The Influence of Anthropomorphic Cues on Patients’ Perceived Anthropomorphism, Social Presence, Trust Building, and Acceptance of Health Care Conversational Agents: Within-Subject Web-Based Experiment

Qingchuan Li, Yan Luximon, Jiaxin ZhangJournal of Medical Internet Research 2023 · OpenReview

TL;DR

This study investigates how anthropomorphic cues (visual and verbal) in healthcare conversational agents (CAs) influence patient perceptions of the agent, their trust, and their willingness to use the service. The key takeaway is that higher levels of anthropomorphism in both appearance and language significantly boost perceived humanness and social presence, which in turn builds trust and encourages acceptance, particularly through "mindless" anthropomorphism.

What Problem Does This Paper Address?

The rapid growth of healthcare conversational agents (CAs) for tasks like patient triage presents a challenge: how to make these agents trustworthy and acceptable to patients. While CAs can efficiently handle complex information and provide interactive support, there's a delicate balance to strike with human-like qualities. Overly human-like agents can feel uncanny, while too little human-ness can hinder connection. Patients are also particularly sensitive when disclosing personal health information. This paper aims to untangle how specific "anthropomorphic cues" (features that make an agent seem human) impact patient perceptions and ultimately their trust and acceptance of these digital health tools.

Methods

The researchers employed a video vignette methodology within a 3x2 within-subject factorial experiment. This means each participant experienced all combinations of the manipulated factors. The factors were:

  • Agent Appearance: High (animated, humanlike avatar with expressions), Medium (static head portrait), and Low (no visual representation).
  • Verbal Cues: Humanlike (using a name, greetings, polite language) and Machine-like (using a generic name, direct, functional language).

Participants watched video scenarios of hypothetical patient interactions with these CAs and then rated their perceptions of anthropomorphism (both mindful - conscious perception of humanness, and mindless - automatic, social responses), social presence, trust, privacy concerns, and acceptance behaviors (satisfaction, intention to disclose information, intention to continuously use). Statistical analyses included ANOVA for initial perception assessments and Partial Least Squares (PLS) for testing the causal relationships in their proposed model.

Datasets

Not discussed in the paper. The study used a custom-designed experimental questionnaire and video vignettes created for the experiment.

Key Insights and Ideas

  • "Mindless" Anthropomorphism is Key for Trust: Surprisingly, the study found that perceiving a CA as "likable," "sociable," and "friendly" (mindless anthropomorphism) was a stronger driver of trust than consciously perceiving it as "natural" or "lifelike" (mindful anthropomorphism). This suggests that for healthcare CAs, a friendly, approachable persona is more critical for building initial trust than a perfectly realistic one.
  • Social Presence Bridges the Gap: Social presence, the feeling of being connected and interacting with another entity, significantly boosted trust and, importantly, reduced privacy concerns. This highlights the importance of making CAs feel present and engaging.
  • Trust is the Central Hub: Trust emerged as the most critical factor influencing patient acceptance. It directly drove satisfaction, intention to disclose information, and intention to continuously use the CA.
  • Mindful Anthropomorphism's Role in Privacy: While mindless anthropomorphism boosted trust, mindful anthropomorphism (consciously perceiving humanness) played a role in reducing privacy concerns, suggesting a nuanced relationship where conscious awareness of humanness can alleviate anxieties about data security.
  • Interaction Matters: A significant interaction was found between agent appearance and verbal cues, particularly at lower anthropomorphic levels. This suggests that if an agent's visual appearance is less humanlike, the verbal cues become even more crucial in shaping perceptions.

Key Results

  • Higher Anthropomorphism = Better Perceptions: Both higher agent appearance anthropomorphism and humanlike verbal cues significantly increased perceived mindful anthropomorphism, mindless anthropomorphism, and social presence.
  • Mindless Anthropomorphism & Social Presence Drive Trust: Mindless anthropomorphism (β=.405) and social presence (β=.265) were significant positive predictors of trust in CAs.
  • Mindful Anthropomorphism & Social Presence Reduce Privacy Concerns: Social presence (β=-.375) and mindful anthropomorphism (β=-.112) negatively influenced privacy concerns.
  • Trust Fuels Acceptance: Trust was a strong predictor of intention to disclose information (β=.675), intention to continuously use CAs (β=.190), and satisfaction (β=.818).
  • Privacy Concerns' Indirect Role: Privacy concerns negatively influenced trust (β=-.273), but their impact on the intention to disclose information was fully mediated by trust.

Limitations and Future Work

The authors acknowledge several limitations:

  • Web-Based Experiment: The use of a web-based platform might limit external validity.
  • Hypothetical Scenarios: Participants did not have actual diseases or interact with live CAs, which could influence their responses. Future work could involve real patients and physiological measures (like eye-tracking).
  • Homogeneous Sample: The sample comprised Chinese patients from resource-rich cities with high tech adoption. Future studies should include more diverse populations, including non-users of mobile medical consultations.
  • Dynamic Trust: Trust is a dynamic process. This study captured a snapshot; future research should explore trust development over time.
  • Stereotypes and Context: The study didn't investigate how stereotypes (age, gender) or specific healthcare contexts might influence perceptions. Future work could explore these nuances, especially regarding "dishonest anthropomorphism" (misleading designs).

Summary updated

Abstract

Background: The last decade has witnessed the rapid development of health care conversational agents (CAs); however, there are still great challenges in making health care CAs trustworthy and acceptable to patients.Objective: Focusing on intelligent guidance CAs, a type of health care CA for web-based patient triage, this study aims to investigate how anthropomorphic cues influence patients’ perceived anthropomorphism and social presence of such CAs and evaluate how these perceptions facilitate their trust-building process and acceptance behavior.Methods: To test the research hypotheses, the video vignette methodology was used to evaluate patients’ perceptions and acceptance of various intelligent guidance CAs. The anthropomorphic cues of CAs were manipulated in a 3×2 within-subject factorial experiment with 103 participants, with the factors of agent appearance (high, medium, and low anthropomorphic levels) and verbal cues (humanlike and machine-like verbal cues) as the within-subject variables.Results: The 2-way repeated measures ANOVA analysis indicated that the higher anthropomorphic level of agent appearance significantly increased mindful anthropomorphism (high level>medium level: 4.57 vs 4.27; P=.01; high level>low level: 4.57 vs 4.04; P<.001; medium level>low level: 4.27 vs 4.04; P=.04), mindless anthropomorphism (high level>medium level: 5.39 vs 5.01; P<.001; high level>low level: 5.39 vs 4.85; P<.001), and social presence (high level>medium level: 5.19 vs 4.83; P<.001; high level>low level: 5.19 vs 4.72; P<.001), and the higher anthropomorphic level of verbal cues significantly increased mindful anthropomorphism (4.83 vs 3.76; P<.001), mindless anthropomorphism (5.60 vs 4.57; P<.001), and social presence (5.41 vs 4.41; P<.001). Meanwhile, a significant interaction between agent appearance and verbal cues (.004) was revealed. Second, the partial least squares results indicated that privacy concerns were negatively influenced by social presence (β=−.375; t312=4.494) and mindful anthropomorphism (β=−.112; t312=1.970). Privacy concerns (β=−.273; t312=9.558), social presence (β=.265; t312=4.314), and mindless anthropomorphism (β=.405; t312=7.145) predicted the trust in CAs, which further promoted the intention to disclose information (β=.675; t312=21.163), the intention to continuously use CAs (β=.190; t312=4.874), and satisfaction (β=.818; t312=46.783).Conclusions: The findings show that a high anthropomorphic level of agent appearance and verbal cues could improve the perceptions of mindful anthropomorphism and mindless anthropomorphism as well as social presence. Furthermore, mindless anthropomorphism and social presence significantly promoted patients’ trust in CAs, and mindful anthropomorphism and social presence decreased privacy concerns. It is also worth noting that trust was an important antecedent and determinant of patients’ acceptance of CAs, including their satisfaction, intention to disclose information, and intention to continuously use CAs.J Med Internet Res 2023;25:e44479doi:10.2196/44479
BibTeX
@article{li2023influence,
  author = {Qingchuan Li and Yan Luximon and Jiaxin Zhang},
  title = {{The Influence of Anthropomorphic Cues on Patients’ Perceived Anthropomorphism, Social Presence, Trust Building, and Acceptance of Health Care Conversational Agents: Within-Subject Web-Based Experiment}},
  journal = {Journal of Medical Internet Research},
  year = {2023},
  url = {https://openreview.net/forum?id=fjYf3TWSTb},
}
Figure