Table 2.

Challenges for AI application in diabetes care and how they may be overcome with future development

Challenge	Description	Mitigating strategies
Data quality control	data quality may have the following problems: (1) poor quality of the data themselves, (2) poor quality of the data labels, and (3) insufficient data.	ensure the quality of data used in the training process
	AI may amplify implicit bias and discrimination if trained on data reflecting the health-care disparities	train AI algorithms on fair datasets that include and accurately represent social, environmental, and economic factors that influence health
Poor technology design	the initial versions of most AI systems are always challenging to navigate	understand the needs of the end user (for example, patients and providers)
	many EHR vendors did not follow basic usability principles	develop software and applications with input from end users
	patients reported lack of confidence with technology, as well as frustration with design features and navigation of commercially available mobile applications	utilize iterative design process
Lack of clinical integration	application of AI systems in the real world may lead to many unintended outcomes	develop AI algorithms that could be integrated into current clinical and digital workflows
	experts may struggle to develop trust with AI systems	demonstrate explainability analysis of AI systems
	AI systems could also be perceived as encroaching on clinicians’ professional role	support the clinical decision-making of clinicians instead of making solely a competing diagnosis
Privacy concerns	implementing data privacy and security assurances is an overriding issue for the future of AI in medicine, since there are pervasive problems of hacking worldwide	•develop AI algorithms using federated learning or swarm learning •protect closed-loop automated insulin-delivery systems from hacking •ensure an individual’s identity could not be determined by facial recognition or genomic sequences from massive databases
Non-adherence	user adherence is crucial to the effectiveness of AI applications in the real world, which can be affected by convenience, user experience, and true benefits brought by this technology	•use smart design, visible electronic health records •integrate electronic patient-reported outcomes in clinics •explore voice enablement in AI software and applications
Imperfection of laws and regulations	AI in medicine results in legal and regulatory challenges regarding medical negligence attributed to complex decision-support systems	•provide clear guidance on which entity holds liability when malpractice cases involving medical AI applications arise •update the credentials needed for diagnostic, therapeutic, supportive, and paramedical tasks with the deployment of automated AI for specific clinical tasks

AI, artificial intelligence; EHR, electronic health record.