Accountability in the Machine Learning Pipeline: The Critical Role of Research Ethics Oversight

Specific challenges posed by artificial intelligence in research ethics

Background

The twenty first century is often defined as the era of Artificial Intelligence (AI), which raises many questions regarding its impact on society. It is already significantly changing many practices in different fields. Research ethics (RE) is no exception. Many challenges, including responsibility, privacy, and transparency, are encountered. Research ethics boards (REB) have been established to ensure that ethical practices are adequately followed during research projects. This scoping review aims to bring out the challenges of AI in research ethics and to investigate if REBs are equipped to evaluate them.

Methods

Three electronic databases were selected to collect peer-reviewed articles that fit the inclusion criteria (English or French, published between 2016 and 2021, containing AI, RE, and REB). Two instigators independently reviewed each piece by screening with Covidence and then coding with NVivo.

Results

From having a total of 657 articles to review, we were left with a final sample of 28 relevant papers for our scoping review. The selected literature described AI in research ethics (i.e., views on current guidelines, key ethical concept and approaches, key issues of the current state of AI-specific RE guidelines) and REBs regarding AI (i.e., their roles, scope and approaches, key practices and processes, limitations and challenges, stakeholder perceptions). However, the literature often described REBs ethical assessment practices of projects in AI research as lacking knowledge and tools.

Conclusion

Ethical reflections are taking a step forward while normative guidelines adaptation to AI's reality is still dawdling. This impacts REBs and most stakeholders involved with AI. Indeed, REBs are not equipped enough to adequately evaluate AI research ethics and require standard guidelines to help them do so.

The silent trial - the bridge between bench-to-bedside clinical AI applications

As more artificial intelligence (AI) applications are integrated into healthcare, there is an urgent need for standardization and quality-control measures to ensure a safe and successful transition of these novel tools into clinical practice. We describe the role of the silent trial, which evaluates an AI model on prospective patients in real-time, while the end-users (i.e., clinicians) are blinded to predictions such that they do not influence clinical decision-making. We present our experience in evaluating a previously developed AI model to predict obstructive hydronephrosis in infants using the silent trial. Although the initial model performed poorly on the silent trial dataset (AUC 0.90 to 0.50), the model was refined by exploring issues related to dataset drift, bias, feasibility, and stakeholder attitudes. Specifically, we found a shift in distribution of age, laterality of obstructed kidneys, and change in imaging format. After correction of these issues, model performance improved and remained robust across two independent silent trial datasets (AUC 0.85–0.91). Furthermore, a gap in patient knowledge on how the AI model would be used to augment their care was identified. These concerns helped inform the patient-centered design for the user-interface of the final AI model. Overall, the silent trial serves as an essential bridge between initial model development and clinical trials assessment to evaluate the safety, reliability, and feasibility of the AI model in a minimal risk environment. Future clinical AI applications should make efforts to incorporate this important step prior to embarking on a full-scale clinical trial.