Argumentative Human-AI Decision-Making: Toward AI Agents That Reason With Us, Not For Us

AI should explain its reasoning through structured arguments rather than opaque outputs. The paper argues that combining computational argumentation with large language models creates systems where decisions are "contestable and revisable," allowing humans to understand and challenge AI reasoning rather than simply accepting recommendations.

The goal is collaborative decision-making, not AI replacement of human judgment. The paper emphasizes "reasoning with humans rather than for them," suggesting that trustworthy AI in high-stakes domains should engage in dialogue where humans can push back, ask questions, and ultimately retain decision-making authority rather than deferring to AI conclusions.

Current AI systems lack the transparency needed for high-stakes applications. While LLMs are good at processing text, the paper identifies their "opaque nature" as a fundamental barrier to trust and evaluation—a critical problem when AI is used in domains like healthcare, law, or policy where decisions significantly impact people's lives.

Combining two AI approaches could solve transparency problems. By merging computational argumentation (which provides formal, verifiable reasoning) with LLMs (which handle complex, unstructured information), the paper suggests we can build systems that are both capable and interpretable—addressing a key tension in responsible AI development.