The Evolution of Machine Learning

January, 2018 – In New Yorker article A.I. Versus M.D, author and oncologist Siddhartha Mukherjee unpacks the evolution of machine learning as it relates to clinical diagnosis, a topic we are following closely at Openxmed. But first, let’s look at the question: how do physicians diagnose?

Diagnosis: the Science

Doctors use patient history and a physical exam to collect facts. That information is used to generate a list of possible causes. Questions and tests help eliminate or strengthen hypotheses. As the list narrows, doctors refine their assessment. Finally, definitive lab or imaging tests are used to confirm diagnosis.

Diagnosis: the Art

There’s also a component of diagnosis that can only be described as an art and it comes with time and experience. Over time, a doctor is able to diagnose in the same way he or she recognizes a commonplace object – without the need to assess in detail, differentiate, confirm – they just know.

This distinction between science and art, is what British philosopher Gilbert Ryle called the difference between “knowing that” and “knowing how”. The former speaks to factual knowledge, the latter to experiential. It’s the difference between knowing that a bike has 2 wheels, pedals and handlebars to propel you forward and steer, versus knowing how to get that bike from point A to B, by balancing, avoiding potholes, leaning forward up a hill.

In his article, Mukherjee examines early efforts to automate diagnosis focused on patterns, the facts – or rule-based algorithms like we see in electrocardiograms or pattern-recognition software used in mammography. Where these first-generation diagnostic tools fell short, however, was in their ability to improve over time, to learn, to move from facts (knowing that), to experience (knowing how).


Machine Learning and Openxmed

At Zenxmed, we are fortunate to have begun partnerships with Dr. Denilson Barbosa (University of Alberta) and Dr. Fred Popowich (University of British Columbia) who are using Natural Language Processing (NLP) to identify and extra high-quality medical evidence from literature.

We believe strongly in continuous learning and refining our ability to diagnose and treat patients, and Artificial Intelligence will no doubt augment our ability to do so. The ability of machines to process datasets at incredible speeds will leave physicians free to take this information and apply it to new discoveries. Additionally, it will allow us more time to connect with our patients, and apply these discoveries to their specific context.

We look forward to the change that is coming, and continuing to build a global community of physicians that can empower one another with these changes.