Safety and Precision AI for a Modern Digital Health System.
Journal:
Yearbook of medical informatics
PMID:
40199284
Abstract
Artificial intelligence (AI) promises to revolutionize healthcare. Currently there is a proliferation of new AI applications that are being developed and beginning to be deployed across many areas in healthcare to streamline and make healthcare processes more efficient. In addition, AI has the potential to support personalized and customized precision healthcare by providing intelligent interaction with end users. However, to achieve the goal of precision AI issues and concerns related to the safety of AI, as with any new technology, must be addressed. In this article we first describe the link between AI and safety and then describe the relation of AI to the emerging study of technology-induced error. An overview of published safety issues that have been associated with introduction of AI are described and categorized. These include potential for error to arise from varied sources, including the data used to drive AI applications, and the design process of AI applications itself. In addition, lack of appropriate and rigorous testing and limited analysis of AI applications during procurement processes has also been reported. Recommendations for ensuring the safe adoption of AI technology in healthcare are discussed, focusing on the need for more rigorous testing and evaluation of AI applications, ranging from laboratory testing through to naturalistic evaluation. The application of such approaches will support safety and precision AI for a modern digital health system.