I think this article points out the fact that the creation of a technology in the lab, and its effective operational deployment are two different problems, both

I think this article points out the fact that the creation of a technology in the lab, and its effective operational deployment are two different problems, both difficult, requiring different skills and resources. An imperfect-but-instructive analogy would be between vaccine development and vaccine delivery. Once a vaccine has been developed and shown to be safe and effective, the hard work is just beginning. In the case of COVID, billions of doses must be produced, and then delivered to people, the delivery requires not just shipping of the doses, but matching the doses with an equal number (billions) of syringes, hypodermic needles, cotton swabs, band-aids, alcohol swabs, etc. People have to be recruited to deliver the doses, systems must be created to manage the demand and the queues, etc. The operational problem of delivering the vaccine to the world is arguably harder than its creation and testing. Likewise, the successful operational rollout of an AI-mediated automated or semi-automated decisioning problem is a complex problem requiring a totally different skillset than that of ML researchers in the lab. Computer systems and human procedures have to be created to manage the automation of the decision; decision results must be tracked and errors fed back to the lab to update models. Radiologists (including new radiologists) will of course be needed to understand the errors and provide correct labels, etc. Trust and mindshare in the medical community has to be built up. These things are not easy.