Edward Christopher Dee, Ryan Carl Yu, Leo Anthony Celi & Umbereen Sultana Nehal
Abstract
Artificial intelligence (AI) and machine learning in healthcare are growing at an unprecedented rate. Myriad uses of medical AI, ranging from tumor identification on imaging to workforce management, make use of a wealth of available healthcare data. These models are becoming increasingly commercially available. However, much of the utility of medical AI depends on the quality of the data models trained on, and critically, the contexts and biases within which these models are created. In this chapter, we first describe a business-informed framework that influences product development and commercialization of these technologies. We describe the consumer side that includes purchasers, end users, and patients. Subsequently, we underscore the pitfalls of the assumption that models trained in one context can be applied to another, that is, the myth of generalizability. We propose solutions to these problems and describe the importance of co-creation and multi-stakeholder engagement in designing medical AI. We highlight the need to define value metrics that consider equity and the mitigation of healthcare disparities. Lastly, we draw attention to open ethical, legal, and policy questions that must be answered as the role of AI in medicine progresses and grows. https://link.springer.com/referenceworkentry/10.1007/978-3-030-58080-3_247-1
