New trends and troubles for AI in medicine
Medicine is a complex field. So complex that any given person can’t know more than a fraction of what’s going on. Keeping up with the latest discoveries is impossible. Machine learning and other forms of artificial intelligence offer a new way of looking at medicine and a great power to automate medical tasks.
At the South by Southwest conference event in Austin, TX, a panel of experts came together to discuss the state of medical AI and how machine learning can benefit both patients and doctors. The discussion was moderated byKay Eron, general manager of health and life sciences at Intel.
The conversation opened with a look at how the panelists found themselves in the machine learning field.Naveen Rao, Ph.D., vice president and general manager of artificial intelligence solutions at Intel, answered that his interest came from a realization that machines weren’t all that different from biological beings. He was also concerned with how skills were so individual.
“It’s always been strange to me that knowledge is locked away inside a few individuals,” he said.
“My mission is to put powerful analytic tools in the hands of every decision maker,” saidBob Rogers, chief data scientist for analytics and AI solutions at Intel. He stated that we need tools to navigate this very complex world we live in.
When asked about current trends, neural networks came up instantly.John Mattison, MD, assistant medical director and chief health information officer, Southern California region, at Kaiser Permanente, explained that engineers are discovering that neural nets have increasingly evolved toward how living brains work. Because of this, he felt there was a real role for looking at biological examples for technical solutions.
Rao backed up this thought, offering that neural networks represent the world in almost the same way the world is built. All data in the world seems to be hierarchical, and people can break it down.
“One of the things that’s changed in machine learning, you could use data to make models, but they had limited utility. You had to do a lot of work up front. What’s exciting in this new generation, it can learn from example data without preprogramming,” said Rogers.
