[Paper Review] Large Language Models in Sport Science & Medicine: Opportunities, Risks and Considerations

Mark Connor, Michael F. O‘Neill|arXiv (Cornell University)|May 5, 2023

Artificial Intelligence in Healthcare and EducationMedicine3 citations

TL;DR

This paper examines the integration of large language models (LLMs) in sports science and medicine, highlighting their potential to enhance clinical decision-making, personalize training programs, and improve knowledge dissemination—particularly in underserved regions. It identifies key risks such as data bias, privacy breaches, and hallucinated outputs, emphasizing the need for ethical alignment and human-in-the-loop validation to ensure safe, equitable applications.

ABSTRACT

This paper explores the potential opportunities, risks, and challenges associated with the use of large language models (LLMs) in sports science and medicine. LLMs are large neural networks with transformer style architectures trained on vast amounts of textual data, and typically refined with human feedback. LLMs can perform a large range of natural language processing tasks. In sports science and medicine, LLMs have the potential to support and augment the knowledge of sports medicine practitioners, make recommendations for personalised training programs, and potentially distribute high-quality information to practitioners in developing countries. However, there are also potential risks associated with the use and development of LLMs, including biases in the dataset used to create the model, the risk of exposing confidential data, the risk of generating harmful output, and the need to align these models with human preferences through feedback. Further research is needed to fully understand the potential applications of LLMs in sports science and medicine and to ensure that their use is ethical and beneficial to athletes, clients, patients, practitioners, and the general public.

Motivation & Objective

• To evaluate the potential applications of large language models (LLMs) in sports science and medicine.
• To identify key risks associated with LLM deployment, including data bias, privacy exposure, and hallucinated outputs.
• To examine the challenges of aligning LLMs with clinical standards and human preferences.
• To advocate for ethical, evidence-based integration of LLMs in sports medicine practice.
• To support equitable access to high-quality sports health information through LLMs in low-resource settings.

Proposed method

• Systematic review of current LLM capabilities in natural language processing and their transferability to sports medicine contexts.
• Analysis of LLM training mechanisms, including pre-training on large text corpora and fine-tuning with human feedback.
• Evaluation of LLM performance in generating clinical recommendations, summarizing medical literature, and personalizing training plans.
• Assessment of risks related to data privacy, model hallucinations, and dataset biases in health-related LLMs.
• Use of qualitative and conceptual frameworks to assess ethical and clinical alignment of LLMs with practitioner needs.
• Identification of gaps in research and implementation, particularly in validation and real-world clinical integration.

Experimental results

Research questions

• RQ1 How can large language models enhance decision-making and personalization in sports medicine and science?
• RQ2 What are the primary risks associated with deploying LLMs in clinical and athletic performance settings?
• RQ3 How can biases in training data affect the reliability and fairness of LLM-generated recommendations in sports health?
• RQ4 What safeguards are necessary to prevent exposure of confidential patient or athlete data through LLM use?
• RQ5 How can LLMs be ethically aligned with clinical guidelines and human preferences in sports medicine?

Key findings

• LLMs show strong potential to support sports medicine practitioners by generating personalized training and recovery recommendations based on textual data.
• LLMs can improve access to high-quality sports health information in developing countries where specialist expertise is limited.
• There is a significant risk of model hallucinations, where LLMs generate plausible but factually incorrect medical or training advice.
• Dataset biases in training data may lead to inequitable or inaccurate recommendations, particularly for underrepresented populations.
• The use of human feedback fine-tuning is essential to align LLM outputs with clinical standards and reduce harmful or misleading content.
• Robust validation and oversight mechanisms are required before LLMs can be reliably used in clinical decision support systems.