Authors

Discussion

Abstract:

Introduction: With the advancement of Artificial Intelligence (AI), it may soon be possible to utilize such technologies to assist scientific research endeavours. One such software is ChatGPT, which has been programmed to acquire humanized intuition and high working efficiency that warrant further enquiries into its utility. Extracting data from publications is a time-consuming process in meta-analyses. Combining the humanized thinking of ChatGPT with data extraction could revolutionize the work process for those in this field. The aim of the present study was to assess the proficiency of ChatGPT to simplify data extraction to conduct a meta-analysis addressing the seroprevalence of SARS-CoV-2 on the African continent.

Methodology: Three versions of a "Task list" were created for ChatGPT to memorize, with the aim of establishing an assembly line approach.

Results: Extracting data based on providing only the paper title in the query was unsuccessful, and even after improving the "Task list" the data extracted by ChatGPT did not match the original text.

Conclusion: Based on the aforementioned results two reasons for the failure of extraction were identified: 1) the limited database of ChatGPT restricts its searching ability; 2) providing only the paper title does not provide enough information for searching. ChatGPT is then compared to the classical data extracting tool, and showing that this AI still needs to be improved.