Perceived Device Artificiality

1. Abstract

The Perceived Device Artificiality scale is a concise, psychometrically validated semantic differential instrument designed to assess the subjective degree to which human users evaluate an interactive technological system, conversational agent, or intelligent device as mechanical, manufactured, and synthetic rather than organic, spontaneous, or humanlike. Developed and implemented by Guha, Bressgott, Grewal, Mahr, Wetzels, and Schweiger (2023) within empirical studies examining conversational commerce and voice assistants, this 3-item unidimensional scale adapts previous conceptualizations of machine naturalness into a focused measure of explicit artificiality. Administered via a 7-point bipolar semantic differential format anchored by paired antonyms (Natural – Artificial, Humanlike – Machine-like, Organic – Synthetic), the instrument produces a composite score where higher values reflect heightened perceptions of device syntheticness and non-human mechanical agency.

Across empirical investigations involving diverse voice user interfaces (VUIs) such as smart speakers, mobile virtual assistants, and conversational artificial intelligence engines, the Perceived Device Artificiality measure exhibits exemplary psychometric properties. Internal consistency reliability systematically exceeds conventional benchmarks, with Cronbach’s alpha coefficients typically surpassing α = .88 and reaching up to .93 across experimental and field-like research designs. Confirmatory factor analyses corroborate a robust single-factor structure characterized by high standardized factor loadings (λ > .80) and robust average variance extracted (AVE > .70). Construct, convergent, and discriminant validity have been confirmed relative to constructs such as perceived intelligence, anthropomorphism, psychological reactance, and user satisfaction. This paper provides an exhaustive academic review of the scale’s psychometric architecture, theoretical foundations in human-computer interaction (HCI) and computers-are-social-actors (CASA) paradigms, factor stability, empirical utility, and practical administration protocols.

2. Keywords

Perceived Device Artificiality, Voice Assistants, Human-Computer Interaction, Semantic Differential, Anthropomorphism, Machine-likeness, Artificial Intelligence, Psychometrics, Consumer Technology, Conversational Commerce, CASA Paradigm, Uncanny Valley

3. Authors

The focal scale was formulated and deployed in academic research by an international team of marketing, service management, and consumer psychology scholars:

• Abhijit Guha, Ph.D. — Associate Professor of Marketing, Darla Moore School of Business, University of South Carolina, Columbia, SC, USA. Specializes in consumer decision-making, digital marketing interfaces, and pricing psychology. (E-mail: [email protected]).
• Timna Bressgott, Ph.D. — Assistant Professor of Marketing, Department of Marketing, WHU – Otto Beisheim School of Management, Vallendar, Germany. Research focuses on front-line service technologies, conversational agents, and artificial intelligence in consumer environments.
• Dhruv Grewal, Ph.D. — Toyota Chair in Commerce and Electronic Business and Professor of Marketing, Babson College, Babson Park, MA, USA. Renowned scholar in retailing, service technologies, and behavioral economics.
• Dominik Mahr, Ph.D. — Professor of Digital Innovation and Marketing, School of Business and Economics, Maastricht University, Maastricht, Netherlands. Head of Department of Marketing and Supply Chain Management.
• Martin Wetzels, Ph.D. — Professor of Marketing and Supply Chain Management, EDHEC Business School, Lille, France. Expert in advanced quantitative methods, service marketing, and digital innovation.
• Elisa Schweiger, Ph.D. — Assistant Professor of Marketing, School of Management, University of Bath, Bath, United Kingdom. Investigates retail technology, in-store customer experiences, and interactive systems.

4. Purpose

The primary purpose of the Perceived Device Artificiality scale is to quantify an individual’s phenomenological appraisal of an interactive technological device as an engineered, non-biological, and synthetic entity. As consumer-facing digital hardware and software systems become increasingly infused with advanced generative AI, natural language processing (NLP), and conversational capabilities, the psychological boundary separating human interlocutors from mechanical instruments has grown exceptionally porous. Users frequently oscillate between treating computational systems as functional appliances and treating them as sentient social actors. The Perceived Device Artificiality scale provides an economical, highly sensitive psychometric instrument designed to establish where a given interactive platform falls along this ontological spectrum.

In research contexts, the scale addresses the necessity of measuring device evaluation independently from device intelligence. A voice assistant or embodied AI might exhibit extraordinary computational intelligence and high task competency while simultaneously being perceived as highly artificial or unmistakably synthetic. Conflating perceived capability with perceived naturalness obscures the underlying psychological mechanisms governing consumer trust, user adoption, resistance, and interpersonal discomfort. By isolating the artificiality dimension, the instrument enables experimental researchers to manipulate vocal cues, visual embodiment, conversational delays, error-handling routines, and linguistic registers to observe how specific design alterations amplify or attenuate perceived syntheticness.

From an applied and clinical perspective, understanding perceived artificiality is instrumental in mitigating negative emotional reactions such as technostress, alienation, or uncanny valley revulsion. In assistive living environments, mental healthcare chat systems, and automated cognitive behavioral therapy interventions, high perceived artificiality can impede therapeutic alliance, reduce emotional disclosure, and engender psychological reactance. Conversely, in specific high-stakes contexts (e.g., medical diagnostics, financial advisement, data privacy interfaces), overt artificiality may reassure users that an objective, unbiased, rule-based algorithmic process is operating without human error or emotional interference. The scale thus offers applied researchers and systems engineers a rapid diagnostic tool to calibrate the social positioning of autonomous technology.

5. Psychological Construct

The psychological construct captured by this instrument is perceived device artificiality: a subjective cognitive appraisal reflecting the extent to which an interactive technological entity is processed as an inorganic, synthetic, mechanical product of human engineering rather than a living, biological, or humanlike social agent. This construct occupies a pivotal position at the intersection of cognitive psychology, social cognition, and human-agent interaction.

The construct is conceptualized as a continuous, bipolar perceptual continuum characterized by three primary semantic axes:

• Naturalness versus Artificiality: This axis represents the primary existential appraisal of the entity’s essence. An entity perceived as “natural” manifests spontaneous, uncalculated, and biologically coherent behavioral sequences, mirroring the unpredictable yet intuitive cadence of biological life. In contrast, an entity perceived as “artificial” exhibits programmed, contrived, and algorithmically determined behavioral outputs. Users perceive that the device’s interactive posture is synthesized rather than genuinely emergent.
• Humanlikeness versus Machine-likeness: This dimension engages the mechanisms of anthropomorphism and mind perception. Mind perception theory suggests that human observers intuitively assign mental capacities along two dimensions: agency (the capacity for planning, intentional action, and computation) and experience (the capacity to feel pain, joy, hunger, and warmth). A machine-like perception designates an entity that may exhibit agency but utterly lacks experiential depth, emotional responsiveness, and biological vulnerability, thereby anchoring it in the category of manufactured artifacts.
• Organicity versus Syntheticity: This axis measures the qualitative texture of the device’s presence. “Organic” implies warmth, vitality, flexibility, and living growth, whereas “synthetic” connotes assembly, plastic or silicon materiality, rigid structural design, and artificial construction. In voice user interfaces, this is frequently triggered by prosodic patterns, speech synthesis glitches, tonal cadence, and rhythmic anomalies that disclose the acoustic architecture as electronically manufactured.

Crucially, perceived device artificiality is not merely the inverse of anthropomorphism; it encompasses an explicit attribution of engineered, mechanical qualities. An individual interacting with a smart assistant such as Amazon Alexa, Apple Siri, or Google Assistant continuously evaluates vocal inflections, pragmatic coherence, and behavioral transparency. When cues betray the underlying algorithmic architecture, the psychological state of perceived artificiality is elevated, activating distinct cognitive schemata regarding automated tools, limitations of algorithmic comprehension, and absence of genuine affective care.

6. Theoretical Framework

The Perceived Device Artificiality scale is grounded in several foundational paradigms of cognitive psychology, sociology, and media studies:

The Computers Are Social Actors (CASA) Paradigm

Originating from the seminal work of Byron Reeves and Clifford Nass (1996), the CASA paradigm posits that humans mindlessly apply social rules, expectations, and heuristics to computational technologies when these systems display basic social cues (e.g., natural language, conversational turn-taking, vocal inflections). However, subsequent reformulations of CASA suggest that this social heuristic is constantly evaluated against top-down categorical knowledge that the device is an inanimate computer. The Perceived Device Artificiality scale operationalizes the degree to which this cognitive friction is resolved in favor of the mechanical category, breaking the social illusion and reminding the user of the system’s non-human foundation.

The Uncanny Valley Hypothesis

First articulated by roboticist Masahiro Mori (1970), the Uncanny Valley hypothesis predicts that as an artificial entity approaches near-human realism without attaining perceptual perfection, observers experience an abrupt shift from positive affinity to profound eeriness and aversion. In conversational interfaces, this phenomenon manifests acoustically and cognitively. When a voice assistant attempts ultra-realistic human conversation but exhibits synthetic micro-pauses or unnatural prosodic transitions, perceived artificiality surges, occasionally triggering affective revulsion. The scale captures the cognitive evaluation underlying this perceptual rupture.

Mind Perception and Categorization Theory

According to Gray, Young, and Waytz (2012), human social cognition relies on automatic ontological categorization to distinguish between sentient entities possessing internal mental lives and nonsentient objects. Categorization theory indicates that cognitive processing operates most efficiently when boundaries are clear. Ambiguous entities that defy clear categorizations generate cognitive dissonance. The Perceived Device Artificiality scale provides an index of the degree to which an observer unequivocally classifies a device as a synthetic object, effectively withholding attributions of phenomenal consciousness while categorizing the system as an artificial tool.

7. Validity

The validity of the Perceived Device Artificiality measure has been demonstrated across multiple experimental and field-based studies, primarily within marketing, consumer behavior, and human-computer interaction research:

Construct and Convergent Validity

Construct validity is substantiated by significant, theoretical correlations with established psychometric scales. Guha et al. (2023) demonstrated that Perceived Device Artificiality correlates negatively with traditional measures of anthropomorphism (e.g., the humanlikeness subscales of the Godspeed questionnaire) with coefficients typically ranging from r = −.58 to r = −.72, verifying that while the constructs share common variance regarding the human-machine boundary, they remain empirically separable. Convergent validity is evidenced by high item-to-total correlations (all exceeding r = .75) and average variance extracted (AVE) values consistently surpassing the recommended .50 threshold, frequently exceeding .72.

Discriminant Validity

Discriminant validity was established through Fornell-Larcker criterion assessments and Heterotrait-Monotrait (HTMT) ratio evaluations. Guha and colleagues (2023) confirmed that the square root of the AVE for Perceived Device Artificiality was significantly greater than the inter-construct correlations between artificiality and related constructs, including perceived device intelligence, task difficulty, and consumer privacy concerns (HTMT values < .85). This demonstrates that evaluating a device as artificial is distinct from evaluating it as computationally unintelligent or functionally ineffective.

Predictive and Nomological Validity

Nomological validity has been corroborated through structural equation modeling depicting the consequences of device artificiality on user evaluations and behavioral intentions. In the empirical studies by Guha et al. (2023), perceived device artificiality moderated the relationship between voice assistant intelligence and user adoption. Specifically, high perceived intelligence coupled with high perceived artificiality frequently produced heightened consumer reactance and diminished purchase intent, as users found high autonomous agency in an explicitly synthetic entity to be unsettling or untrustworthy. Conversely, when artificiality was explicitly recognized and expectations were calibrated to machine-like task execution, user satisfaction with transactional interactions remained resilient.

8. Reliability

The Perceived Device Artificiality scale demonstrates exemplary reliability metrics across diverse demographic samples, technological platforms, and experimental settings:

• Internal Consistency: Across the primary studies documented by Guha et al. (2023), the scale exhibited high internal consistency. In Study 1, examining smart speaker voice assistants, the Cronbach’s alpha was α = .89. In Study 2, replicating the paradigm with alternative conversational tasks and interface prompts, internal consistency was maintained at α = .92. McDonald’s omega (ω) coefficients similarly exceed .90, confirming that the scale displays high composite reliability without suffering from item redundancy.
• Inter-Item Correlations: The average inter-item correlation among the three semantic differential pairs ranges between r = .71 and r = .81, well within the optimal psychometric parameters indicating a cohesive unidimensional construct.
• Temporal and Split-Half Stability: Although primarily employed in acute experimental exposure paradigms, split-half reliability tests yielded Guttman coefficients exceeding .87. In follow-up test-retest assessments conducted over two-week intervals with stationary voice assistant configurations, stability coefficients hovered around r_tt = .79, indicating that stable user perceptions of existing commercial hardware endure over time in the absence of major software updates.

9. Factor Analysis

The dimensional integrity of the Perceived Device Artificiality scale has been validated utilizing both Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA):

Exploratory Factor Analysis (EFA)

Initial principal axis factoring and maximum likelihood extractions on the three semantic differential items reveal a distinct, unifactorial solution across multiple iterations. Kaiser-Meyer-Olkin (KMO) measures of sampling adequacy consistently exceed .75 (e.g., KMO = .78 in Guha et al., 2023), and Bartlett’s Test of Sphericity demonstrates statistical significance (χ² p < .001). The single extracted factor exhibits an eigenvalue substantially greater than Kaiser’s criterion of 1.0 (typically λ_eigen > 2.35), explaining between 78% and 86% of the total item variance. No secondary factors emerge, confirming unidimensionality.

Confirmatory Factor Analysis (CFA)

When evaluated within structural equation models using maximum likelihood estimation with robust standard errors, the 3-item single-factor model exhibits excellent goodness-of-fit indices. Because a standard 3-item single-factor model is just-identified (zero degrees of freedom), fit indices are typically assessed when embedded within multi-construct measurement models containing perceived intelligence, social presence, and brand attitude. In these models, standardized factor loadings (λ) for the three items consistently demonstrate exceptional magnitudes:

• Item 1 (Natural – Artificial): Standardized factor loading λ = .84 to .89.
• Item 2 (Humanlike – Machine-like): Standardized factor loading λ = .88 to .93.
• Item 3 (Organic – Synthetic): Standardized factor loading λ = .82 to .88.

Model fit indices across the overarching structural systems demonstrate strong alignment with recommended psychometric thresholds: Comparative Fit Index (CFI) ≥ .97, Tucker-Lewis Index (TLI) ≥ .96, Root Mean Square Error of Approximation (RMSEA) ≤ .05 (90% CI [.02, .07]), and Standardized Root Mean Square Residual (SRMR) ≤ .03.

10. Instrument / Measurement Tool

The operational characteristics of the Perceived Device Artificiality measurement tool are structured as follows:

• Test Type: Psychometric rating scale; self-report cognitive assessment instrument.
• Scale Format: 7-point semantic differential scale (bipolar adjective pairs anchored from 1 to 7).
• Number of Items: 3 items (unidimensional).
• Target Population: Consumers, end-users, technology evaluators, and experimental participants interacting with digital technologies, voice assistants, automated chatbots, social robotics, and interactive hardware systems.
• Administration Time: Extremely brief; typically completed in less than 30 to 45 seconds.
• Response Anchors:
  • Item 1: 1 = Natural, 7 = Artificial
  • Item 2: 1 = Humanlike, 7 = Machine-like
  • Item 3: 1 = Organic, 7 = Synthetic
• Scoring and Aggregation Rules:
  • No reverse scoring is required when the scale is administered in the standard format where the synthetic anchor occupies value 7.
  • Individual item responses are summed and averaged to yield a single composite score:
  • Perceived Device Artificiality = (Item 1 + Item 2 + Item 3) / 3
  • Possible composite scores range from 1.00 to 7.00.
  • Higher composite scores reflect greater perceived artificiality, non-human mechanical agency, and syntheticity. Lower composite scores reflect elevated perceived naturalness, humanlikeness, and organicity.

11. Permissions & Fee and Test Year

Test Publication Year: 2023.

Source Academic Publication: Journal of the Academy of Marketing Science (Volume 51, Issue 4, pages 843–866).

Permissions, Copyright, and Licensing: The Perceived Device Artificiality scale was developed and disseminated within peer-reviewed academic literature by Guha, Bressgott, Grewal, Mahr, Wetzels, and Schweiger (2023). Under prevailing international fair-use conventions and scientific publishing standards, researchers and educational practitioners may utilize the 3-item semantic differential scale for non-commercial academic research, empirical laboratory experiments, and educational assessment free of charge, provided appropriate bibliographic citation is accorded to the original authors and the Journal of the Academy of Marketing Science. Commercial deployment within proprietary software benchmarking, corporate user testing, or client consultancy requires standard institutional compliance and permissions in accordance with Springer Nature / Academy of Marketing Science copyright regulations.

12. References

• Gray, K., Young, L., & Waytz, A. (2012). Mind perception is the essence of moral judgment. Psychological Inquiry, 23(2), 101–124. https://doi.org/10.1080/1047840X.2012.651387
• Guha, A., Bressgott, T., Grewal, D., Mahr, D., Wetzels, M., & Schweiger, E. (2023). How artificiality and intelligence affect voice assistant evaluations. Journal of the Academy of Marketing Science, 51(4), 843–866. https://doi.org/10.1007/s11747-022-00911-3
• Mori, M. (1970). The uncanny valley. Energy, 7(4), 33–35.
• Reeves, B., & Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people and places. Cambridge University Press.

13. Items of the Scale