Key points are not available for this paper at this time.
Recent achievements in respect of Artificial Intelligence (AI) open up opportunities for new tools to assist medical diagnosis and care delivery. However, the typical process for the development of AI is through repeated cycles of learning and implementation, something that poses challenges to our existing system of regulating medical devices. Product developers face tensions between the benefits of continuous improvement/deployment of algorithms and keeping products unchanged. The latter more easily facilitates collecting evidence for safety assurance processes but sacrifices optimisation of performance and adaptation to user needs gained through learning-implementation cycles. The challenge is how to balance potential benefits with the need to assure their safety. Governance and assurance processes are needed that can accommodate real-time or near-real-time machine learning. Such an approach is of great importance in healthcare and other fields of application. AI has stimulated an intense process of learning as this new technology embeds in application contexts. The process is not only about the application of AI in the real world but also about the institutional arrangements for its safe and dependable deployment, including regulatory experimentation involving new market pathways, monitoring and surveillance, and sandbox schemes. We review the key themes, challenges and potential solutions raised at two stakeholder workshops and highlight recent attempts to adapt the laws for AI-enabled medical devices (AIeMD) with a special focus on the regulatory proposals in the UK and internationally. The UK regulatory trajectory shows signs of alignment with the US thinking, and yet the European Union model is still the most closely aligned framework.
Li et al. (Mon,) studied this question.