A new (augmented) reality: Overcoming challenges to using AI in athlete development  

“Not everything that can be counted counts, and not everything that counts can be counted” – Cameron (1963) 

The current generation of analysts in sports like baseball (arguably the most data-obsessed sport on the planet) have access to data their statistically curious relatives from a generation ago could scarcely imagine. But has this interest and influx of data increased our ability to predict and influence the outcomes that matter in sport?  

In the most recent issue of Psychology of Sport and Exercise, our research group (with our colleagues Adam Kelly and Alex McAuley in the UK) contributed a critical review of studies applying Artificial Intelligence (AI) techniques (for example, machine learning and/or deep learning) in athlete development contexts like talent/athlete identification, selection, and development. We won’t be including the full review in this article as this work is still developing, and more research is needed to clarify the benefits and limitations of these technologies. (If you’d like to explore our research in more depth, please see Baker et al., 2025). Instead, we’ll be providing guidance in how (and how not) to use them as well as highlighting the challenges associated with these (and other) emerging technologies in high-performance sport.  

Challenge 1: The Underlying Assumptions – What is the basis of AI? 

AI approaches to analysis and forecasting rely on large datasets of historical information like team or individual performance to predict future events. While this approach can make sense for short-term predictions, using it in athlete development contexts can be problematic. Why? Because it assumes two things:  

• We are already collecting the most relevant information for athlete development. Unfortunately, current research from our team and others shows significant gaps in our knowledge in this area.  

• That things that were relevant in previous generations are equally applicable in current and future ones. The extensive time horizons involved in athlete development make predictions substantially more complex and difficult. Data from past generations may no longer match the cohorts they are trying to predict. For example, the social environments the athletes developed in may be appreciably different.  

Challenge 2: Establishing validity and reliability – What is the value of these technologies? 

One of the greatest perceived benefits of emerging technologies including AI is the rapid rate at which these advance. On the surface, this adaptability seems like a strength, for example, new tools can respond quickly to new developments. But this rapid pace creates a major problem: it makes it difficult to determine their validity and reliability. Researchers need to compare performance results over time with multiple cohorts. This is impossible to do when technologies do not remain stable over time. 

Challenge 3: Data Management – Who should access this data? 

The explosion of data available to scientists and practitioners through revolutionary technologies like computer vision has led to increased concern about data ownership (storage, accessibility, usage, and sale of athlete data). Unfortunately, athletes are often the last to be considered in these discussions, despite being the most at risk of being negatively affected or exploited via the data they generate. While some sports organizations like the NBA and FIFA are leading efforts to give athletes more control over their data, the reality is complex. Often, new technologies are created not just to provide data to end-users, but to build large databases that can be monetized for other purposes like gambling, fan engagement, and broadcasting. Removing a software company’s ability to use the data generated through their devices may undermine the quality of the technology by, for example, limiting updates to software.  

Challenge 4: Increasingly specialized skills – Who can understand the technology? 

As technology continues to re-shape the world of high-performance sport, it is also re-defining the skills needed to operate in this environment. Athlete development policies often promote ‘holistic’ approaches that emphasize this process as multi-factorial and multi-disciplinary, yet the knowledge and methods being used are becoming increasingly specialized. This creates a disconnect between those who understand and use these approaches (like analytical teams) and those who do not (often coaches, trainers, etc.). Ultimately, AI-driven performance analytics will be as valuable as the extent to which coaches and other practitioners can interpret and apply the insights. Prioritizing clear and timely communication between roles or levels in a research team or organization will ensure data does more than sit in a database or report.  

Overcoming these challenges: The potential of AI 

 “AI systems should be designed with the intention of augmenting, not replacing, human contributions”  

Jarrahi (2018) 

Although many of these challenges apply to all new technologies, they are particularly relevant for those AI. A practical way forward is to recognize where AI adds the most value by addressing areas where humans are less efficient. Computers have the capacity to generate, synthesize and/or recall knowledge faster than any human, while humans outperform computers in more subjective tasks. However, prioritizing the cues humans and computers pick up on respectively, and how that information gets used may provide the best of both worlds.  

Research from human judgement and decision-making in medicine and law indicates that computer-generated advice improves the quality of human decision-making (Grgic-Hlaca et al., 2019; Wang et al., 2016). Those working in athlete selection contexts could consider how to integrate this type of blended approach. For instance, AI may be useful for exploring patterns in data in less-biased and more-reliable ways, while humans may be more effective at assessing more subjective elements of athlete development such as relationships, emotional states, character, drive, and resilience. Together, they provide a stronger approach than either provides separately.  

A 21st century approach to athlete development will need to take advantage of emerging technologies while overcoming traditional limitations of previous work. Historically, sport science has been largely siloed into disciplines like physiology, psychology, and biomechanics, which has limited how research questions have been explored. The mountains of new data emerging on a near daily basis provide researchers and other interest-holders the capacity to build better, more holistic models of athlete development and performance. Greater depth and breadth in the types of data available for those working in sport science will also allow development and testing of more complex models of relationships among variables.  

The availability and accessibility of large datasets also increase the potential value of AI to level the playing field beyond countries who have traditionally dominated in the realm of sport science.  Currently, AI in sport depends heavily on extensive datasets, sophisticated infrastructure, and specialized expertise, which means its benefits are concentrated in countries with robust sport science systems. However, if low-cost, scalable AI solutions can be paired with globally accessible datasets, these technologies have the potential to democratize performance analytics. Strategic implementation could support nations with fewer resources in narrowing the competitive gap. 

AI approaches continue to advance at an impressive rate and integrating them into aspects of athlete training, selection, and development will become commonplace. However, thoughtful integration will ensure they are used to improve on current concerns instead of contributing additional noise to an already noisy system.  

Further Reading from SIRC 

Sport after Moneyball: Exploring sports analytics and the digital economy – https://sirc.ca/blog/sports-analytics/  

Beyond the buzzwords: A practical guide to AI for sport leaders – https://sirc.ca/articles/beyond-the-buzzwords-a-practical-guide-to-ai-for-sport-leaders/  

References 

Baker, J., Cattle, A., McAuley, A., Kelly, A., & Johnston, K. (2025). Will artificial intelligence solve the riddle of athlete development? A critical review of how AI is being used for athlete identification, selection, and development. Psychology of Sport and Exercise, 102978. https://doi.org/10.1016/j.psychsport.2025.102978  

Cameron, W. B. (1963). Informal sociology: A casual introduction to sociological thinking. Random House. 

Grgic-Hlaca, N., Engel, C., & Gummadi, K. P. (2019). Human decision making with machine assistance: An experiment on bailing and jailing. Proceedings of the ACM on Human-Computer Interaction, 3(CSCW), 1–25. https://doi.org/10.1145/3359280  

Jarrahi, M. H. (2018). Artificial intelligence and the future of work: Human-AI symbiosis in organizational decision making. Business Horizons, 61(4), 577–586. https://doi.org/10.1016/j.bushor.2018.03.007  

Wang, D., Khosla, A., Gargeya, R., Irshad, H., & Beck, A. H. (2016). Deep learning for identifying metastatic breast cancer. arXiv. https://doi.org/10.48550/arXiv.1606.05718