Developing an evidence-based framework for reducing epistemic trespassing when using generative artificial intelligence: a mixed methods study

Generative artificial intelligence (GenAI) is a machine learning model that mimics human language and, by using probability, generates content across a range of formats. GenAI is disrupting numerous industries, including academia. It can be used across the entire research pipeline, from ideation and literature review to data analysis and manuscript preparation; it is easily accessible, deployable, and offers productivity gains to researchers. However, this raises questions about its implications for research integrity, especially if GenAI is used to substitute specialist technical expertise such as statistical methodology or data interpretation. This tension between achieving productivity gains and maintaining research integrity is particularly significant in healthcare research where methodological and design errors and misinterpretation of data can severely impact health outcomes and quality of life. Security, rigor, accuracy and robustness are all vital components to health research as findings may directly impact patient outcomes, inform clinical guidelines, and shape future research funding. Methodological errors, no matter the cause, in this domain can damage the credibility of research in the eyes of the public if these errors lead to the publication of misleading information and changing recommendations. Healthcare researchers minimise these errors by working in multidisciplinary teams, with individuals of various specialised backgrounds. Insights from clinical, statistical, behavioural science, and economics backgrounds are combined to develop hypotheses and analyses that better detect, quantify and interpret the complex relationships between variables ensuring high quality methodology, more impactful conclusions, and limiting downstream consequences. GenAI can improve access to specialised knowledge, allowing researchers to engage in complex methodologies outside their speciality (e.g., generating data analysis code for non-statistical researchers). This offers enormous benefits, allowing researchers to pursue complex and robust methodologies, however the potential costs of this approach are unclear. Information generated by GenAI reflects the most probable response given the data it was trained on, not specialised knowledge of a topic. The initial answers generated by GenAI can be incorrect or misinterpreted by individuals with limited expertise, resulting in a form of epistemic, or knowledge, trespassing where researchers make judgements outside their core domain. Epistemic trespassing may inadvertently introduce subtle but consequential errors. This research programme proposes to explore how GenAI is used outside one’s specialty within healthcare research. Findings aim to inform recommendations and an initial evidence-based framework on effective GenAI use. Initially, a survey will be administered to healthcare researchers across research experience levels, aiming to provide insight into GenAI’s role in the research pipeline and their views and experience of the benefits and risks associated with its use. Public perception of GenAI use in healthcare research and perceived impact on healthcare will be investigated. An experimental study will be conducted to assess how healthcare PhD students utilise GenAI to conduct data analysis. This experiment will implement statistical anomalies that, if not caught, impact results and interpretation, assessing whether students are managing these risks properly. This proposed programme of research will provide evidence on how healthcare researchers (including student learners) are implementing GenAI in their workflow, particularly in fields outside their core domain: a poorly understood area of research with impacts to rigor, integrity, interdisciplinary research, and ethical implications. These findings aim to inform an initial framework for utilising GenAI outside one’s core domain, particularly data analytics, effectively informing policy developments on the education and training of future and current healthcare researchers.