Understanding Attitudes and Trust of Generative AI Chatbots for Social Anxiety Support

Research Background and Issues

• Identified Problems or Challenges: Social Anxiety (SA) is becoming increasingly prevalent, while traditional mental health support methods face challenges such as a shortage of professionals, healthcare limitations, and poor accessibility. Furthermore, although Generative AI (GenAI, e.g., intelligent chatbots) offers new interactive platforms, its trustworthiness and effectiveness in mental health applications remain questionable, particularly regarding the dynamics of trust in SA support, which have not been thoroughly explored.
• Research Significance: Understanding individuals' attitudes and trust toward GenAI chatbots in SA support is crucial for technological design, user adoption, and innovation in mental health interventions.
• Research Motivation and Related Work: While existing studies have explored methods to enhance the credibility of AI in mental health, most focus on the reliability and transparency of the models themselves, neglecting how trust varies based on users' personal experiences and application contexts. This research aims to fill the gap in understanding the dynamics of trust formation among SA users toward GenAI.

Solution

• Methods or Solutions: This study adopts a mixed-methods approach, combining surveys and interviews to deeply explore users' trust and willingness to use GenAI chatbots for SA support.
  • Survey: A questionnaire was administered to 159 participants to analyze the relationship between the severity of social anxiety, prior interactions with GenAI, and trust and willingness to use the technology.
  • Interviews: In-depth semi-structured interviews were conducted with 17 participants in groups to uncover nuanced characteristics of trust and attitude formation.
• Innovations:
  1. Proposed and validated a multidimensional trust measurement framework in the context of SA support (including emotional trust and cognitive trust).
  2. Differentiated user trust and adoption willingness based on symptom severity and usage experience.
  3. Combined quantitative and qualitative analyses to reveal the multidimensional nature of trust dynamics, including prioritization factors between technical trust and emotional support.
• Implementation Steps and Key Techniques:
  1. Designed and implemented the SPIN scale to measure SA severity.
  2. Combined Kruskal-Wallis tests and linear regression in survey analysis to predict the relationship between trust and willingness to adopt.
  3. Applied Gaussian Mixture Model (GMM) clustering to identify characteristics of groups with ambiguous attitudes.
  4. Used a general inductive approach to code and perform thematic analysis on interview results.

Research Findings

• Specific Findings:
  • Trust and Willingness to Use: There is a significant positive correlation between trust and willingness to use, especially among individuals with severe SA symptoms.
  • Types of Trust:
    1. Emotional Trust: For individuals with severe SA symptoms, users are drawn to the empathy and non-judgmental conversational tone of GenAI chatbots. These users prioritize emotional support, such as comfort and emotional understanding.
    2. Cognitive Trust: Individuals with milder SA symptoms place greater emphasis on the technical feasibility of GenAI, such as information accuracy, model reliability, and technical transparency.
  • Experience Differences: Users with extensive experience using GenAI are more inclined to adopt it, while those with no prior experience with the technology exhibit greater hesitation.
  • User Classification: GMM analysis identified three characteristic groups among neutral-attitude users:
    1. High-severity SA users reliant on emotional support.
    2. Long-term, high-frequency users with milder symptoms who prioritize technical feedback.
    3. Casual users without specific usage purposes, with weak associations to SA support.
• Comparison with Existing Solutions and Advantages:
  • Compared to single-dimensional trust studies, this research proposes a contextualized model of trust, offering a detailed analysis of the layered mechanisms of trust in emotional and cognitive dimensions.
  • Provides more tailored support strategy recommendations addressing the specific mental health needs of SA individuals.
• Experimental or Evaluation Results:
  • Both the survey and interviews confirmed the potential of GenAI tools in SA support, but limitations such as AI "hallucinations" and memory deficiencies need attention.
  • Participants with more severe symptoms showed higher levels of trust and acceptance of GenAI, closely linked to necessity-driven motivations.
  • Long-term and high-frequency users demonstrated higher trust levels, reflecting the impact of usage experience on adoption willingness.
• Limitations and Future Directions:
  • Limitations:
    1. Reliance on self-reported data, which may be subject to subjective bias or memory errors.
    2. Demographically concentrated sample, potentially affecting the generalizability of results.
    3. Lack of consideration for cultural differences in trust formation.
  • Future Research Directions:
    1. Further exploration and optimization of emotional trust design to enhance user experience and long-term outcomes.
    2. Cross-cultural comparative studies to validate the universality of trust dynamics.
    3. Incorporation of physiological and behavioral data to quantify the process of emotional trust formation.
    4. Exploration of new paradigms for collaboration between AI and human psychotherapists to address ethical risks in technology applications.

These findings provide critical references for the design and implementation of future mental health technologies and enrich research on the dynamics of trust between AI and humans.