There is a lot of variation in risk-mitigating AI development and deployment techniques right now. The evidence-based medicine movement arose from the documentation of unjustified diversity in clinical procedures. Evidence-based medicine is the deliberate, explicit, and thoughtful application of current best evidence in making practise decisions. According to experts, an Evidence Based AI development and deployment movement requires a similar philosophy and implementation approach.
One of these experts is Joachim Roski, a principal at Booz Allen Hamilton’s health practise. He’ll present case studies next month at HIMSS22 in his educational session “Making a Case for Evidence Based AI,” in which he’ll show how Evidence Based AI development and deployment movement practises could have prevented several high-profile AI failures. He’ll also go over several fundamental design ideas and features for developing Evidence Based AI.
However, if the often-hyped expectations for AI solutions are not matched by greater performance, there are warning signals of a “techlash” developing. AI solutions that consistently underestimate illness burden in non-Caucasian populations, poor performance in cancer diagnostic support, and limitations in delivering at scale are only a few examples of these issues.