Maarten de Rijke
Whether information retrieval should be done by a system that "writes the document ID" instead of by one that searches a vector index.
De Rijke has been one of the architects of generative retrieval — a methodological alternative to the standard "embed the query, search a vector index" paradigm, in which a sequence-to-sequence model is trained to directly emit the identifier of the relevant document, with no separate retrieval step. The question this opens is whether the retriever and the generator should ever have been separate components in the first place, or whether the whole pipeline was an artifact of how the field happened to evolve. For ai100, the consequence is structural: if generative retrieval keeps getting better, the assumption that "retrieval" and "generation" are distinguishable layers — which underlies most current evaluation — may not hold for the next generation of engines.