Moral Association Graph: A Cognitive Model for Automated Moral Inference.

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Computational Linguistics

Background:

• Automated moral inference is crucial in AI, but current language model approaches are complex and computationally expensive.
• Existing methods often lack alignment with psychological theories of moralization.

Purpose of the Study:

• To introduce a novel, psychologically-inspired cognitive model for automated moral inference.
• To demonstrate the efficacy and interpretability of this new model compared to state-of-the-art methods.

Main Methods:

• Developed the Moral Association Graph (MAG), a cognitive model based on word association networks and psychological data.
• Evaluated MAG against comprehensive datasets for moral norm inference, concept moral judgment, and in-context moral inference.

Main Results:

• MAG demonstrates competitive performance against state-of-the-art language models in automated moral inference tasks.
• The model provides interpretable outputs, unlike many complex contemporary AI approaches.
• MAG shows potential for informing short-term moral change.

Conclusions:

• The Moral Association Graph (MAG) presents a viable, simpler, and psychologically-grounded alternative for automated moral inference.
• MAG's interpretability and performance highlight its utility in AI research and application.
• This model opens new avenues for understanding and potentially influencing moral dynamics.