Presenter: Michael Weinold from ecoinvent Association
This talk examines where AI-assisted life cycle assessment collides with non-negotiable ISO and academic requirements for source traceability, pedigree-based uncertainty, and transparent documentation. It walks through the concrete failure modes of a typical LLM document pipeline - table and figure extraction, chunking, embedding retrieval, and citation hallucination during synthesis - and explains why these are architectural limits of LLMs rather than bugs to be patched.