Recognizing and Avoiding Epistemic Injustice When Using AI Tools for Reviews of Literature

Abstract

Global academic publishing is transforming, as many research outputs are shared through open access models to increase access globally. With this move, the magnitude of research available to an average researcher or layperson has increased exponentially. Therefore, making sense of the mass amount of literature has become an overwhelming aspect of the preliminary step in research known as the review of literature. In conjunction with the increase of available peer-reviewed literature, the emergence of AI tools has enabled many investigators to obtain and summarize available documents with limited intimate knowledge of each study. Leaning on the use of AI tools can come with risks that need to be explored. Many scholars use AI tools with the knowledge and understanding that AI can have built in biases in the algorithms, affecting how and what it extracts from the research, as well as biases in the human designed lenses in which it summarizes the data. However, many investigators are using these tools without this understanding, or the knowledge that generative AI can hallucinate, or create what appear to be facts from nonexistent literature. The objective of this paper is to examine what is known about biases in AI and how these can lead to epistemic injustices in the early stages of research. The critical nature of understanding AI as part of information literacy will be emphasized. This study would contribute to a growing body of knowledge about the possible effects of AI when used in extracting and summarizing extant literature.