The flap about IBM Watson’s Oncology Expert Advisor at MD Anderson is causing quite a stir. A Forbes article called the benching of Watson a “setback for artificial intelligence in medicine”. MD Anderson’s business process issues aside, the failure of Watson to meet its goals was neither a setback for AI nor a surprise. Here’s why.
Not a Setback
Watson’s Oncology Expert Advisor, an example of Prescriptive Analytics, reportedly provided 90% accuracy, which is, on the surface, an impressive accomplishment. Is 90% enough? Yes, as a clinical decision support tool. But, just like the GPS system that suggests a driver turn the wrong way on a one-way street, the expert must be cognizant of the patient’s circumstances and needs, and use the advice to reach their own conclusion. In this regard, AI is making a valid contribution and the efforts to date are indicative of great things to come for AI in health.
Not a Surprise
The conclusions from MD Anderson that the Oncology Expert Advisor project did not meet its goals and that significant updating is required are no surprise.
Cognitive systems such as Oncology Expert Advisor require training. And, like a physician’s own training, cognitive systems need their own version of Continuing Education. As knowledge increases, the accuracy based on information from one point in time will decrease against constantly evolving standards.
Quertle’s own BioAI™ platform which powers the Qinsight™ literature discovery solution, is supported by processes (both automated and human-in-the-middle) that ensure up-to-date accuracy and discoverability. Furthermore, Qinsight‘s goal is decision support through leading you, the expert, to the right literature quickly and accurately. This is accomplished through Descriptive and Predictive Analytics. As such, it is a robust platform with lasting current relevance.