Evaluating AI Alignment Using Adapted Clinical Empathy Assessments

Cassandra Feilbach

Keywords: AI Alignment, Empathy, Human-AI interaction, Ethical AI

TL;DR: We adapt clinical empathy assessment tools to create a framework for evaluating AI alignment and understanding human-like social reasoning in AI systems.

Abstract:

We propose a novel approach to AI alignment evaluation by adapting a validated
human empathy assessment clinical tool for use with large language models and
other AI systems. The original assessment, designed to measure empathy in
humans, has been applied to AI to quantify a model’s potential alignment with
societal interests. Early experiments suggest the method provides a scalable,
repeatable baseline for AI empathy measurement, with implications for AI safety
and governance.

Submission Number: 6