AI in Research: Ethical Considerations

Recent advancements in Artificial Intelligence (AI) have given rise to new ethical challenges in healthcare and biomedical research. With a bottom-up approach, this presentation will focus on three challenges arising from local Singaporean case studies in non-ideal practical settings. 

The first case study involves a deep learning system to detect the progression of diabetic retinopathy found to have bias, and discusses whether it is ever justifiable to implement a biased AI when utility and equity are at odds. 

The second discusses issues surrounding human involvement with a deep learning system used for the detection of lumbar spinal stenosis (LSS), and tackles how to evaluate when humans can be “kicked out of the loop”. 

The final case study analyses the risks involved in the implementation strategy proposed for a machine learning triage tool estimating mortality after emergency admissions where a randomised control trial is not feasible.