The rise of generative AI has sparked debates across professions, including within the library and information sciences, about its potential to transform specific tasks. Among the many specialised tasks that information professionals perform, cataloguing and creating metadata records are particularly complex.
Many wonder whether AI tools like ChatGPT can perform cataloguing or create metadata with reasonable accuracy and reliability. Some have conducted experiments to test AI’s capability in this area. Such experiments include using ChatGPT for LCSH Subject assignment, identifying errors in catalogue records, creating MARC records using RDA or DC, and creating MARC21 using RDA.
Most of these experiments point to some potential and some shortcomings. As Moulaison-Sandy and Coble argue: “The notion that freely-available, general-purpose AI systems are able to solve cataloguing problems easily, with the click of a button, if only the right prompt is created, is problematic to perpetuate – at least for now.” They rightly highlight the limitations of AI in addressing the challenges of cataloguing. Moreover, there are also several key ethical considerations such as bias, privacy and transparency, that need to be addressed when using AI to create metadata.
Some enthusiasts have developed custom versions of ChatGPT for cataloguing. Here are three custom ChatGPT models that claim to assist with cataloguing tasks and generating MARC records. I personally haven’t tested their accuracy, but you can experiment with them. While there is no perfect AI solution for cataloguing yet, these tools are evolving, and companies that develop library software have started incorporating AI features in their products. However, there is still a long way to go.
- MARC-AI Cataloger: creates RDA-compliant MARC records.
- CatalogerGPT: creates MarcEDIT format MARC records from book contents, and determines LCSH, LCC and DDC numbers.
- Cataloging Assistant (DDC and LC Cutter): Catalogue library materials using DDC and LC Cutter.
