The Effects of Warmth and Competence Perceptions on Users' Choice of an AI System

Title of the Paper

The Effects of Warmth and Competence Perceptions on Users’ Choice of an AI System

Bibliographic Information

• Subject Area: User choice and perception of AI systems
• Keywords: Artificial intelligence, warmth perception, competence perception, user choice, trust, human-computer interaction

Research Background and Problem

• Identified Problems or Challenges:

  • With the widespread application of artificial intelligence (AI) systems in recommendation and decision-support domains, users face the daily challenge of choosing among multiple AI systems, such as selecting a music streaming app or navigation software.
  • Traditional perspectives suggest that users primarily base their system choices on "competence," i.e., the system's strength or accuracy of performance.
  • However, social psychology research indicates that users' judgments of people and organizations are often influenced by the dimension of "warmth," such as perceptions of friendliness, sincerity, and trustworthiness.
  • Existing research has largely focused on improving AI competence, with limited studies on how users perceive and choose AI systems with varying levels of warmth and competence.

• Significance:

  • As AI becomes integrated into various fields (e.g., housing valuation, car insurance recommendations, movie recommendations), understanding how users choose systems based on warmth and competence dimensions is crucial for designing AI systems that better meet user needs and improve human-computer interaction success rates.
  • Additionally, whether users' preference for warmth can surpass the traditional emphasis on system competence remains an underexplored question.

• Research Motivation and Related Work:

  • Previous studies have shown that warmth and competence are two fundamental dimensions in social judgments, influencing perceptions of robots and other forms of artificial intelligence.
  • Social psychology suggests that warmth often takes precedence over competence in interpersonal judgments, while competence is prioritized in judgments of organizations.
  • This paper focuses on whether these findings apply to AI systems, particularly those without obvious social characteristics (e.g., navigation or recommendation systems without physical or virtual embodiments).

Proposed Solution

• Proposed Methods or Solutions:

  • Investigated how users' perceptions of AI systems' warmth and competence influence their choice behavior.
  • Conducted six scenario-based survey experiments, manipulating descriptions of AI systems' warmth and competence characteristics. The experimental scenarios covered three main domains: housing valuation, car insurance plans, and movie recommendations.
  • Tested the impact of explicit expressions of competence and warmth (e.g., "high competence/low warmth" vs. "high warmth/low competence" combinations) on user choices.

• Innovative Contributions:

  • Systematically demonstrated for the first time the influence of the "warmth" dimension on AI system choice, suggesting it may take precedence over the "competence" dimension.
  • Introduced a survey on users' perceptions of "humanization" and "organization" of systems, further elucidating the relationship between choice preferences and perceptions.
  • Provided practical insights into the implications of warmth prioritization for AI system design, offering profound suggestions on balancing user trust and system performance.

• Implementation Steps and Key Techniques:

  1. Scenario Design: Designed six independent experiments to test user choice behavior under combinations of high/low warmth and high/low competence.
  2. Manipulated Variables:
     • Warmth: Expressed through descriptions of the system's primary beneficiaries (e.g., users vs. brokers) and language tone (friendly vs. indifferent).
     • Competence: Reflected through algorithm types (advanced neural networks vs. basic decision trees) and training data scale (e.g., millions vs. thousands of data points).
  3. Measured user choice preferences, evaluations of system warmth and competence, and confidence levels post-choice.
  4. Examined the relationship between different domain scenarios and subsequent user perceptions of "humanization" and "organization," exploring psychological factors (e.g., regret likelihood) influencing user behavior.

Research Findings

• Specific Findings:

  1. Warmth-First Effect:
     • Across the three domains (housing valuation, car insurance recommendations, movie recommendations), users tended to choose AI systems with high warmth, even when those systems had lower competence.
     • When faced with conflicting dimensions (e.g., "high warmth-low competence" vs. "high competence-low warmth"), the majority of users still preferred high-warmth systems.
  2. Avoidance of Low-Warmth Systems:
     • Users were more inclined to avoid systems perceived as low in warmth, even if the alternative options had lower competence.
  3. Association Between Warmth and Humanization:
     • Users were more likely to perceive high-warmth systems as "human-like" rather than "organization-like."
     • Warm systems tended to influence positive evaluations of other performance aspects through a "halo effect" of perceived sincerity.
  4. Confidence Not Compromised:
     • Users who chose high-warmth systems reported confidence levels comparable to those who chose high-competence systems.

• Comparison with Existing Solutions and Advantages:

  • Challenged the traditional "competence-first" assumption, highlighting the importance of warmth as a judgment criterion.
  • Validated the cross-domain applicability of social psychology findings on perception behavior to AI systems.

• Experimental or Evaluation Results:

  • Across the six experiments, the warmth-first effect was statistically significant under most conditions (e.g., p<0.05).
  • Warmth-oriented design descriptions increased system selection rates among users (e.g., emphasizing user benefits or employing friendly language during system design).

• Limitations and Future Directions:

  • Limitations:
    • Limited expression methods for manipulating warmth and competence, not covering all possible characteristics (e.g., privacy protection or developer background).
    • Reliance on short-term laboratory studies; ecological validity needs verification in real-world applications.
    • In high-risk domains (e.g., healthcare), competence may take precedence.
  • Future Directions:
    • Test the dynamic effects of warmth and competence in longer-term, more complex interactions.
    • Design more comprehensive decision models incorporating factors such as service cost and data privacy.
    • Explore the moderating effects of personalized user traits (e.g., relationship-oriented vs. utility-oriented) on preferences for warmth and competence.

Conclusion

This study systematically analyzed the impact of perceptions of warmth and competence on AI systems, particularly highlighting the precedence of warmth in user choice trends. The findings reveal that users consider not only functional performance but also the social and moral intentions of systems when making choices. These results provide developers with new design perspectives for creating user-friendly AI, emphasizing the importance of demonstrating trustworthy intentions to enhance acceptance and user satisfaction.