The Knowledge Placement Problem

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Keynote Presentation at the The 25th International Conference on Knowledge Engineering and Knowledge Management (EKAW 2026) https://ekaw2026.di.unito.it/invited-speakers Abstract Knowledge in modern AI systems resides in many substrates: model weights, prompts and context windows, retrieval indexes, agent memories, knowledge graphs and databases, callable tools, and people. Foundation models have expanded our ability to make knowledge machine-usable and to extract, transform, and transfer it between representations. Yet these transformations are costly and lossy. Deciding which knowledge belongs where, and when it should move, is now a pervasive design problem, addressed through fine-tuning, retrieval, editing, memory, and extraction. These decisions are often guided by trial and error, software defaults, and intuition, resulting in knowledge infrastructures that can be brittle, difficult to audit, and expensive to maintain. In this talk, I frame the knowledge placement problem as a central challenge for knowledge engineering and outline a research agenda for understanding how placement decisions shape the reliability, cost, provenance, and maintainability of AI systems. I argue that this is the example of the kind of problem that knowledge engineering as field should be addressing.