The Role of AI in Peer Support for Young People: A Study of Preferences for Human- and AI-Generated Responses

Title of the Paper

"The Role of AI in Peer Support for Young People: A Study of Preferences for Human- and AI-Generated Responses"

Paper Information

• Subject Area: Application of Artificial Intelligence in Adolescent Mental Health Support
• Keywords: Artificial Intelligence (AI), Chatbots, Large Language Models (LLM), Human-AI Interaction (HAII), AI-Mediated Communication (AI-MC), Mental Health, Peer Support, Social Support, Adolescents

Research Background and Issues

• Problems and Challenges:
  • Adolescents increasingly rely on online platforms for mental health support, but peers providing support often withdraw due to lack of experience or boundaries.
  • Emerging generative AI technologies (e.g., ChatGPT and Large Language Models) are rapidly integrated into everyday tools, potentially impacting this field significantly, yet their potential and limitations require further exploration.
• Significance:
  • The prevalence of adolescent mental health issues, such as depression and anxiety, has become a significant challenge, and AI may offer a feasible solution to provide support and reduce barriers.
  • Supportive communication on sensitive topics like suicide and anxiety requires deeper consideration of social cognition and ethics.
• Motivation and Related Work:
  • Existing research highlights the benefits of supportive communication on social media, including anonymity, ease of use, immediacy, and the ability to connect with individuals with similar experiences.
  • Generative AI has demonstrated potential in helping humans provide more empathetic responses (e.g., specific platforms like Crisis Text Line), but it still faces challenges such as misunderstanding subtle human emotions and potential algorithmic biases.

Research Objectives and Questions

The authors propose the following two research questions (RQs):

1. What are adolescents' preferences for human- and AI-generated supportive responses?
2. How do adolescents' preferences vary based on the topic of the initial help-seeking information (e.g., relationships, depressive tendencies, self-expression, or physical health)?

Solution

• Methodology:
  • Surveyed 622 adolescents aged 18-24 through an online questionnaire to examine their preferences for four sources of supportive responses (AI, peers, adult mentors, and clinical psychologists).
  • Four help-seeking scenarios: relationship issues, suicidal tendencies, difficulties with self-expression, and physical health problems.
  • Combined quantitative questionnaires (e.g., "I like this response") with open-ended questions to understand underlying reasons, using a blind testing design to reduce source bias.
• Data Analysis Strategy:
  • Employed correlation analysis, analysis of variance (ANOVA), and qualitative thematic analysis to extract the deeper logic behind adolescents' preferences.

Experimental Results and Analysis

Quantitative Results

1. Overall Preferences:

   • Responses from adult mentors performed best, followed by AI-generated responses, while peer-generated responses were the least favored.
   • Regardless of the source, adolescents perceived a positive correlation between responses with "fewer issues" and those deemed "more helpful," with AI particularly excelling in topics related to "relationships, self-expression, and physical health."

2. Differences Across Topics:

   • Relationship Issues: Adolescents preferred AI-generated responses, considering them more constructive and empathetic, as well as gentler and more specific compared to others.
   • Suicidal Tendencies: For this sensitive topic, responses from adult mentors were most favored, as they demonstrated clear care and encouraged further interaction. AI-generated responses, which often directly recommended seeking professional help, were perceived as colder.
   • Self-Expression: AI-generated responses were highly appreciated for boosting confidence and providing practical suggestions.
   • Physical Health: AI-generated responses dominated again, with participants noting their ability to accurately identify issues and offer comprehensive advice.

Qualitative Results

Themes extracted from open-ended responses include:

• Characteristics of AI: Participants appreciated AI's ability to provide caring, specific advice and an encouraging tone. AI responses were described as "understanding" and "non-judgmental."
• Advantages of Adult Mentors: They were perceived as genuinely listening and offering advice based on life experience.
• Limitations of Peers and Psychologists: Peer responses were often considered "too brief" and lacking depth; psychologist responses were deemed overly professional and unnatural in the context of peer interactions.

Research Findings

• Key Conclusions:
  • AI-generated responses are particularly favored by adolescents for non-sensitive topics (relationships, self-expression, physical health).
  • For sensitive topics like suicide, human-generated responses, especially from adult mentors, are superior. AI should be seen as a supplementary tool rather than a complete replacement.
• Advantages and Significance:
  • This study provides empirical support for the role of AI in mental health support, demonstrating its potential to occupy a significant position in young people's online interactions.
  • Offers specific directions for designing AI support tools to optimize empathy and trustworthiness.
• Limitations and Future Directions:
  • The sample may not represent the entire U.S. adolescent population; broader demographic studies are needed.
  • Further experiments in dynamic interaction scenarios are required, rather than relying solely on static questionnaire environments.
  • Exploration of real-time monitoring of inappropriate AI-generated responses and solutions for unsafe effects is necessary.
  • Future research could investigate changes in trust levels when participants are aware that the content is AI-generated.

Practical Implications and Design Suggestions

1. Training Tools for Human Supporters: AI can provide optimized supportive suggestions to peers or mentors, helping them enhance empathy and communication skills.
2. Personalization and Transparency: Research how to adjust AI designs to better understand adolescents' needs while improving its ability to explain its reasoning to users.
3. Ethical Design in Sensitive Contexts: AI's response content and referral logic for topics like suicidal tendencies require extra caution.
4. Long-Term Effect Studies: Monitor whether AI-human interactions have positive or negative impacts on mental health over time.

Conclusion

Generative artificial intelligence holds great potential in the field of adolescent mental health support but requires careful consideration for sensitive issues. AI should be regarded more as a tool to assist humans rather than a complete replacement solution.