Building AI systems in healthcare is hard because it involves interacting with and changing complex human organisations.
Sepsis Watch is a system used to predict and prevent cases of sepsis in patients at Duke Hospital in North Carolina.
AI is hard in healthcare. Making it work involves humans and machine and organisations. Anybody who has tried to implement ML in any organisation will recognise this:
" in order to be successful, AI interventions must always be thought of as sociotechnical systems, in which social context, relationships, and power dynamics are central, not an afterthought. AI and machine learning technologies are often looked to as being the key to a solution. However, all too often potential solutions remain just that—potential solutions, which may work in theory, given pre-set conditions."
Having tried and largely failed in the past to build a system that would do exactly this I couldn't agree more how important it is to think of the entire 'sociotechnical system'. Often, building the model is the easiest part. This article focuses on all of the hard parts after that.