Know Thyself, Improve Thyself: Personalized LLMs for Self-Knowledge and Moral Enhancement.

Area of Science:

• Artificial Intelligence
• Ethics
• Human-Computer Interaction

Background:

• Current artificial moral advisor (AMA) proposals often rely on static, predetermined values or require extensive introspective self-knowledge from users.
• Personal morality is dynamic and can shift over time, posing a challenge for existing AMA frameworks.

Purpose of the Study:

• To propose personalized large language models (LLMs) trained on individual data as a novel approach to artificial moral advising.
• To explore how LLM-based AMAs can account for the dynamic nature of personal morality and foster self-knowledge and self-creation.

Main Methods:

• Conceptual proposal for LLM-based AMAs trained on user-specific data (writings, interactions).
• Utilizing past and present user data to infer and articulate evolving values and preferences.
• Leveraging AI for reflection on personal identity, desired future self, and actionable goals.

Main Results:

• LLM-based AMAs could potentially offer personalized moral guidance by reflecting user's dynamic values.
• These systems may enhance users' self-knowledge and aid in processes of self-creation and personal development.
• The proposed approach addresses limitations of AMAs based on fixed values or demanding introspective capabilities.

Conclusions:

• Personalized LLMs offer a promising avenue for developing artificial moral advisors that adapt to individual moral evolution.
• Further technical development is needed to ascertain the feasibility of these LLM-based AMAs.
• This approach represents a significant advancement over static or introspection-reliant AMA systems.