Beyond Disposition: AI Knowledge Predicts Anthropomorphization of a Language Model Better Than Personality Traits in Lay and Expert Populations

Paper Title

Beyond Disposition: AI Knowledge Predicts Anthropomorphization of a Language Model Better Than Personality Traits in Lay and Expert Populations

Publication Info

• Topic area: Human-AI interaction and anthropomorphism
• Keywords: Anthropomorphism, AI knowledge, personality traits, AI literacy, human-computer interaction, large language models, LaMDA, moral care, intention to use, AI expertise

Background and Problem

• Problem / challenge: Anthropomorphism of AI systems is widespread but poorly understood, particularly regarding the role of dispositional traits versus AI knowledge in shaping these perceptions.
• Significance: Understanding anthropomorphism is critical for responsible AI use, as it influences trust, adoption, and moral care for AI systems, potentially leading to ethical and practical challenges.
• Motivation and related work: Previous research suggests dispositional traits like loneliness, need for cognition, and need for structure may predict anthropomorphism, but evidence is mixed. AI knowledge has been proposed as a developmental factor to counteract anthropomorphic tendencies, yet empirical studies comparing its role to dispositional predictors are limited.

Solution

• Proposed approach: Comparative analysis of dispositional traits and AI knowledge as predictors of anthropomorphism in lay and expert populations using vignette-based studies.
• Novelty:
  1. Direct comparison of dispositional traits (need for cognition, need for structure, loneliness) and AI knowledge as predictors of anthropomorphism.
  2. Inclusion of both laypersons and AI experts to examine differences in anthropomorphic tendencies across populations.
  3. Investigation of anthropomorphism’s mediating role in shaping intention to use and moral care for AI systems.
• Procedure and key techniques:
  • Two vignette-based online studies with excerpts from a conversation between LaMDA and a Google engineer.
  • Measurement of anthropomorphism, AI knowledge, dispositional traits, and social perceptions using validated scales.
  • Path modeling to assess direct and mediated relationships between predictors, anthropomorphism, and behavioral outcomes.

Results

• Concrete findings:
  • AI knowledge negatively predicted anthropomorphism in both lay and expert samples (β = -0.28 for laypersons, β = -0.39 for experts).
  • Dispositional traits (need for cognition, need for structure, loneliness) did not significantly predict anthropomorphism.
  • Anthropomorphism positively predicted moral care (β = 0.50 for laypersons, β = 0.39 for experts) but had mixed effects on intention to use (positive for laypersons, non-significant for experts).
• Advantage over baselines: AI knowledge emerged as a stronger predictor of anthropomorphism than dispositional traits, challenging prior assumptions in anthropomorphism theory.
• Experiments / evaluation:
  • Study 1: General public sample (N = 307) recruited via Prolific.
  • Study 2: AI expert sample (N = 130) from a master’s program in AI.
  • Measures included anthropomorphism, AI knowledge, dispositional traits, moral care, and intention to use.
  • Path models demonstrated good fit (CFI = 1.00, RMSEA < 0.01).
• Limitations and future work:
  • Correlational design limits causal inference; future experiments are needed.
  • Vignette-based paradigm may lack ecological validity; interactive studies are recommended.
  • Demographic differences between samples (e.g., age, gender) may confound results; matched samples are suggested.
  • Further exploration of subtle anthropomorphic cues and less human-like AI agents is needed.

Summary

This study demonstrates that AI knowledge is a stronger predictor of anthropomorphism than dispositional traits like need for cognition, need for structure, and loneliness, across both lay and expert populations. Higher AI knowledge consistently reduced anthropomorphic perceptions of LaMDA, while anthropomorphism influenced moral care and intention to use differently for laypersons and experts. The findings highlight the importance of AI literacy in shaping responsible interactions with anthropomorphic AI systems and suggest practical implications for policymakers, users, and designers to mitigate risks associated with human-like AI behavior. Future research should explore interactive settings, broader demographic samples, and the role of AI literacy interventions.