Dr. Gary Wilkerson doesn’t see artificial intelligence-enabled robots replacing humans in the field of athletic training in the future, but he does see AI already impacting how research and practice management are done in the present and the immediate future.
And the University of Tennessee at Chattanooga professor is working to prepare his graduate athletic training students for that reality.
“In our field, we are specialists in preventing injuries, rehabilitation and then, certainly, preventing secondary injury after the rehabilitation process is complete; and we send people back to sport participation,” said Wilkerson, also an award-winning researcher of more than 30 years.
“That is a very, very complex problem in terms of the many, many factors that are interacting with one another to either increase or decrease risk.
“Historically, for over a century, the dominant paradigm has been null hypothesis. significance testing, with statistical significance generally defined as a P value of less than 0.05, and that’s been paired with a reductionistic approach to research design that emphasizes randomly assigning people to either an experimental group or a control group. The inherent limitation of that approach is you can study only one or two things at a time, and you can’t really get at the interactions among factors.”
So, as is the case in many disparate fields, AI can benefit athletic training in the rapid and complex analysis of a large database of many multiple variables toward a machine-learning approach to diagnosis and treatment of injury, Wilkerson said.
“The thing that we’re clearly moving toward in our field is decision trees where you classify somebody based on a strongest factor, and then within the subgroups look at what are the next strongest factors,” he said. “Those outputs could either be coming from a formal machine learning method or they could be more guided in their construction using procedures that historically have been viewed as statistics.”
About UTC Department of Health and Human Performance
The specific machine learning method to use for development of such models is currently a challenge for researchers, Wilkerson said.
“One of the big distinctions we’re grappling with right now is that many of the machine learning methods are not transparent in how the output is produced, so they’re being referred to as ‘black box models’,” he said. “It gives you an algorithm, but there’s no way to trace how that algorithm was created from the data. We need a way to understand the rationale for why a particular factor was included or excluded from the model.
“Another big limitation right now is that there are very, very few people in our field who possess the mathematical expertise and the computer engineering expertise to actually implement these machine models. You have people in the computer engineering realm who are very good at running these, but they typically understand little about the input data. For machine learning to realize its full potential, there has to be collaboration between the machine learning expert and the domain expert.”
In his research methods course, Wilkerson’s athletic training graduate students are introduced to the reality of AI as a research and clinical tool.
“We are very clearly seeing a transition away from traditional hypothesis testing to machine learning algorithms that can handle extremely large numbers of variables in cases that just exceed human capacity to consider so many different factors,” he said. “I’m trying to help them to understand that there are some basic concepts they must grasp to be able to evaluate whether machine learning output is going to be useful for their clinical decision making.”
Another graduate course is based on knowledge—including of relevant AI—and skills used on the business side of athletic training.
“The healthcare finance and administration course is for students who wind up working in orthopedic clinics or hospitals or corporate industrial health management programs to understand how continuous quality improvement is a critically important skill to have, to be able to demonstrate that you’re identifying inefficiencies within a system and that you can demonstrate the monetary value of the improvements that you’ve made,” Wilkerson said. “That involves extremely large amounts of clinical data for patient status, their diagnosis, but also then their health insurance claims and their workers’ compensation claims and their time loss away from work. Analyzing factors like that and identifying strategies for quality improvement is an important administrative function.”
One former student took that expertise to a position he now has with the UT College of Medicine Orthopaedic Residency program. “He’s essentially helping to supervise resident research projects and helping them with their data analysis, most of which are using machine learning algorithms in the output,” Wilkerson said.
Among athletic training university faculty, generally, there is growing recognition of the growing role of AI in the classroom—which is also a subject of research Wilkerson is conducting now.
“I’m collaborating with some colleagues at three other institutions where we’ve sent out a survey to faculty members to gauge their perspectives on how important this is, how prepared they are for it, and we’re in the process of developing that manuscript right now,” he said. “I would say probably about 20 to 25% recognize it and are actually taking steps in that direction, but it’s inevitable that it’s going to be a hundred percent.”
Undergraduate studies in UTC Health and Human Performance
Graduate studies in UTC Health and Human Performance
About CHAI–the UTC Artificial Intelligence Initiative
Are you a UTC faculty member or scholar researching an AI field? Email us so that we can share your story: UTC-AI@utc.edu