Opening Doors, Expanding Possibilities

The course has traditionally challenged students to design and develop the knowledge, skills, and abilities necessary to work with individuals, performing exercise testing for health and fitness and developing exercise programs following guidelines from the American College of Sports Medicine and the National Strength and Conditioning Association. By focusing on the principles and guidelines for exercise testing and programming, students learn marketable skills that will carry them into a career or continuing education.

However, Kinesiology Assistant Clinical Professor Brooks Mobley knows the reality of AI, and he wants students to know what works and what doesn’t.

“In my Exercise Assessment, Prescription & Programming class, we use AI to evaluate health and fitness performance data, as well as various health metrics, to determine how accurately AI can assess this information,” Mobley explained.

For example, he will walk through an AI conversation in class where he shares information about a fictitious individual who might be seeking a health assessment. Data includes height, weight, age, career, self-reported exercise levels, body composition, and more.

“We use AI to generate exercise programs based on an individual’s health data, performance measures, and personal goals,” Mobley explained. “As a class, we then review the AI-generated output to determine whether the information is accurate, aligns with what we have learned, and is supported by scientific literature.”

Mobley even demonstrates how you can add information to AI platforms to produce better results, signifying that an AI-based exercise program is only as reliable as the data it receives and the accuracy of that data.

“Lastly, each student receives blinded health and performance data from our TigerFit database,” he said. “Students are tasked with creating an AI-generated exercise program using that data, then evaluating the accuracy of the AI’s output. They conclude by writing a report discussing what they liked, what they did not, and what they would change.”

For students in the class, it’s an eye-opening experience in seeing the value of accurate data and the need for a human-to-human interaction to produce the best possible exercise program.

Nick Mauldin, a graduate student in the School of Kinesiology, took Mobley’s course and completed the project involving AI-generated exercise programs.

“We were given guidelines to ‘steer’ the AI model in the right direction, but ultimately our job was to interpret and critique the AI output for accuracy,” Mauldin said. “I learned that AI often misses key details that someone may not be able to recognize, and uses many buzzwords in its responses, most likely to appeal to users.”

Mauldin noticed several downsides to the AI-developed program, notably that AI doesn’t have a good ability to understand the nuances of different exercise variables and thus has a difficult time getting away from its algorithmic response.

“Overall, the assignment taught me that AI is a great tool for speed and efficiency, but it needs to be utilized in conjunction with experience to get a high-quality result,” Mauldin said. “I also learned how to better spot mistakes. Specifically, AI usually makes mistakes with numbers and calculations and generally gets overall concepts correct. This is important for me to be aware of when utilizing AI in the future.”

Mauldin said the biggest benefit of AI is the speed with which you can get information back from it.

“As a learning tool or a teacher, it is extremely useful as you can have unlimited access to something to answer your questions and to correct your understanding of certain topics,” he said. “The speed at which you can get information is something that will be hard to replicate anywhere else.”

The biggest challenge, Mauldin said, is getting an AI platform to correct its mistakes.

“If you can catch an error that AI has made, whether it be through a conceptual idea or a calculation, it is extremely difficult to get the AI model to recognize and correct the mistake,” he said. “You can explain the mistake; however, AI will claim to recognize it and repeatedly make the same error. This can make doing mathematical tasks difficult.”

Mauldin also noted how important it is to have a preliminary understanding of the topic you are working with in AI, because without that, it can sometimes be difficult to factcheck.

“AI platforms may provide citations or links that do not work, and the rate at which you receive information would make it difficult to check it all anyway,” he explained. “If you are uninformed, you tend to be at the mercy of the information it gives you.”