Psychology of Sport & Exercise 82 (2026) 102978

Contents lists available at ScienceDirect

Psychology of Sport & Exercise

journal homepage: [Link]/locate/psychsport

Will artificial intelligence solve the riddle of athlete development? A critical

review of how AI is being used for athlete identification, selection,
and development
Joseph Baker a,* , Antonia Cattle a , Alex McAuley b , Adam Kelly b , Kathryn Johnston a
a
Tanenbaum Institute for Science in Sport, University of Toronto, Toronto, Ontario, Canada
b
Research for Athlete and Youth Sport Development (RAYSD) Lab, Birmingham City University, Birmingham, United Kingdom

A R T I C L E I N F O A B S T R A C T

Keywords: Introduction: The last decade has seen a rapid increase in the use of artificial intelligence (AI) approaches such as
Sport science machine learning and deep learning in the sport sciences. However, despite the increased interest in this area, the
Machine learning scope and utility of, and challenges associated with, these approaches are relatively unknown.
Artificial intelligence
Methods: This critical review aimed to scan sport science research for articles using AI in athlete development
Review
Deep learning
contexts (i.e., talent/athlete identification, talent/athlete selection, and talent/athlete development). Through
database, external reference lists, book chapters, and other relevant resource searching, information from eligible
articles was extracted and form the basis of the current review.
Key takeaways: The use of AI was prominent in three main areas: improving athlete assessment, athlete selection
and classification, and athlete development and training. These technologies have been used in a variety of ways
and appear to have potential value for those working in this area. The challenges associated with these ap­
proaches are also discussed.
Conclusion: AI in the context of athlete development allows for access to more data, more easily, and with greater
statistical complexity, than ever before. Importantly, a balanced approach that embraces both innovation and
critical evaluation seems necessary to ensure these tools enhance, rather than disrupt, the athlete identification,
selection, and development landscape.

In 2003, Michael Lewis’s Moneyball became an international best- become nearly ubiquitous as the adjective of choice when describing a
selling book, detailing how the Oakland A’s professional baseball team new (or re-packaged) technology. While the hype around AI is almost
used modern statistical approaches to gain advantages over their less certainly overblown (at least for the moment; Floridi, 2024), there have
informed or dubious competitors (Lewis, 2003). Fast forward nearly a been important developments in how these approaches have been used
quarter of a century later, and the integration of advanced statistics has in sport settings (for a review see McAuley et al., 2024). In parallel, the
become mainstream and nearly universal in high-performance and rapid increase in the amount of data and information from many do­
professional sport, as well as widely used in many levels of develop­ mains of high-performance sport have created a ‘perfect storm’ of data
mental and recreational sport. Their value for improving predictive availability and increased capacity for data management.
models in these settings, however, remains debatable (see Jones et al., It is reasonable to assume that the increased availability of infor­
2021 for an example). mation would expand our understanding of athlete development and
Statistical approaches to predicting human behaviour and perfor­ improve our predictions of athletes in talent identification contexts (e.g.,
mance, and the tools underpinning them, have continued to evolve. The making selections about an athlete future potential based on available
past decade has seen a rapid emergence in artificial intelligence (AI) (i. metrics). Indeed, increases in the use and quality of technology to
e., computer assisted approaches such as deep-learning or machine monitor athlete performance and inform athlete training (i.e., athlete
learning) in nearly all aspects of sport science, and the use of AI has telemetry) should be a boon for researchers in this area, particularly

This article is part of a special issue entitled: PSE 25th Anniversary published in Psychology of Sport & Exercise.
* Corresponding author.
E-mail address: [Link]@[Link] (J. Baker).

[Link]
Received 25 February 2025; Received in revised form 9 August 2025; Accepted 20 August 2025
Available online 15 September 2025
1469-0292/© 2025 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC license ([Link]
nc/4.0/).
J. Baker et al. Psychology of Sport & Exercise 82 (2026) 102978

when using study designs where sample sizes are limited and informa­ reviewed journal articles (Grant & Booth, 2009; Reese, 2023). They
tion quality is at a premium (e.g., research on domain experts; Barra­ can also be helpful for laying the foundation for future, more structured
clough et al., 2022; Johnston et al., 2018). work in the space, especially for fast-growing topics where interpreta­
In this critical review, we focus on how AI has been applied in the tion, synthesis, or theory-building is needed, such as the present topic for
context of athlete/talent development, identification, and selection AI and ML in sport. For the review, a blended approach was used to
(collectively referred to as ‘athlete development contexts’). Broadly, we search and retrieve articles, mirroring (in the early stages) that of a
use AI to reflect analytical approaches drawn from machine learning scoping review, while also applying the creative liberties allotted with a
(ML), and deep learning (DL). These approaches use computer-based critical discussion format.
approximations of human intelligence or decision-making to generate The first step involved identifying the broad focus of our search (i.e.,
analyses and interpretations at a more rapid rate than possible by a how AI has been applied in the context of athlete development, talent
human (for an introduction, see Chmait & Westerbeek, 2021). In this identification, and athlete selection) and required a search within and
rapidly changing landscape, understanding how these technologies have beyond sport journals. To minimize author biases, we considered a large
been used by researchers in athlete development contexts will be useful database of articles recently collected by our research team for a scoping
for identifying the value of different approaches in various contexts, as review of AI usage in sport. This database was collected following a
well as the challenges and opportunities for future research. scoping review process, which involved searching the databases of
Although we were interested in how this broad category of ap­ Scopus, SPORTDiscus, and MEDLINE using the following search terms
proaches have informed research in athlete development, it is worth [“machine learning”, OR “deep learning” OR “neural networks”]. To
noting the differences between these approaches. Typically, this differ­ ensure these terms were captured in context of sport, these terms were
ence is one of specificity in how the programs work. AI is broadly also searched with [sport OR sports OR sporting] and a list of Olympic
defined as ‘aspects of human intelligence executed by machines’ (e.g., sports (to capture a broad range of specific sports). These terms were
Dindorf et al., 2022). The common link between intelligence (as then used to search the title, abstract, and keywords from articles in the
demonstrated by humans or other animals) and artificial intelligence (as databases. The title and abstract of these sources were then read by two
demonstrated by machines) is that the ‘behavior’ is demonstrated by an members of the research team to determine if they met the initial in­
‘intelligent agent’, that is, an agent capable of (a) perceiving their clusion criteria of (a) examining some element of AI, (b) tracking
environment, (b) taking actions to maximize their chance of success at a athlete-related outcomes for at least 1 year, (c) focusing on high-
goal, and (c) improving future performance through learning (Bini, performance sport, and (d) available in full text and in English. If the
2018; Poole et al., 1998). Within AI, ML refers to approaches to analyses article met the criteria, it was moved to the full-text reading phase,
where computers ‘learn’ from existing datasets and take that learning to where two of the authors read the full-text to confirm it met the pre­
subsequent analyses (Dindorf et al., 2022). This type of analysis does not liminary inclusion criteria. Only articles that were not published in
require explicit programming, instead, the computer learns patterns English or not peer-reviewed articles were removed. From this database,
within the datasets provided and transfer these patterns to new data. we conducted a ‘rapid review’ (Smela et al., 2023) of key articles, which
Deep learning is an approach to ML that uses neural networks (i.e., were identified through a search of key terms of ‘athlete development’,
models of the structure between variables) to develop multi-layered ‘talent identification’, and ‘athlete selection’.
approaches to the analysis of large datasets. After identifying these articles from the larger review, external
Below, we build upon and extend reviews in related areas. Dindorf sources were identified from book chapters, research briefs, and con­
and colleagues;(2022), for example, conducted a bibliometric analysis of ference proceedings. All sources of information were then compiled into
1125 articles on AI in sport. A bibliometric analysis is a broader statis­ a tentative final list. The final step in our process involved three mem­
tical approach that aims to capture the structure, trends, and impact of a bers of the author team meeting to discuss how the articles could be
research area by using quantitative software tools to assess patterns of grouped into meaningful categories to be discussed for the context of
knowledge production and accumulation (Donthu et al., 2021). Dindorf this paper. Below, an overview of the landscape of prior research is
et al. highlighted the rate at which research in this area is growing, provided in three broad areas. Following this, we assessed this back­
noting that the field has generally been dominated by a small number of ground relative to gaps in our understanding, opportunities for future
articles that account for 32.52 % of total citations, largely from China, work, and challenges for those working in this area.
the USA, and Germany. The authors also noted a relative lack of infor­
mation (at least from their search strategy) related to AI in the fields of 2. Findings
sports pedagogy, sports sociology, and sports economics. Unlike bib­
liometric analyses, this review has a more specific focus – how these 2.1. Improving athlete assessment
statistical advances have been used in athlete identification (i.e., recog­
nizing participants who are more likely to become high-performers in Several studies focused on the development or testing of new models
the future), development (i.e., monitoring and predicting changes in for assessing athletes. While these did not always focus on development
performance over time), and selection (i.e., choosing participants who contexts specifically, they were included in our discussion because of
have displayed a requisite measure of performance on some outcome of their relevance for improving the accuracy or complexity of athlete
value). assessment processes, which occurs regularly during the timeframe of
athlete development. Often, ML approaches were used to explore the
1. Methods inter-relations between variables at greater levels of complexity than
permitted in traditional statistical approaches. For instance, Duncan
A critical review approach was chosen to discuss how AI is being et al. (2023) used a ML approach to consider the value of various
used in the context of athlete development. This method allowed for a anthropometric, technical, and psychological factors, as well as coach
widespread search using both formal and information strategies to scan, ratings to understand technical skills of youth English male grassroots
search, retrieve and report on articles (and other information sources) soccer players. Out of 40 different predictors, total functional movement
within and beyond a confined search. While critical reviews (and similar screening score (which reflects an assessment of fundamental movement
formats like conceptual essays) are criticized for their degree of bias, patterns) was the most important feature in predicting technical soccer
lack of structure, and lack of empirical findings, they can be helpful for skills, followed by coach rating of player skills for their age, years of
casting a broad net around articles in a broad research area (in this case, playing experience, and age at peak height velocity.
spanning sport sciences, computer sciences, mathematical sciences, etc.) In another example, Cameron et al. (2024) used ML to develop a set
and allowing for more flexible searches in resources beyond peer- of novel performance metrics for quantifying a player’s overall

2
J. Baker et al. Psychology of Sport & Exercise 82 (2026) 102978

contribution to team success in rugby league. These included measures youth rugby union (Owen et al., 2022), and (d) high and low performers
of both offensive (e.g., wins created) and defensive (e.g., losses created) in volleyball (Musa et al., 2023). While less common, there has been
play that could be combined into a global metric of net wins added. The some exploration of these approaches in individual sports including
value of these types of indicators is not limited to elite levels of devel­ rowing (Liu et al., 2023), fencing (Aresta et al., 2022), archery (Musa
opment, they may have important implications for tracking progress et al., 2019), skateboarding (Ab-Rasid et al., 2024), and tennis (Siener
across the extensive timeline of high-performance skill acquisition. et al., 2021). While these studies provide insight into the differences
Kohn et al. (2024), for instance, evaluated a multidimensional battery of between selected and non-selected (or equivalent) cohorts, they fail to
strength measures, functional movement abilities, and anthropometric capture how sports, and athlete development endpoints, evolve over
characteristics in a cohort of collegiate baseball players. Their results time. Indeed, the main purpose of athlete selection is to recruit young
delineated the contributions of different combinations of factors to athletes with the greatest potential to excel in the long-term, which in­
in-game baseball performance (e.g., the amount of variance accounted volves predicting how sports will change in the future. Predicting how
for by anthropometric factors versus functional movement scores). By sports will change over time is critical for understanding the charac­
using a validated assessment and multidimensional approach, these data teristics needed to excel across different sports, and for developing
may be important for establishing comparative benchmarks throughout appropriate environments for these characteristics to be acquired. Un­
the athlete development pathway. fortunately, much of this work is cross-sectional or over shorter time
New approaches have also been applied in performance analysis in periods (e.g., single seasons), and there are limits to value of these de­
individual sports. In a recent study of the velocity-time curve in signs for informing athlete development models over considerably
sprinting, for example, Tam and Yao (2024) found support for two ML longer timeframes.
models that were more accessible and accurate despite using fewer
variables. Another investigation (Lai et al., 2018) used a network anal­ 2.3. Athlete training and development
ysis to explore how the quality of a player’s opponents affects their
long-term ranking and performance. Such approaches to athlete as­ Within athlete development contexts, AI has also been used to assess
sessments have led to several professional sports clubs and organizations participation and training patterns. For instance, Barth et al. (2019) used
investing significant resources (i.e., human, financial, time) into AI ap­ a methodical approach that applied decision trees and gradient boosting
plications, despite little empirical research demonstrating their to previously published data to detect differences between sports per­
effectiveness. formers at international- and national-level with regards to main-sport
and other-sport practice volume. Compared to the more traditional
2.2. Athlete selection and classification statistical approach used in previous studies, the author’s concluded AI
was superior and claimed such findings provide evidence to support
In addition to studies of how AI could be used to improve the investigators re-analyzing previously published data using more so­
assessment of athletes, several researchers have focused on how these phisticated contemporary approaches. Continuing in a similar vein,
tools may improve athlete selection contexts. For instance, Jauhiainen Barth et al. (2020) used a supervised ML method to explore how su­
et al. (2019) used a dataset from 14-year-old Finnish male youth soccer pervised and coach-guided practice in athletes’ primary and other sports
players (n = 951) to train a ML algorithm to detect possible future ‘elite’ impacted their performance and success at attaining international
players (i.e., those who subsequently signed for an academy at 16-year­ medals. The study revealed that ‘coach-led practice in other sports’ until
s-old). The most accurate model included tests of technical skills, speed, the age of 14 was the most significant factor associated with achieving a
and agility. Similarly, Kelly et al. (2022) used a cross-validated Lasso podium finish. The authors concluded that the combined results of these
regression to examine characteristics of selected (i.e., offered a profes­ two studies emphasize the value of combining traditional statistical
sional contract) and deselected (i.e., not offered a professional contract) approaches with AI to enhance testing and uncover patterns between
male U18 England academy soccer players. Out of 53 analyzed features, variables.
a greater perceived ability to handle performance and developmental In another example, Güllich et al. (2019) used AI to analyze differ­
pressures, along with a strong capacity to organize and engage in quality ences between the developmental biographies of athletes who had won
practice, were significant factors in securing a professional contract. Olympic and world titles compared to international athletes who did not
In cricket, Jones and colleagues (2019, 2020) have conducted several win any major medal. Using a ML approach known as pattern recogni­
studies using ML approaches to explore the value of developmental tion analysis, the authors found that more successful athletes: (a)
variables in predicting skill level. In one study (Jones et al., 2019) they experienced an early negative life experience close to a significant
used non-linear pattern recognition to analyze 93 features from senior positive sports milestone, (b) placed a higher relative importance on
international and professional (i.e., First-Class County) male spin bow­ sport over other life aspects, (c) exhibited stronger obsessiveness/­
lers in England. A subset of twelve developmental features, including an perfectionism and sport-related ruthlessness/selfishness, and (d) had
earlier engagement in cricket, greater quantity of domain-specific coaches who more appropriately aligned with their physical and psy­
practice and competition, and superior adaptability to new levels of chosocial requirements. The use of AI was highlighted by the authors as
competition (e.g., capacity to overcome developmental challenges), being a main strength of the study and suggests developmental model­
discriminated between the international and professional groups. The ling could be a fruitful area of further research in sport.
authors used a similar approach to analyze differences between ‘su­ Recently, Anderson et al. (2022) used AI in a longitudinal, pro­
per-elite’ (i.e., predetermined high-profile senior international) and spective design to holistically model the development of Olympic
‘elite’ (i.e., predetermined high-profile domestic senior professional) weightlifters. The authors employed four ML algorithms (i.e., Naïve
English male cricket batters (Jones et al., 2020) on 658 features. The Bayes, J48 decision tree, Support Vector Machine and K-nearest neigh­
groups were differentiated by a subset of 18 variables, with ‘super-elite’ bour) based on a multidisciplinary framework of expertise themes. The
batters undertaking a larger volume of skills-based practice that was summary model produced from this analysis selected a critical set of
both more random and varied in nature at age 16 years (see also Brown nine features that classified high performers and low performers, with
et al., 2024). 91 % average accuracy. Collectively, these studies showcase the utility
In addition to the studies discussed above on soccer and cricket, ML of AI for supporting prospective, longitudinal, and multidisciplinary
techniques have been especially prominent in team sports contexts such methodological approaches in studies of athlete development.
as predicting (a) different types of movements in handball players
(Oytun et al., 2020), (b) future career attainment in rugby league (Till
et al., 2016), (c) who gets selected at regional talent selection camps in

3
J. Baker et al. Psychology of Sport & Exercise 82 (2026) 102978

3. Challenges to using AI in athlete development contexts Fonseca and colleagues (2022) illuminated in their narrative review of
data ownership for individual sport athletes, it is usually the athletes
For good or bad, these technologies are rapidly becoming part of who are negatively affected (and sometimes exploited) by the data that
many athlete development contexts around the world. One example was they themselves generate, making it even more critical their voices are
highlighted in a news report in January 2024 (Smith, 2024); Premier heard in data ownership decisions.
League football club Chelsea will use a new app – aiScout – that allows As a result of these growing concerns for ethical, respectful, and
players to upload video clips of themselves doing specific drills and skills rigorous data ownership practices, regulations, laws, and agreements
that clubs can assess far away from the playing field. Other professional (within and beyond sport) are integrating data ownership clauses and
clubs are following suit. While these approaches can be criticized in a practices. For example, the National Basketball Association (NBA)
variety of ways (e.g., lack of consideration of negative effects, rapid banned the use of wearable technology data in contract negotiations and
implementation without establishing validity and reliability of the player transactions as a way to mitigate the wrongful use and exploi­
outputs, potential for increasing the divisions between ‘have’ and ‘have tation of athlete data (Socolow & Jolly, 2017). On the one hand, this
not’ organizations and countries), these criticisms may overshadow the presents a hopeful step forward in athlete advocacy, but on the other
potential for positive changes to an athlete development system with hand, it creates questions regarding third party access for monetization,
well-known flaws and limitations. In the section below we discuss like enhancing broadcasting viewership and fan engagement, and
several challenges and opportunities related to the use of AI in sport further use and sharing of athlete data.
contexts.
3.3. Increasing specialization of skills
3.1. Establishing validity of emerging methods
A final challenge relates to the specialized knowledge necessary to
The reality of technological advancements in high performance sport work and thrive in the context of AI. Current theories and policies
is one of constant change. This can present challenges for scientists and emphasize a ‘holistic’ or ‘systems-level’ approach to athlete develop­
product developers trying to establish clear evidence for the validity, ment (Gulbin et al., 2013; Thompson et al., 2022), which requires
reliability, sensitivity, and specificity of new measures (Saah & Hoover, diverse knowledge of multiple areas. Simultaneously, the knowledge
1997). However, many of these new technologies (including aiScout) and methods being used in this field are increasing at incredible speeds,
have a key advantage over traditional measures – greater ease of which has the potential to create barriers between those who understand
accessing athlete data. This has been a critical limitation to traditional and use these approaches (e.g., analytical teams) and those who do not
research designs in high-performance sport, which has relied on small (e.g., coaches, trainers). This places pressure on both groups, the former
samples of hard to access groups. Concerns about measurement quality may be ill equipped to explain the complexity of their analysis to a group
may be offset to some extent by the power of large samples, which are without their training (i.e., the curse of knowledge; Pinker, 2015), the
more representative of the populations they are meant to reflect. latter may feel less inclined to ask questions when they do not under­
However, determining the value of new sources of data will remain a stand due to fears of embarrassment.
challenge for several reasons. First, new approaches to data acquisition In a review of AI in sport, Hammes et al. (2022) highlighted five
and modeling assume that we know what the key sources of information broad challenges researchers face, which are also appropriate here.
are. While this issue may be less important in data analytics settings These challenges relate to (a) assessing the right types and amounts of
where the outcome of interest is performance, the extensive time hori­ data for AI to be useful, (b) linking the AI and sport communities to
zon for predictions during development make this substantially more allow creation of, and access to, the data required, (c) allowing practi­
complex and difficult. Relatedly, approaches to ML that use existing tioners such as coaches and trainers to maintain a sense of control, (d)
datasets may be using data from cohorts that are appreciably different ensuring AI outputs are explainable and that end users to not require
from the ones they are trying to predict (i.e., cohort effects). As a result, specialized knowledge, and (e) remembering AI outputs are predictive
things that were relevant in the past for explaining why certain athletes/ rather than explanatory, which constrains their capacity to directly
groups were successful may no longer be relevant because the contexts inform theory development and/or testing.
have changed (see Baker et al., 2022).
4. Designing the future: directions for further research
3.2. Emerging issues in data management
AI has the potential to revolutionize how athletes train, how coaches
Arguably the most significant change in many fields of sport science coach, and how scouts evaluate. At the same time, embracing these
as a result of AI, is the amount of available data. Advancements in techniques without a thoughtful and proactive approach could waste
computer vision have provided non-invasive measures of biomechanical time (best case) and/or lead to poor policy built on inaccurate or
information (e.g., Fitzpatrick et al., 2023; Mecheri et al., 2016), the use incomplete information (worst case). While future research in this area
of lidar and doppler radar allow for collection of heart rate data at a may wish to embrace the obvious advantages AI will bring to athlete
distance (Bouazizi et al., 2022) and the integration of 3D tracking sys­ identification and development contexts, researchers will also need to
tems allows athlete tracking and assessment in arenas, stadiums, and acknowledge the potential limitations of these approaches.
ballparks. While previous research would require the recruitment of
athletes for sport science research, often involving invasive data 4.1. Determining the limits of AI
collection methods, the future of data collection will largely be ‘outside
the athlete’. Despite the rhetoric about the potential for widely available data
As a direct result of this influx of data generated by AI, there are leading to an ‘end of theory’ (Anderson, 2008), there is little chance
growing concerns about the ownership of the data that is being collected these technologies will be a panacea, at least for figuring out the com­
and analyzed in droves. Ownership, in this sense, refers to the storage, plex process of athlete development. A human-machine symbiosis
accessibility, usage, and sale of data generated by athletes. As a result, (Jarrahi, 2018; Licklider, 1960) is much more realistic, as many AI ex­
interest-holders find themselves in challenging positions for deter­ perts hold the position that humans combined with computers will
mining rightful owners of the data. Unfortunately, it seems that even perform better on many things than either one alone (Davenport, 2016;
though the data is generated by athletes (either through wearable Siegel, 2016; Yun et al., 2021), and in the words of Jarrahi (2018) “AI
technology, or external motion capture), the athletes themselves are systems should be designed with the intention of augmenting, not
rarely, decision makers in the process of determining ownership. As replacing, human contributions” (p.1).

4
J. Baker et al. Psychology of Sport & Exercise 82 (2026) 102978

For the best effect/outcome, it is important to determine the et al., 2018). Finally, and perhaps most critically, this increased data
strengths and weaknesses of humans and computers for a given task. On comes with significantly less demand on athletes, in most situations.
the one hand, computers have already surpassed humans in accuracy While future research is needed to explore the benefits and draw­
and speed in accomplishing some quantitative tasks (Parry et al., 2016). backs of these technologies in these contexts, providing guidance in how
A computer has capabilities for knowledge generation, knowledge syn­ (and how not) to use it would be a valuable next step. It is obviously
thesis, and memory recall, at a rate far beyond that of any human or beyond the scope of this paper to discuss all the ways AI will affect
group of humans (Simon, 1982). Humans on the other hand, outperform athlete development contexts, from increasing the depth and breadth of
AI in the more subjective and qualitative tasks, and likely will for a long data available for sport scientists and practitioners and improving ap­
while. By utilizing the best of both humans and computers’ strengths, proaches to statistical analysis and data modelling (Bai & Bai, 2021) to
there can be powerful synergistic benefits. This may be in part related to influencing the training of athletes, coaches, and other practitioners
the notion that a computer (with appropriate information) might be less (Passmore & Woodward, 2023) and predicting future trends in sport
prone to certain types of bias when forming judgments and making science and performance (Naik et al., 2022). Despite the considerable
decisions (Bersoff, 2001; Kamiran & Calders, 2010). In their work, hype about the value of AI and the extent to which it will solve many of
Grgić-Hlača and colleagues (2019), ran a series of experiments to the world’s most complex problems (Floridi, 2024), it is difficult to
investigate whether machine generated advice improves the quality of imagine a single area in sport science (indeed, in the world more
human decisions in a judicial context. Findings from their experiment generally) that will remain as it was before the advent of these tech­
indicate there are certain conditions where the advice of computers can nologies. A thoughtful and rational discussion may provide important
help with the decision quality and accuracy. The authors conclude there clues as to how AI can be integrated to improve athlete identification,
is benefit from designing and using powerful interfaces between development, and selection contexts. While the riddle of athlete devel­
advice-giving machines and advice-receiving human decision makers. opment remains unsolved, at least for the foreseeable future, AI will help
Athlete scouting is a perfect example how augmentation is possible us contribute more data, more easily, and with greater statistical
for ‘real world’ difficult decision-making in complex and dynamic en­ complexity than ever before. Importantly, a balanced approach that
vironments. While a computer can assess and crunch numbers on an embraces both innovation and critical evaluation is necessary to ensure
athlete and his/her/their performance statistics in a very reliable way, these tools enhance, rather than disrupt, the athlete identification, se­
no advances in computer vision can ‘see’ the character qualities that a lection, and development landscape.
scout can glean from meeting with the athlete, speaking with his/her/
their parent(s), coach(es), etc. When blended, however, utilizing the CRediT authorship contribution statement
scout’s expert eye with the number-crunching capabilities of the com­
puter, there is likely greater strength than either system alone. Take the Joseph Baker: Writing – review & editing, Writing – original draft,
example of medical judgment and decision-making. In the context of Conceptualization. Antonia Cattle: Writing – review & editing, Writing
cancer detection in the images of lymph node, a fully AI approach had a – original draft, Methodology. Alex McAuley: Writing – review &
7.5 % error rate, whereas the pathologists in the sample had a 3.5 % editing, Writing – original draft. Adam Kelly: Writing – review &
error rate. When examining how the pathologists faired when editing, Writing – original draft. Kathryn Johnston: Writing – review &
combining inputs from both AI and the experts, this resulted in an error editing, Writing – original draft, Methodology, Conceptualization.
rate of 0.5 % (an 85 % improvement; Wang et al., 2016). A key area of
future work will be determining which aspects of these tools are
ground-breaking and revolutionary, which are useful extensions and Declaration of competing interest
time-saving options of currently available tools, and which are distrac­
tions that lead researchers away from actual solutions to the challenges The authors declare that they have no known competing financial
in this area. Relatedly, given the complex process of athlete (and human) interests or personal relationships that could have appeared to influence
development, it will be important to understand the psychosocial risks the work reported in this paper.
associated with AI-driven datafication, especially in relation to the
treatment of young athletes in environments increasingly focused on Data availability
performance outcomes (see Lupton & Williamson, 2017; Millington &
Millington, 2015). No data was used for the research described in the article.

4.2. Integrated approaches to holistic athlete development

References
Traditional approaches to sport science have been largely siloed by Ab-Rasid, A. M., Musa, R. M., Abdul-Majeed, A. P. P., Musawi-Maliki, A. B. H.,
discipline (e.g., physiology, psychology, biomechanics). The limits of a Abdullah, M. R., Razmaan, M. A. M., & Abu-Osman, N. A. A. (2024). Physical fitness
uni-dimensional method for understanding athlete performance and and motor ability parameters as predictors for skateboarding performance: A logistic
regression modelling analysis. PLoS One, 19(2), Article e0296467. [Link]
development are well known (e.g., Johnston et al., 2018). However, the 10.1371/[Link].0296467
rapid availability of new data may help researchers and other stake­ Anderson, C. (2008). The end of theory: The data deluge makes the scientific method
holders to build more holistic models of athlete wellbeing, performance, obsolete. June 23rd Wired Magazine.
Anderson, D. N. J., Gottwald, V. M., & Lawrence, G. P. (2022). Capturing the holistic
and development. Increases in the amount of availability data across a profile of high performance olympic weightlifting development. Frontiers in Sports
wider range of categories would allow researchers to embrace methods and Active Living, 4. [Link]
grounded in complexity science or other integrated approaches to Aresta, S., Bortone, I., Bottiglione, F., Di Noia, T., Di Sciascio, E., Lofù, D., Musci, M.,
Narducci, F., Pazienza, A., Sardone, R., & Sorino, P. (2022). Combining
human performance and development. The potential value of AI is re­ biomechanical features and machine learning approaches to identify fencers’ levels
flected in the greater ease at which the increasingly available informa­ for training support. Applied Sciences, 12(23), Article 12350. [Link]
tion in many aspects of sport science is integrated into manageable 10.3390/app122312350
Bai, Z., & Bai, X. (2021). Sports big data: Management, analysis, applications, and
datasets. Importantly, the accessibility of much of this information is
challenges. Complexity, 2021(1), Article 6676297. [Link]
broader than ever, which, over time, should level the playing field 6676297
beyond countries who have traditionally dominated in the realm of sport Baker, J., Johnston, K., Wojtowicz, M., & Wattie, N. (2022). What do we really know
science. There is also a greater breadth of data available, meaning more about elite athlete development? Limitations and gaps in current understanding.
British Journal of Sports Medicine, 56(23), 1331–1332.
complicated models of interactions amongst variables can be consid­ Barraclough, S., Till, K., Kerr, A., & Emmonds, S. (2022). Methodological approaches to
ered, which has been a noted limitation of previous research (Johnston talent identification in team sports: A narrative review. Sports, 10(6), 81.

5
J. Baker et al. Psychology of Sport & Exercise 82 (2026) 102978

Barth, M., Emrich, E., & Güllich, A. (2019). A machine learning approach to “revisit” Jones, J., Johnston, K., Farah, L., & Baker, J. (2021). Predicting seasonal performance in
specialization and sampling in institutionalized practice. Sage Open, 9(2). https:// professional sport: A 30-year analysis of sports illustrated predictions. Sports, 9(12),
[Link]/10.1177/2158244019840554, 2158244019840554. 163. [Link]
Barth, M., Güllich, A., Raschner, C., & Emrich, E. (2020). The path to international Kamiran, F., & Calders, T. (2010). Classification with no discrimination by preferential
medals: A supervised machine learning approach to explore the impact of coach-led sampling. In In proceedings of the 19th ACM conference on information and knowledge
sport-specific and non-specific practice. PLoS One, 15(9), Article e0239378. https:// management (pp. 1–10). [Link] ACM.
[Link]/10.1371/[Link].0239378 Kelly, A. L., Williams, C. A., Cook, R., Jiménez, S. L., & Wilson, M. R. (2022).
Bersoff, D. M. (2001). Why good people sometimes do bad things: Motivated reasoning A multidisciplinary investigation into the talent development processes at an English
and unethical behavior | emerald insight. The Next Phase of Business Ethics: Integrating football academy: A machine learning approach. Sports, 10(10), 159. [Link]
Psychology and Ethics, 3, 239–262. [Link] org/10.3390/sports10100159
Bini, S. A. (2018). Artificial intelligence, machine learning, deep learning, and cognitive Kohn, J. N., Lochhead, L., Feng, J., Bobb, R., & Appelbaum, L. G. (2024). Strength, speed,
computing: What do these terms mean and how will they impact health care? The and anthropometric predictors of in-game batting performance in baseball. Journal
Journal of Arthroplasty, 33(8), 2358–2361. [Link] of Sports Sciences, 42(8), 720–727. [Link]
arth.2018.02.067 02640414.2024.2363679
Bouazizi, M., Yamamoto, K., & Ohtsuki, T. (2022). Heartbeat detection using 3D lidar Lai, M., Meo, R., Schifanella, R., & Sulis, E. (2018). The role of the network of matches on
and mimo doppler radar. In ICC 2022-IEEE international conference on predicting success in table tennis. Journal of Sports Sciences, 36(23), 2691–2698.
communications. IEEE. [Link] [Link]
Brown, T., Cook, R., Gough, L. A., Khawaja, I., McAuley, A. B., & Kelly, A. L. (2024). Lewis, M. (2003). Moneyball: The art of winning an unfair game. WW Norton and
Exploring the multidimensional characteristics of selected and non-selected white Company.
British and British South Asian youth cricketers: An exploratory machine learning Licklider, J. C. R. (1960). Man-computer symbiosis. IRE Transactions on Human Factors in
approach. All About You, 4(2), 718–734. [Link] Electronics, (1), 4–11. [Link] HFE-1.
Cameron, S., Radwan, I., & Mara, J. (2024). What the eyes don’t see: An objective Liu, J. W., Chen, S. H., Chen, C. H., & Huang, T.-H. (2023). Constructing an artificial
assessment of players’ contribution to team success in men’s rugby league. Research intelligence strategy algorithm for the identification of talented rowing athletes. Soft
Quarterly for Exercise & Sport. [Link] Computing, 27, 1743–1750. [Link]
Chmait, N., & Westerbeek, H. (2021). Artificial intelligence and machine learning in Lupton, D., & Williamson, B. (2017). The datafied child: The dataveillance of children
sport research: An introduction for non-data scientists. Frontiers in Sports and Active and implications for their rights. New Media & Society, 19(5), 780–794.
Living, 3, Article 682287. [Link] McAuley, A. B. T., Baker, J., Johnston, K., & Kelly, A. L. (2024). Artificial intelligence in
Davenport, T. H. (2016). Rise of the strategy machines. MIT Sloan Management Review, 58 talent identification and development in sport. In C. Dindorf, E. Bartaguiz,
(1), 29. F. Gassmann, & M. Fröhlich (Eds.), Artificial intelligence in sports, movement, and
Dindorf, C., Bartaguiz, E., Gassmann, F., & Fröhlich, M. (2022). Conceptual structure and health (pp. 255–270). Springer. [Link]
current trends in artificial intelligence, machine learning, and deep learning research Mecheri, S., Rioult, F., Mantel, B., Kauffmann, F., & Benguigui, N. (2016). The serve
in sports: A bibliometric review. International Journal of Environmental Research and impact in tennis: First large-scale study of big hawk-eye data. Statistical Analysis and
Public Health, 20(1), 173. Data Mining: The ASA Data Science Journal, 9, 310–325. [Link]
Donthu, N., Kumar, S., Mukherjee, D., Pandey, N., & Lim, W. M. (2021). How to conduct sam.11316
a bibliometric analysis: An overview and guidelines. Journal of Business Research, Millington, B., & Millington, R. (2015). ‘the datafication of everything’: Toward a
133, 285–296. [Link] sociology of sport and big data. Sociology of Sport Journal, 32(2), 140–160.
Duncan, M. J., Eyre, E. L. J., Clarke, N., Hamid, A., & Jing, Y. (2023). Importance of Musa, R. M., Abdul Majeed, A. P. P., Suhaimi, M. Z., Abdullah, M. R., Razman, M. A. M.,
fundamental movement skills to predict technical skills in youth grassroots soccer: A Abdelhakim, D., & Abu Osman, N. A. (2023). Identification of high-performance
machine learning approach. International Journal of Sports Science & Coaching, 19, volley-ball players from anthropometric variables and psychological readiness: A
1042–1049. [Link] machine-learning approach. Journal of Sports Engineering and Technology, 237(4),
Fitzpatrick, A., Stone, J. A., Choppin, S., & Kelley, J. (2023). Analysing hawk-eye ball- 317–324. [Link]
tracking data to explore successful serving and returning strategies at wimbledon. Musa, R. M., Abdul Majeed, A. P. P., Taha, Z., Chang, S. W., Nasir, A. F. A., &
International Journal of Performance Analysis in Sport, 24(3), 251–268. [Link] Abdullah, M. R. (2019). A machine learning approach of predicting high potential
org/10.1080/24748668.2023.2291238 archers by means of physical fitness indicators. PLoS One, 14(1), Article e0209638.
Floridi, L. (2024). Why the AI hype is another tech bubble. Philosophy & Technology, 37 [Link]
(4), 128. [Link] Naik, B. T., Hashmi, M. F., & Bokde, N. D. (2022). A comprehensive review of computer
Fonseca, N., Marques, M. C., & Esteves, D. (2022). Data ownership in individual sports: vision in sports: Open issues, future trends and research directions. Applied Sciences,
Narrative review. Journal of Physical Education and Sport, 22(5), 1214–1218. https:// 12(9), 4429. [Link]
[Link]/10.7752/jpes.2022.05153 Owen, J., Owen, R., Hughes, J., Leach, J., Anderson, D., & Jones, E. (2022). Psychosocial
Güllich, A., Hardy, L., Kuncheva, L., Laing, S., Evans, L., Rees, T., Abernethy, B., Côté, J., and physiological factors affecting selection to regional age-grade rugby union
Warr, C., & Wraith, L. (2019). Developmental biographies of olympic super-elite and squads: A machine learning approach. Sports, 10(3), 35. [Link]
elite athletes: A multidisciplinary pattern recognition analysis. Journal of Expertise, 2 sports10030035
(1), 23–46. [Link] Oytun, M., Tinazci, C., Sekeroglu, B., Acikada, C., & Yavuz, H. U. (2020). Performance
19_2_1_Gullich.html. prediction and evaluation in female handball players using machine learning
Grant, M. J., & Booth, A. (2009). A typology of reviews: An analysis of 14 review types models. IEEE Access, 8, 116321–116335. [Link]
and associated methodologies. Health Information and Libraries Journal, 26(2), ACCESS.2020.3004182
91–108. [Link] Parry, K., Cohen, M., & Bhattacharya, S. (2016). Rise of the machines: A critical
Grgić-Hlača, N., Engel, C., & Gummadi, K. P. (2019). Human decision making with consideration of automated leadership decision making in organizations. Journal of
machine assistance: An experiment on bailing and jailing. Proceedings of the ACM on Business Research, 41(5), 571–594. [Link]
Human-Computer Interaction, 3(CSCW), 1–25. [Link] Passmore, J., & Woodward, W. (2023). Coaching education: Wake up to the new digital
Gulbin, J. P., Croser, M. J., Morley, E. J., & Weissensteiner, J. R. (2013). An integrated and AI coaching revolution. International Coaching Psychology Review, 18(1), 58–72.
framework for the optimisation of sport and athlete development: A practitioner [Link]
approach. Journal of Sports Sciences, 31(12), 1319–1331. [Link] Pinker, S. (2015). The sense of style: The thinking person’s guide to writing in the 21st century.
02640414.2013.781661 Penguin Books.
Hammes, F., Hagg, A., Asteroth, A., & Link, D. (2022). Artificial intelligence in elite Poole, D. I., Goebel, R. G., & Mackworth, A. K. (1998). Computational intelligence, 1.
sports: A narrative review of success stories and challenges. Frontiers in Sports and Oxford University Press.
Active Living, 4, 1466. [Link] Reese, S. D. (2023). Writing the conceptual article: A practical guide. Digital Journalism,
Jarrahi, M. H. (2018). Artificial intelligence and the future of work: human-AI symbiosis 11(7), 1195–1210. [Link]
in organizational decision making. Business Horizons, 61(4), 577–586. [Link] Saah, A. J., & Hoover, D. R. (1997). “Sensitivity” and “specificity” reconsidered: The
org/10.1016/[Link].2018.03.007 meaning of these terms in analytical and diagnostic settings. Annals of Internal
Jauhiainen, S., Äyrämö, S., Forsman, H., & Kauppi, J.-P. (2019). Talent identification in Medicine, 126, 91–94. [Link]
soccer using a one-class support vector machine. International Journal of Computer 00026
Science in Sport, 18(3), 125–136. [Link] Siegel, R. (2016). 20 years later, humans still no match for computers on the chessboard.
Johnston, K., Wattie, N., Schorer, J., & Baker, J. (2018). Talent identification in sport: A October 24th. NPR [Link]
systematic review. Sports Medicine, 48(1), 97–109. [Link] 99162905/20-years-later-humans-still-no-match-for-computers-on-the-chessboard.
017-0803-2 Siener, M., Faber, I., & Hohmann, A. (2021). Prognostic validity of statistical prediction
Jones, B. D., Hardy, L., Lawrence, G., Kuncheva, L. I., Brandon, R., Bobat, M., & methods used for talent identification in youth tennis players based on motor
Thorpe, G. (2020). It ain’t what you do—It’s the way that you do it: Is optimizing abilities. Applied Sciences, 11(15), 7051. [Link]
challenge key in the development of super-elite batsmen? Journal of Expertise, 3(2), Simon, H. A. (1982). Theories of bounded rationality. In H. A. Simon (Ed.), Models of
144–168. [Link] bounded rationality: Behavioral economics and business organization, 1 pp. 408–423).
Jones, B. D., Hardy, L., Lawrence, G., Kuncheva, L. I., Du Preez, T., Brandon, R., Such, P., MIT Press. [Link]
& Bobat, M. (2019). The identification of “game changers” in England cricket’s Smela, B., Toumi, M., Świerk, K., Francois, C., Biernikiewicz, M., Clay, E., & Boyer, L.
developmental pathway for elite spin bowling: A machine learning approach. Journal (2023). Rapid literature review: Definition and methodology. Journal of Market
of Expertise, 2(2), 92–120. [Link] Access & Health Policy, 11(1), Article 2241234. [Link]
20016689.2023.2241234

6
J. Baker et al. Psychology of Sport & Exercise 82 (2026) 102978

Smith, A. (2024). Chelsea leave man utd in their wake with ‘revolutionary’ new recruit- Till, K., Jones, B. L., Cobley, S., Morley, D., O’Hara, J., Chapman, C., Cooke, C., &
ment tool. Mirror. [Link] Beggs, C. B. (2016). Identifying talent in youth sport: A novel methodology using
ad-revolutionary-new-31918442. higher-dimensional analysis. PLoS One, 11(5), Article e0155047. [Link]
Socolow, B., & Jolly, I. (2017). Game-changing wearable devices that collect athlete data 10.1371/[Link].0155047
raise data ownership issues. World Sports Advocate, 15(7), 15–17. Wang, D., Khosla, A., Gargeya, R., Irshad, H., & Beck, A. H. (2016). Deep learning for
Tam, C. K., & Yao, Z. F. (2024). Advancing 100m sprint performance prediction: A identifying metastatic breast cancer. arXiv. [Link]
machine learning approach to velocity curve modeling and performance correlation. arXiv.1606.05718
PLoS One, 19(5), Article e0303366. [Link] Yun, Y., Ma, D., & Yang, M. (2021). Human–computer interaction-based decision support
Thompson, F., Rongen, F., Cowburn, I., & Till, K. (2022). The impacts of sports schools on system with applications in data mining. Future Generation Computer Systems, 114,
holistic athlete development: A mixed methods systematic review. Sports Medicine, 285–289. [Link]
52(8), 1879–1917. [Link]