Dr. Youngsam Chun Presents at the 2026 The Responsible AI Forum (TRAIF)

On May 19th, 2026, Generative AI, LLMs and Societal Value Alignment track travel grant winner Dr. Youngsam Chun presented at the 2026 The Responsible AI Forum (TRAIF) held at Amerikahaus in Munich. During the Philosophical and Technical Foundations of Value Alignment session, Dr. Chun presented: “MORAL-GRAPH: A Geometric and Graph-based Framework for Human-LLM Value Alignment”.

abstract excerpt: Large language models (LLMs) increasingly mediate ethically sensitive decisions in public services, healthcare, education and content moderation, shaping outcomes that directly affect social well-being and public trust. Despite substantial progress in preference learning and safety fine-tuning, it remains insufficiently understood how moral values are structurally encoded within embedding spaces and how divergences between human moral judgement and LLM outputs can be measured in an interpretable and policy-relevant manner. MORAL-GRAPH is a geometric and graph-based framework that represents moral judgement as structured projections onto explicit moral axes. Grounded in moral prototype theory, which conceptualizes moral categories as organized around central prototypes rather than fixed definitions, virtue–vice moral axes (for example, Care–Harm and Fairness–Cheating) are constructed using contrastive centroids derived from curated seed embeddings. By combining deterministic geometric modeling with participatory human evaluation, MORAL-GRAPH provides an interpretable and actionable approach for aligning generative AI systems with societal values and informing policy-oriented AI governance.

The Responsible AI Forum 2026 and travel grants were funded by the European Union. More on alignAI.

Watch a video of the talk below.