Information retrieval or document retrieval? Terminological confusions and unrealistic goals in information science, exemplified in relation to generative artificial intelligence

ChatGPT and related technologies have revived an old issue in information science (IS) concerning information retrieval (IR) versus document retrieval. Since 1950, the term IR has primarily been used as a misnomer for document retrieval. This problematic terminology reflects a desire to go beyond documents and provide, in response to user queries, not lists of documents but direct answers. Only with the emergence of large language models such as ChatGPT has the goal of directly informing users appeared to many as justifiable in relation to IR. Such models, however, still depend on input in the form of documents. A basic problem with large language models is their inability to establish a valid connection between their answers and the sources on which they are based. Whereas scholarly norms dictate that all claims be explicitly supported by the sources and arguments used, this cannot be done satisfactorily by ChatGPT, which represents a fundamental limitation of this technology. Neglecting the documentary basis in all forms of IR is naïve, and the core concept in IS should be understood as document retrieval. Recognizing this distinction is essential for enabling users to maintain control over the search and to perform “source criticism.”