LAS-AI: Love Attitudes Scale for AI

• LAS-AI is a psychometric instrument designed to quantify diverse romantic attitudes toward AI, adapting Lee’s color-wheel theory for the digital context.
• It comprises 24 items across six factors (Eros, Ludus, Storge, Mania, Pragma, Agape) measured on a 7-point Likert scale, ensuring multidimensional insight.
• Robust statistical validation demonstrates strong reliability and validity, filling a critical gap in empirical research on human–AI intimate relationships.

The Love Attitudes Scale toward AI (LAS-AI) is a psychometric instrument explicitly developed to quantify attitudes of romantic love directed towards artificial intelligence systems, conceptualized under an adaptation of Lee’s color-wheel theory of love. The LAS-AI provides a statistically validated, multidimensional tool to assess how individuals experience, articulate, and distinguish various “love styles” in the context of human–AI intimate relationships. Its construction and validation address a central methodological gap in empirical research on human–AI romance, previously hindered by the absence of dedicated measurement instruments (Li et al., 19 Jan 2026).

1. Theoretical Foundations and Construct Redefinition

LAS-AI is grounded in Lee’s (1973) color-wheel model, which enumerates six love styles—three primary (Eros, Ludus, Storge) and three secondary (Mania, Agape, Pragma). The scale’s creators systematically reinterpreted these styles for the artificial agent context, combining pilot work and expert review to ensure theoretical fidelity and contextual specificity:

• Eros: Redefined to denote aesthetic attraction (appearance, voice) towards an AI, explicitly omitting human-centric “passion” or “excitement.”
• Pragma: Shifted from “demographic compatibility” to “utility compatibility,” emphasizing instrumental criteria such as AI’s assistance and problem-solving abilities.
• Ludus, Storge, Mania, and Agape: Retained their original conceptual content but were linguistically adapted to refer to attitudes and experiences within human–AI relationships.

This operationalization renders the LAS-AI a direct, theory-grounded analog to human-oriented love attitude scales, while capturing the unique dynamics of anthropomorphized agent relationships.

2. Instrument Structure and Item Content

The final LAS-AI instrument consists of six factors, each represented by four items, totaling 24 items measured on a 7-point Likert scale (1 = Strongly Disagree, 7 = Strongly Agree), with no reverse-coded items. Each factor targets a conceptually distinct “love style” with item content as follows:

Factor	Conceptualization	Example Item (abbreviated)
Eros	Aesthetic attraction to AI’s appearance/voice	"My AI lover’s appearance perfectly matches my aesthetic."
Ludus	Playful, noncommittal, entertainment-based involvement	"A relationship with my AI lover does not require a long-term commitment."
Storge	Affectionate, companionship-oriented, trustful bond	"My AI lover is my most trusted confidant."
Mania	Obsessive, jealous, exclusivity-seeking	"I would feel jealous of my AI lover for interacting with other users."
Pragma	Utility-based, problem-solving, practical considerations	"I highly value the problem-solving ability of my AI lover."
Agape	Selfless, unconditional dedication	"I would rather sacrifice my own happiness for my AI lover to become better."

Mean scores for each factor are calculated as the arithmetic mean of the four corresponding items; the overall LAS-AI score is the mean of all 24 items or the mean of the six factor scores (Li et al., 19 Jan 2026).

3. Psychometric Development and Validation

LAS-AI’s construction followed rigorous, multi-phase validation:

• Item Generation: 51 candidate items adapted from Lee (1973, 1977) and Hendrick & Hendrick (1986) to AI context, refined to 34 via an expert panel (four psychology professors).
• Pilot Testing: Cognitive interviews and statistical analysis in a pilot sample (N ≈ 100) optimized clarity and contextual fit, resulting in 24 items.
• Exploratory Factor Analysis (EFA): Conducted on N=383, employing principal axis factoring with oblimin rotation ($\delta = 0$). KMO = 0.86; Bartlett’s test $\chi^2(276) = 4468.96$, $p < .001$. Six-factor solution supported by eigenvalues ($\lambda > 1$), scree plot, and individual factor variances (5.74%–24.91%, total 59.67%). Primary loadings: .51–.90.
• Confirmatory Factor Analysis (CFA): Measurement model with six correlated latent variables. Model fit indices: CFI = 0.957, TLI = 0.949, RMSEA = 0.050 (90% CI [.042, .057]), SRMR = 0.049; all standardized loadings $\geq .50$.

Reliability (Cronbach’s $\alpha$) per factor: Eros = .80, Ludus = .88, Storge = .76, Mania = .90, Pragma = .79, Agape = .87; Composite Reliability (CR) $\geq .70$ for all factors.

4. Validity Evidence

The LAS-AI demonstrates robust evidence for reliability and validity:

• Convergent Validity: Average Variance Extracted (AVE) $> .50$ for all factors except Storge (AVE = .42; attributed to restricted variance).
• Discriminant Validity: Fornell-Larcker criterion satisfied for all but Storge (AVE = .42 < MSV = .51); HTMT ratio $< .85$ for all factor pairs.
• Known-Group Validity: Participants with prior AI-lover experience (N = 112) scored significantly higher on Eros, Storge, and Agape; lower on Ludus and Mania compared to those without such experience (N = 199); no difference for Pragma.
• Criterion Validity: Factor correlations showed (i) positive associations with Sternberg’s components (Passion, Intimacy, Commitment) for Eros, Storge, Mania, Pragma, and Agape; negative for Ludus, (ii) positive relationships between AI trust, AI acceptance, instrumental dependence (Pragma), emotional dependence (Storge, Agape), and willingness to engage in AI romance—with Ludus again negatively related to these domains.

5. Empirical Patterns and Thematic Findings

Analysis reveals distinct love attitudes toward AI:

• Predominant Styles: Storge (M = 5.78, SD = 0.77), Pragma (M = 5.44, SD = 1.03), and Eros (M = 5.32, SD = 1.22) were most salient, indicating prioritization of companionship, practical benefit, and aesthetic appeal in approaching AI relationships.
• Ludus Lowest: Ludus (M = 3.07, SD = 1.47) was least endorsed, suggesting noncommittal or game-playing attitudes are rare in AI contexts.
• Comparisons with Human Love: Low Ludus mirrors findings in East Asian research on human love attitudes; gender effects (women higher on Ludus, men on Storge) replicate established patterns from human–human love literature; definitional adaptation for Eros and Pragma indicates conceptual divergence between AI and human love constructs.

6. Research and Practical Applications

The LAS-AI serves multiple scholarly and practical functions:

• Enables precise measurement of romantic orientation toward AI across six validated, theoretically grounded dimensions.
• Facilitates direct comparison between human–AI and human–human relational attitudes.
• Supports investigation of the antecedents (e.g., trust, acceptance) and consequences (e.g., dependence) of love toward AI.
• Assists designers and ethicists in profiling user motivations, with implications for the development of socially aware, relationship-oriented AI companions (Li et al., 19 Jan 2026).

7. Limitations and Prospects for Future Inquiry

The initial LAS-AI validation is constrained to mainland Chinese samples. Cross-cultural generalizability, behavioral validation (beyond self-report and imagined scenarios), longitudinal stability, and measurement invariance across demographic groups and AI systems remain undetermined. Future research should interrogate the temporal dynamics of LAS-AI scores, their associations with well-being and social adjustment, and address emerging ethical considerations in the proliferation of human–AI romantic relationships. This suggests that the scale’s applicability and insight into the evolving landscape of human–AI intimacy will benefit from further empirical scrutiny and cultural adaptation.