Thesis Title
Compassionate AI: A Care Ethics Approach to Value Alignment in Generative AI

Research Description
This thesis, Compassionate AI: A Care Ethics Approach to Value Alignment in Generative AI, explores how care ethics and compassion can address the value alignment problem in generative AI.

Current alignment methods often neglect the relational and emotional dimensions essential to human well-being, especially in sensitive fields like mental health. This research argues that care ethics offers a more human-centered foundation for AI alignment, while compassion serves as both an ethical principle and a practical tool for guiding AI behavior.

Using mental health AI language agents as a case study, the thesis highlights the need for AI systems to recognize and respond to human vulnerability. Finally, it proposes a framework for designing compassionate AI, addressing ethical challenges such as bias, transparency, and user trust, and suggests future research directions for applying this approach across other domains.

Supervisors
Dr. Aaron Snoswell, QUT