Background

Artificial intelligence (AI) represents the latest industrial revolution, bringing potentially boundless opportunities with transformative advances within medicine, including injury and illness prediction, imaging interpretation and outcome measures, among others.1

AI entered elite sports approximately 30 years ago2 with a particular focus initially on two areas: sports performance and predicting injury.3 AI applications are now growing at a formidable rate, with new applications such as assisting player recruitment, officiating, tactical decision-making and broadcasting.4

The influence of AI within elite sport is impossible to ignore, which led the International Olympic Committee (IOC) to develop the Olympic AI Agenda (OAIA), outlining various focus areas where AI should be leveraged to produce positive benefits. We discuss two priorities below.

Prioritising athlete safety

Embracing AI will transform athlete welfare by offering reactive and predictive applications. World Rugby recently introduced instrumented mouthguards (iMG), which signal in real-time where impacts could indicate head trauma.5 Their use is mandated for elite competitions and codified in rugby laws, with compliance enabling a temporary substitution for head injury assessment. iMGs complement sport concussion assessment tool 66 by providing an additional safeguard whereby head trauma may otherwise go undetected. While currently imperfect, with World Rugby building confidence in iMG use,7 AI support could indicate expected outcomes from impacts, offering directives regarding the next steps of the welfare protocol.

Precedence for real-time monitoring of athlete biometric data to identify event risk and predict athlete injury was highlighted in a recent IOC consensus statement on recommendations and regulations for sport events in the heat,8 with international federations encouraged to proactively consider athlete safety. Proof of a principle example where metrics are used to predict exertional heat stroke is represented in figure 1. AI methods also support the recognition of athlete motion actions to predict injury.9 Collecting and analysing more data, in real-time, will enable at-risk athletes to be properly identified, driving policy change and sports-specific welfare protocols. High-profile incidents such as the collapse of Jonny Brownlee in the 2016 World Triathlon series, and others in football, could be avoided by observing metrics in real-time, allowing athletes to push their limits, safe in the knowledge that their health is monitored and safeguarded.

Figure 1

Proof of principle traffic light system to predict exertional heat stroke (2023 SAFRA Singapore Bay Run—10 km).

Enhancing clean competition

AI also bolsters fair competition by advancing the Athlete Biological Passport (ABP), which monitors haematological variables over time to reveal doping effects. Various analyses have enabled us to understand erythropoietin and its effects on sports performance,10 to identify altered red blood cell production, and associated biomarkers following erythropoietin injections11 and altitude adaptation.12 In future, sequencing hundreds, or even thousands, of gene transcripts and their analysis informed by AI have the potential to allow authorities to accurately determine whether an athlete used prohibited substances, trained at altitude or has an underlying medical condition.

The latest Antidoping Rule Violations Report for 2020 revealed that 0.67% of samples were adverse analytical findings, 66% of which led to sanctions.13 These findings were produced without AI support with the World Anti-Doping Agency yet to implement a formal policy despite investments into projects researching AI use for antidoping. Adverse analytical findings increased to 0.77% in 202214; however, the true prevalence of doping is likely to be significantly higher.15 Future research avenues, including integrative ‘omics’ technologies16 and integrating with relevant confounders, such as altitude exposure12 performance profiling,17 will be supercharged by AI techniques. Machine learning and deep learning will enable the Bayesian approach to the ABP to become more efficient, accurate and reliable. These methods will allow for the identification of subtle trends or anomalies in large and complex data sets, which may be challenging and indeed time-consuming for humans to detect. AI models can be trained on historical ABP data to predict personalised models of an athlete’s biological profile, anticipating expected ranges of biomarkers given their relevant contextual factors, enabling more accurate and sensitive detection of deviations. These models can be updated over time as new research becomes available, improving their accuracy and reducing the risk of false positives or negatives. They can also integrate with other data, for example, training information or environmental conditions, producing a more comprehensive understanding of changes in an athlete’s biological profile. In future, deep learning techniques may also help to identify other biomarkers that should be monitored for doping, of which we are not yet aware.

Understanding risk

AI is expected to revolutionise our understanding of elite sports, including genetic talent identification,18 genetic predisposition to musculoskeletal injury19 and concussion.20 However, we must recognise the risks, including data privacy and hacking, data integrity and manipulation, and ethical issues. The use of genetic and biometric data is now commonplace in society, yet collection of such information represents a significant risk. Protection and management of data are paramount. Athletes must have confidence that their information is safe, cannot be manipulated and will only be used for specifically agreed purposes. Ensuring these criteria will require investment in data infrastructure and strict policy. Regulatory hurdles are one example of related challenges, as variation in policy is highly likely due to territorial differences regarding AI understanding, sentiment and trust.21 We therefore suggest that initial discussions focus on enhancing human welfare as a shared common interest between jurisdictions. We hope this will develop mutually beneficial and aligned AI uses, and associated legislation, which will also support changes in public perception.

Effective data management is non-negotiable, as are the Olympic values. AI use must align with these values and, we suggest, in the ‘spirit’ of sport. Technological impact on sport evolution is evidenced by advanced footwear, whereby performances pre and postintroduction of these technologies are incomparable.22 Governing bodies should therefore scrutinise how AI may impact the spirit of sport. Will artistic ‘beauty’ be lost in favour of scientific components? Will AI reframe history? For example, past performances (ie, those judged without AI support) could be analysed under a new AI-supported system, resulting in cases where athletes wrongly received medals, potentially causing controversy. So, we propose a simple question. How can governing bodies integrate future technologies to enhance elite competition while maintaining the spirit of sport? World Rugby achieved this using iMGs with no material sacrifice, unlike sports, which compromised game flow and spectator enjoyment.23 This should be the benchmark, but if sacrifice is necessary, then what are we willing to accept?

Conclusion

The OAIA represents an important milestone in evolving elite sports, recognising AI as a core component of our future. Stakeholders, led by the IOC, also including governing bodies, athletes, industry and academics, have a pivotal opportunity to shape the Olympic future. Box 1 summarises some key issues needing consideration by Olympic stakeholders to advance the AI agenda. We must prepare our own environments by changing processes and structures to be more dynamic. However, we require collaboration to solve approaching challenges and to harness technologies that improve our sports without losing our spirit. The time to take the lead is now or we risk missing the opportunity to create a better future through the power of sport.

Box 1 Summary of key issues

Education

Coach education courses do not promote the skills sought by sporting directors24 with artificial intelligence (AI) emphasising the need for contemporary methods.

Future practice depends on understanding what needs to be measured, how to use technologies to obtain such data and how this data informs training. For example, the football xG (or expected goals) metric is increasing in prominence. xG uses several variables, including distance and angle, to indicate the likelihood of a shot resulting in a goal, offering insights into player and team qualities and associated tactics.

Understanding how AI models and algorithms have been trained and their relevance to athletes will facilitate AI use.

AI use must remain athlete-centred, guiding athlete learning and aiding decision-making rather than overriding it.

Synthesising AI insights, coaching knowledge and athlete perspectives is fundamental to data optimisation and long-term athlete development.

Governance

Variation in policy and regulation is highly likely due to territorial differences regarding AI understanding, sentiment, and trust.

Ensuring fair access to technologies is likely to have considerable financial implications.

Administrative processes must consider the current lifespan of some AI technologies.

AI could improve a nation’s performance despite their contextual limitations. However, if unaddressed, regulation and associated financial resources could lead wealthier nations to markedly excel, further widening the gap in competition.

Collaboration between sporting organisations, governments and industry will be necessary to resolve these issues.

Speed of change

AI may augment sports, for example, through changing how sports are officiated and judged or by changing formations or styles of play. For example, football recently introduced semiautomated offside technology to support referee decision making, and we are also seeing AI-driven tactical adjustments in major sports, including basketball and football.

These changes may raise the bar for success, drive training innovations or increase injury prevalence through accepting further risk in medal pursuit.

AI may facilitate evolutions in physiological requirements for sports based on additional or reprioritised metrics.

Stakeholders must be prepared for the speed and unknown directions of change.

We should prepare for the future with forward-thinking, open-mindedness and curiosity.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

Not applicable.

Acknowledgments

ITK wrote the first draft of the manuscript, and both authors critically revised each version until both authors could approve the final manuscript. YPP is the guarantor.