The International Committee of the Red Cross (ICRC) has pointed out that AI systems can generate incorrect or completely false archival references and explains how to research correctly. This makes the renowned aid organization the latest in a growing list of institutions to publicly address the problems that AI hallucinations mean for them. If a document cannot be found, "it does not mean that the ICRC is withholding information," the organization assures in a recent statement. AI-generated references are a possible explanation why one should continue searching through other channels. Previously, complaints from other libraries had already become public.
In its statement, the ICRC feels compelled to point out that AI systems "do not research, do not verify sources, or check inf…
The International Committee of the Red Cross (ICRC) has pointed out that AI systems can generate incorrect or completely false archival references and explains how to research correctly. This makes the renowned aid organization the latest in a growing list of institutions to publicly address the problems that AI hallucinations mean for them. If a document cannot be found, "it does not mean that the ICRC is withholding information," the organization assures in a recent statement. AI-generated references are a possible explanation why one should continue searching through other channels. Previously, complaints from other libraries had already become public.
In its statement, the ICRC feels compelled to point out that AI systems "do not research, do not verify sources, or check information." The content is generated solely based on statistical patterns, which is why they "invent" catalog numbers, descriptions of documents, or even references to platforms: "Because their purpose is to generate content, they cannot indicate that information does not exist." Instead, they would "invent" details that appear plausible but have no basis in an archive. Reliable information can only be found in official catalogs, such as those of the ICRC, or directly in libraries.
Those responsible assure that they are committed to transparency and are available for inquiries and help with research. This at least indicates the difficulties that false references entail and the issues they pose for archives. Librarians had already pointed this out more explicitly several weeks ago to the US magazine 404. There was talk of a lot of confusion and a noticeable loss of trust, which were caused by non-existent titles from AI-generated texts. At the same time, the references also mean a lot of unnecessary extra work, as research has to be done for a while each time before it becomes clear that a title does not exist.
(mho)
Don’t miss any news – follow us on Facebook, LinkedIn or Mastodon.
This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.