JMIR Publications

ABSTRACT

Background:

As the use of social media for the dissemination of healthcare information becomes more prevalent, the expanding role of Artificial Intelligence (AI) and Machine Learning (ML) in intervening in this process becomes inevitable, which raises multiple ethical concerns. This paper delves into the ethical use of AI and ML in the context of healthcare information on Social Media Platforms (SMPs). It examines these technologies from the perspective of Fairness, Accountability, Transparency, and Ethics (FATE), with emphasis on computational and methodological approaches that ensure their responsible application.

Objective:

We aim to find and compare existing solutions that address components of FATE in the domain of AI in healthcare on SMPs by in-depth exploration of computational methods, approaches, and evaluation metrics used in various solutions. Additionally, we present an assessment of the evidence strength supporting each solution.

Methods:

Our research methodology, based on the approach by Kofod-Petersen [1] and conforming to the PRISMA Extension for Scoping Reviews [2], involved a comprehensive search across PubMed, Web of Science, and Google Scholar. We employed a strategic search using specific filters to identify relevant research papers dated from the year 2012 onward, focusing on the intersection and union of different literature sets. The inclusion criteria were focused on studies primarily concerned with FATE in healthcare discussions on SMPs, those presenting empirical results, and covering definitions, computational methods, approaches, and evaluation metrics.

Results:

We decomposed the FATE principles, offering clear and concise definitions aligned with the AMIA ethical principles where applicable. This decomposition was further segmented into dedicated subsections, where we provided specific computational methods and conceptual approaches designed to enforce each aspect of FATE in the context of AI-driven healthcare on SMPs. Our review highlights the intricate interactions between these principles and identifies limitations in their practical implementations.

Conclusions:

While a diverse array of approaches and metrics exists to address FATE in AI for healthcare on SMPs, each comes with inherent limitations. The application of these methods often intersects with other ethical principles, potentially leading to conflicts. Currently, there is no unified, well-established solution for ensuring the comprehensive integration of FATE principles in this field. Therefore, careful consideration is essential when employing existing methods, balancing their benefits against potential ethical trade-offs. This highlights the ongoing need for research and development of more comprehensive solutions to uphold FATE principles in the evolving landscape of AI in healthcare on SMPs.