AI Regulation in Healthcare: can we achieve effective regulation?
Background: The implementation of AI systems for the healthcare industry requires special considerations. These systems, namely narrow AI, are already assisting us in diagnosis, in processing of vast array of healthcare data and in clinical administration.
Healthcare data deals with sensitive personal information through access to patient medical records and consequently privacy is one of the main concerns. The use of patient information from medical records for research purposes, despite de-identification, poses questions about consent and data collection, and must comply with the Australian Privacy Principles.
A further challenge occurs when deep learning algorithms are applied to medical data where it is often not clear how the system reaches a prediction for particular outcomes. We will need to grapple with issues of trustworthiness and explainability. Clinicians who use these systems should have a deep understanding of these issues and will be required to obtain informed consent from their patients when such systems are applied.
Regulation of AI in healthcare is complex and warrants collaboration between many stakeholders (designers, manufacturers, ethics and law experts, data scientists and clinicians). It is imperative that clinicians are involved early in the regulation process. The recognised trust between patients and their clinicians, places additional responsibility on clinicians to be well informed so that they can ensure safety, reliability, validity, privacy and consent around the implementations of new AI systems in clinical care.
This presentation will discuss the challenge of effective regulation and will touch on both the Australian and International experience.
Aims: The aim of this presentation is to encourage discussion and awareness around this important topic of AI Regulation in Healthcare
Methods: This presentation is in discussion format
Results: There are no specific results because this is not a research project
Conclusions: Same applies here as mentioned above