Evidence Summary
A Review of:
Brzustowicz, R.
(2023). From ChatGPT to CatGPT: The Implications of Artificial Intelligence on
Library Cataloging. Information Technology and Libraries, 42(3). https://doi.org/10.5860/ital.v42i3.16295
Reviewed by:
Andrea Miller-Nesbitt
Associate Librarian
Schulich Library of Physical Sciences, Life
Sciences, and Engineering
McGill University
Montreal, Quebec, Canada
Email: andrea.miller-nesbitt@mcgill.ca
Received: 3 Mar. 2024 Accepted:
16 May 2024
2024 Miller-Nesbitt. This is an Open Access
article distributed under the terms of the Creative Commons‐Attribution‐Noncommercial‐Share
Alike License 4.0 International (http://creativecommons.org/licenses/by-nc-sa/4.0/), which permits
unrestricted use, distribution, and reproduction in any medium, provided the
original work is properly attributed, not used for commercial purposes, and, if
transformed, the resulting work is redistributed under the same or similar
license to this one.
DOI: 10.18438/eblip30524
Objective – To evaluate the potential of ChatGPT as a tool for improving
efficiency and accuracy in cataloguing library records.
Design
– Observational, descriptive study.
Setting
– Online, using ChatGPT and the WorldCat catalogue.
Subject –
The Large Language Model (LLM) ChatGPT.
Methods –
Prompting ChatGPT to create MARC records for items in different formats and
languages and comparing the ChatGPT derived records versus those obtained from
the WorldCat catalogue.
Main results – ChatGPT was able to generate MARC records, but the accuracy of
the records was questionable, despite the authors’ claims.
Conclusion – Based on the results of this study, the author concludes
that using ChatGPT to streamline the process of cataloging could allow library
staff to focus time and energy on other types of work. However, the results
presented suggest that ChatGPT introduces significant errors in the MARC
records created, thereby requiring additional time for cataloguers to correct
the error-laden records. The author correctly stresses that if ChatGPT were
used to assist with cataloguing, it would remain important for professionals to
check the records for completion and accuracy.
This article resulted in three separate
letters to the editor criticizing the methods and conclusions (DeZelar-Tiedman,
2023; Amram et al., 2023; Floyd, 2023). All three of these responses assert
that ChatGPT would not be a useful tool for streamlining the cataloguing
process, as the MARC records generated by the LLM are riddled with errors.
These letters to the editor go into technical detail about the shortcomings of
the study from the perspectives of professional cataloguers.
The literature review of the study is very
limited, with only four articles cited in this section. The articles cited
discuss ChatGPT broadly, and uses of ChatGPT more specifically in the context
of higher education and libraries. Given that ChatGPT is a relatively new
technology, and therefore not a great deal of literature exists on the topic
yet, it would have been interesting for the author to zoom out and situate the
current study in the historical context. For example, Weible (1990) and Burger
(1984) discuss automation of cataloguing. More recent articles such as Lowagie
(2023) are also missing from the literature review, however it is possible they
were published after this study was submitted for review.
This study was appraised using the CAT
critical appraisal tool by Perryman and Rathbun-Grubbs (2014). The research
objective is clear, and the methodology is appropriate, if limited (as
discussed below). The first item selected by the author to test ChatGPT as a
cataloguing tool is one that is relatively simple (i.e. readily available,
English language). Subsequent items used become increasingly complex (i.e.
different formats, non-English language items, non-Latin characters), with the
final item chosen being one with no existing WorldCat entry.
This study shows that ChatGPT can create
catalogue records albeit records with significant errors. The author maintains
the importance of reviewing and editing MARC records created by ChatGPT to
ensure that they are complete (e.g. replacing placeholder content) and correct
any biases or inaccuracies. In their letters to the editors, DeZelar-Tiedman
(2023), Amram et al., (2023), and Floyd, (2023) maintain that the time required
to correct the inaccuracies in the ChatGPT-created records would negate any perceived
efficiencies.
Various limitations in ChatGPT itself are
discussed. For example, the risk of bias based on the data used to train the
LLM, questions around copyright infringement and crediting intellectual labour
creating catalogue records, issues of privacy, and ownership of generated
content are identified as issues that need to be carefully considered before
ethically using ChatGPT for cataloguing. The author does not mention any
limitations in the study itself. Regarding the methods section, the sample size
is too small to be generalizable and the author does not provide the prompts
used to query ChatGPT. Regarding the conclusions, the fact that the ChatGPT
generated records differ from the WorldCat records, in some instances quite
significantly, is not addressed.
This study, despite its shortcomings, does
contribute to a conversation about the use of Artificial Intelligence and Large
Language Models in cataloguing. Doing a similar study with a larger sample
size, more transparent methodology, and more critical interpretation of results
would be interesting.
Amram, T.,
Malamud, R. G., & Hollingsworth, C. (2023). Response to "From ChatGPT
to CatGPT". Information Technology and Libraries, 42(4). https://doi.org/10.5860/ital.v42i4.16983
Brzustowicz, R.
(2023). From ChatGPT to CatGPT: The Implications of Artificial Intelligence on
Library Cataloging. Information Technology and Libraries, 42(3). https://doi.org/10.5860/ital.v42i3.16295
Burger, R. H.
(1984). Artificial Intelligence and Authority Control. Library Resources and
Technical Services, 28(4), 337-45.
DeZelar-Tiedman,
C. (2023). Response to "From ChatGPT to CatGPT". Information Technology and Libraries, 42(4). https://doi.org/10.5860/ital.v42i4.16991
Floyd, D.
(2023). Response to "From ChatGPT to CatGPT". Information Technology
and Libraries, 42(4). https://doi.org/10.5860/ital.v42i4.16995
Lowagie, H.
(2023). From Bias to Transparency: Ethical Imperatives in AI-Based Library
Cataloging. Retrieved from https://repository.ifla.org/handle/123456789/2841
Perryman, C.
& Rathbun-Grubb, S. (2014). The CAT: a generic critical appraisal tool. In
JotForm – Form Builder. Retrieved 21 Aug. 2014 from http://www.jotform.us/cp1757/TheCat
Weibel, S. L.
(1990). Automated cataloging: implications for libraries and patrons. 27th
Clinic on Library Applications of Data Processing, (pp. 67-80). Retrieved from http://hdl.handle.net/2142/1294