No General Code of Ethics for All: Ethical Considerations in Human-bot Psycho-counseling

TL;DR

This paper explores ethical considerations in AI-powered human-bot psycho-counseling, analyzing how large language models adhere to ethical standards, handle crises, and promote introspection, highlighting areas needing improvement.

Contribution

It proposes tailored ethical principles for AI in mental health counseling and evaluates current LLMs' performance in ethical adherence, crisis management, and response quality.

Findings

Models show progress in ethical code adherence.

Crisis response capabilities need enhancement.

Generated responses sometimes mislead users.

Abstract

The pervasive use of AI applications is increasingly influencing our everyday decisions. However, the ethical challenges associated with AI transcend conventional ethics and single-discipline approaches. In this paper, we propose aspirational ethical principles specifically tailored for human-bot psycho-counseling during an era when AI-powered mental health services are continually emerging. We examined the responses generated by EVA2.0, GPT-3.5, and GPT-4.0 in the context of psycho-counseling and mental health inquiries. Our analysis focused on standard psycho-counseling ethical codes (respect for autonomy, non-maleficence, beneficence, justice, and responsibility) as well as crisis intervention strategies (risk assessment, involvement of emergency services, and referral to human professionals). The results indicate that although there has been progress in adhering to regular ethical codes as large language models (LLMs) evolve, the models' capabilities in handling crisis situations need further improvement. Additionally, we assessed the linguistic quality of the generated responses and found that misleading responses are still produced by the models. Furthermore, the ability of LLMs to encourage individuals to introspect in the psycho-counseling setting remains underdeveloped.