AI Doctor

CHAPTER 3Barriers to AI Adoption in Healthcare

IN THE LAST CHAPTER, we discussed the many issues there are when it comes to building medical algorithm. Less than perfect data in the training, validation, or clinical deployment of models can result in the issues we touched on, such as model bias, gaps in performance, interpretability issues, lack of explainability, and more. We need to remember that these aren't the only issues that keep AI models from being developed or used in healthcare. A myriad of other technical, economic, regulatory, and business barriers exist (Figures 3.1 and 3.2). Many of these have yet to be addressed sufficiently so that the applications of AI in medicine can truly take off. In this chapter, we'll discuss many of those barriers and speculate on how they could be overcome.

According to a survey of over 12,000 participants by consultancy PriceWaterhouse Coopers (PwC), a lack of trust and the need for the human element were the biggest hurdles to using AI in healthcare.² Another survey by Klynveld Peat Marwick Goerdeler (KPMG) in 2020 revealed a number of areas of concern for healthcare executives in regards to AI.³ One of these areas is that of talent. At the time of writing, only 47% of healthcare employees say that their employers offer AI training courses, a figure which is much lower than we see in other industries. This may be why only 67% of healthcare workers support AI adoption, which makes healthcare the lowest ranking industry. We can't build ...