AI, augmented reality are tools, not the solution in medicine
NEW YORK — In a broad sense, AI is the ability to create programs that mimic elements of human intelligence and has been around for decades but made possible by three recent advances: massive computing power, improvements in machine learning techniques and big data, Justin Ko, MD, clinical associate professor at Stanford School of Medicine said in a session here.
He added that technological advances cannot solve all problems in the real world and joked with the audience, asking how many attendees came to his talk at the American Academy of Dermatology Summer Academy Meeting to hear how computers would replace the need for clinicians.
“To create an algorithm, you need a lot of data and a label of interest, essentially the answer key,” Ko said. “You can train a model to figure out the best way to match that data with the answers with a set of rules. You can then test the model by applying it to complete the task you created, and then you see how it learned. You can do this for any task you can imagine.”
It is powerful, as it can uncover complex relationships in data in a way that traditional statistics and the human brain cannot.
“When we start to gather new data streams — like from genetics, the microbiome, social media feeds, EHR data — we can start to interrogate the linkage between health and disease and essentially engage precision and personalized medicine,” Ko added.
AI performs reliably in the first task or the millionth. It can be trained to form a specific task and is geared in ways the human brain is not.
“It possesses no threat to us or our jobs. It’s a tool, not a solution and is dependent on us to make it work,” he said.
Moreover, augmented reality can help improve workflow and offer insight in clinical decisions; Ko suggests thinking of augmented reality like a virtual digital assistant.
Instead of spending most of the patient visit gathering information, you could spend it connecting with the patients to make sure they feel seen and heard, he said.
“Could deploying a virtual clinical system bring more humanity to your practice?” he asked.
The quality of data used is of the utmost importance, and is also a concern with this technology. “AI will magnify biases within the data set, so those need to be known.
“What we need to do is make sure high quality data are representative and inclusive of the populations we want to study,” Ko said.
Patients without access to care are unlikely to be part of any data sets but have the most potential to benefit from this technology and access, he added.
“We as physicians have to be thoughtful about the implementation that AI bridges, not worsens, these gaps. …We have to be consistent and relentless in advocating for those appropriate guardrails to be in place to protect our patients from harm and of unintended consequences in technology and make sure the algorithms are subject to rigorous external and internal validation.” – by Abigail Sutton