  • Perceived Adaptability of AI (measures flexible, contextual machine learning capabilities)
  • Perceived Quality of AI (measures functional accuracy, performance, and reliability)
  • AI Use Anxiety (measures affective tension, intimidation, and apprehension)
  • Job Insecurity (measures fear of occupational displacement and career threat)
  • Personal Utility of AI (measures perceived individual productivity and efficiency gains)
• Authentic Response Scale: 5-point Likert scale:
  • 1 = Strongly Disagree
  • 2 = Disagree
  • 3 = Neither Agree nor Disagree
  • 4 = Agree
  • 5 = Strongly Agree
• Administration Format: Electronic (online survey platforms, workplace management portals) or paper-and-pencil inventory
• Administration Time: Approximately 5 to 8 minutes
• Scoring and Interpretation Procedures:
  • Calculate separate mean or sum scores for each of the six individual subscales.
  • Higher scores on Perceived Humanlikeness, Perceived Adaptability, Perceived Quality, and Personal Utility indicate favorable, constructive appraisals of workplace AI capabilities and benefits.
  • Higher scores on AI Use Anxiety and Job Insecurity reflect heightened psychological strain, perceived existential threat, and technological aversion.
  • A composite total score is generally not advised, as combining divergent constructs (e.g., anxiety and perceived quality) obscures distinct psychological dynamics; profiling across the six distinct subscale dimensions provides the highest diagnostic precision.

Permissions & Fee and Test Year

The Attitudes Towards Artificial Intelligence at Work Scale (AAAW) was published in 2024 in the Journal of Occupational and Organizational Psychology. The scale was developed under standard scholarly academic conventions.

Licensing and Academic Usage:

• Academic Research: The scale is available for non-commercial academic research, empirical investigations, university-based dissertations, and institutional teaching purposes without fee, provided that appropriate scholarly attribution and standard academic citations are given to Park, Woo, and Kim (2024).
• Commercial and Corporate Use: Any commercial deployment, inclusion in proprietary diagnostic platforms, corporate consulting initiatives, or fee-generating human resource management tools requires formal permission from the copyright holders (the authors and the publisher, John Wiley & Sons Ltd on behalf of The British Psychological Society).
• Original Publication Reference: Park, J., Woo, S. E., & Kim, J. (2024). Attitudes towards artificial intelligence at work: Scale development and validation. Journal of Occupational and Organizational Psychology, 97(3), 920–951. https://doi.org/10.1111/joop.12502

References

• Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340. https://doi.org/10.2307/249008
• Demerouti, E., Bakker, A. B., Nachreiner, F., & Schaufeli, W. B. (2001). The job demands-resources model of burnout. Journal of Applied Psychology, 86(3), 499–512. https://doi.org/10.1037/0021-9010.86.3.499
• Fornell, C., & Larcker, D. F. (1981). Evaluating structural equation models with unobservable variables and measurement error. Journal of Marketing Research, 18(1), 39–50. https://doi.org/10.1177/002224378101800104
• Hobfoll, S. E. (1989). Conservation of resources: A new attempt at conceptualizing stress. American Psychologist, 44(3), 513–524. https://doi.org/10.1037/0003-066X.44.3.513
• Lazarus, R. S., & Folkman, S. (1984). Stress, appraisal, and coping. Springer Publishing Company.
• Park, J., Woo, S. E., & Kim, J. (2024). Attitudes towards artificial intelligence at work: Scale development and validation. Journal of Occupational and Organizational Psychology, 97(3), 920–951. https://doi.org/10.1111/joop.12502
• Reeves, B., & Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people and places. Cambridge University Press.
• Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425–478. https://doi.org/10.2307/30036540

Items of the Scale