The role of Global Sports Organisations (GSO’s) is to protect the integrity of the multi-billion-dollar sports industry through governance and regulation. For example, GSO’s provide governance and regulation for the prevention of doping, financial corruption and match fixing, and child safeguarding, among others. Humanity is on the verge of developing powerful new technologies that will bring about new challenges that could compromise fair-play and athlete wellbeing in elite sport.
This Blog presents a thought experiment on the potential advantages, disadvantages, and unintended consequences of the implementation of advanced artificial intelligence in football.
This blog is adapted from a paper myself and colleagues wrote in 2022. McLean, S., Read, G. J., Thompson, J., Hancock, P. A., & Salmon, P. M. (2022). Who is in control? Managerial artificial general intelligence (MAGI) for football. Soccer & Society, 23(1), 104-109.
What is Artificial General Intelligence (AGI)?
Artificial General Intelligence (AGI) represents the next step in Artificial Intelligence (AI) which will exceed human intelligence in all aspects, making it the most powerful technology ever invented [1]. AI systems in sport are now common for assisting officiating, training, performance prediction/analysis, and injury prediction [2]. Whilst current Artificial ‘Narrow’ Intelligent (ANI) systems can now outperform humans in their own specific domain, they have little intelligence outside of it; hence the term ‘narrow’ intelligence. Thus, the AI algorithm used for officiating would not be effective for injury prediction. Advancing AI’s, will exceed human intelligence on most if not all intellectual tasks; underwritten by their ability to learn, problem-solve, self-regulate, and undertake tasks for which they were not originally designed [3]. This is known as Artificial General Intelligence (AGI). Formal AGI’s do not yet exist but promise to shortly, with predictions around 2040-2070 [4]. The intelligence an AGI will possess is presently incalculable. On average U.S. Students take 8.2 years to earn a PhD [5], however it is estimated that an AGI could accumulate the same knowledge in only a few minutes. At this rate, one AGI could attain could PhDs in virtually every sports science field within a single day. The potential benefits and risks to society have been widely debate, with many focussing on the proposition that humans will no longer be required simply because AGI’s will simply be able to outperform them [6]. Here, we take an envisioned world view to explore this in the realm of sport and consider what could happen when an AGI football manager is developed.
How far are we from AGI’s in sport?
Current ANI systems are principally tools useful in assisting humans, based on their capacity to assimilate and interpret large volumes of data. But this is changing. Standalone forms of ANI are now closer to being fully autonomous. An ANI developed by Google’s DeepMind, named AlphaGo, was programmed to play the ancient Chinese board game of Go [7]. The rules of Go are simple. Two players each take turns to place black or white stones on a board trying to capture the opponent’s stones or surround empty space to secure territory. Despite simple rules, the game is profoundly complex and contains thousands of possible legal positions. Go is thus regarded as the most complex of all board games. In 2016, AlphaGo defeated the 18 times world champion Lee Sedol by four games to one in a five-game series. Match data from AlphaGo indicated that at certain points AlphaGo’s strategising was some 100 moves in advance. To achieve this level of expertise, AlphaGo was trained on data consisting of 30 million moves from 160,000 games of the world’s best Go players. In the context of football, for someone to watch 160,000, 90-minute games would take 240,000 hours (10,000 days), or 27 years nonstop. Whilst AlphaGo is an advanced ANI, it is not AGI. An AGI is anticipated to possess these levels of intelligence across these and many other domains. Given this rate of progress, it is logical to assume that an AGI in football manager (strategist, tactician) is not only feasible but even imminent. Given that football is a multi-billion-dollar industry and ANI is already in widespread use in sport, it is logical to assume that an AGI football manager will quickly appear following the creation of AGI.
How will AGI transform football?
If we believe optimistic projections, a football AGI system would be capable of synthesising every recorded match and training session, player performance profile and health record, every book, blog, magazine piece, and peer-reviewed article on football, available on the internet. This AGI could also use diverse data sources, to train itself and reinforce learning to become the greatest possible source of football knowledge imaginable. It would understand every tactical innovation, recognise game patterns and, to an extent, be able to foresee forthcoming actions. Its value to betting agencies would be incalculable. Beyond the game itself, the AGI manager would recognise how performance is influenced by components within a larger complex system. It would factor in skills assimilation of both acute and chronic expression and understand how each impactful factor is connected. As such, the AGI manager would possess greater levels of intelligence than the most intelligent human exercise physiologist, sport scientist, biomechanist, nutritionist, sports psychologist, football coach, strength coach, and every other scientific discipline involved with the sport. Importantly, it would understand how to weight these varying factors and their interactions and modify their strengths in real-time. Even the most accomplished humans are presently unable to achieve this feat, if indeed they ever can. Through reinforcement learning and by playing thousands of simulated matches, the AGI manager could devise tactics never conceived and would have a solution to every tactical challenge it faces. It could incorporate a real-time scout mode, assessing thousands of players worldwide and identifying talent more quickly and reliably than any of its competitors. This ensures both future as well as present success. Further, the AGI manager may anticipate injury and identify replacements, perhaps years in advance. The siloing and fragmentation of current within elite sports’ performance departments would be resolved, as it will be all of them. ANI already performs language translation, and the AGI manager could thus communicate fluently to each squad member in their own native language. The above are only examples, the AGI manger will develop innovations of its own not currently feasible to human thinking. The first team to employ an AGI manager would quickly dominate world football. We would likely then see an unprecedented paradigm shift in football, since other teams would inevitably have to follow suit. Soon AGI mangers would be pitted against each other. This will transform the game, perhaps even beyond recognition. As we have seen with the introduction of Video Assistant Referee (VAR), technology changes the character of the game. At heart football is an entertainment business and requires supporter acceptance to flourish. Will the ceiling effects introduced by an AGI manager prove popular with supporters, or will the game lose its character if all action is optimised and so rendered uninteresting? So, it is important at this point to also ask, what could go wrong?
The unintended ’Gift’
Many of the documented risks associated with AGI’s arise because they will seek to achieve their own, self-determined goals in the most efficient way possible [8], and will secure as many supporting resources as feasible to do so. AGI thought leader Nick Bostrom’s thought experiment involves an AGI tasked to maximize the creation of paperclips [3]. Without stopping rules, this system eventually consumes all global resources, in effect destroying the world, in seeking to fulfill its goal of making paper clips. A similar scenario might see the AGI manger transforming the world into its own football academy. Though far-fetched, a more realistic prospect might be an AGI manager who creates large pools of elite players through revolutionary coaching and training techniques. Caring little about player health and wellbeing (if values are not aligned) and with such a large talent pool available, it would instantly discard injured and out of form players as superfluous. Playing careers could diminish from decades to months, weeks, or even one single match. If the AGI manger inexorably pursues its goal of winning football matches, it may do so by employing optimal tactics that leaves football monotonous and uninteresting; and unintentionally destroying the game itself. A major question is whether the now intellectually inferior human players will be able to keep up? Indeed, many issues will arise since human players will be unable to perform as it prescribes. This could be exacerbated by trust failures between players and the AGI manger alongside the inability of a non-human coach to manage the social and psychological dimensions of human team cohesion. Rationally, the AGI manager will then seek to develop its own roboplayers who will outperform their dopey human counterparts. Just as many are concerned that humans will be replaced in multiple forms of work, a major concern for elite sports is that they may no longer involve humans at all. This eventuality will challenge the basic conception of sport itself.
What can be done?
It is critical that work is undertaken now to ensure that advanced ANI and eventually AGI render positive effects in the sporting arena, and that unintended consequences of their introduction are controlled. In relation to the development of AGI systems generally, there is agreement that various forms of restriction are required. These include controls on AGI design, operations, risk management, and controls for any AGI systems use [9]. Failure to develop these appropriate controls before AGI implementation means it may well be impossible to control. Regulations for using advanced ANI, and AGI, in sport are therefore required of GSO’s (e.g., FIFA, World Anti Doping Authority, International Olympic Committee), who may need to start considering appropriate controls now to ensure level playing fields remain. In the same way current controls ensure fair sport, such as banning doping, gambling, and match fixing, the controls around the use of ANI and AGI will be critical. For example, introducing technology expenditure caps, like player salary caps, could ensure that clubs with large budgets cannot simply outspend smaller clubs. A futuristic example may include restrictions on AGI manger-controlled technology directly inserted into human players. We are already seeing numerous wearable devices in sport. It is safe to assume that AGI controlled implantable devices are a logical advancement. Lastly, transparency via open-source AI where information would be shared freely may prevent an AGI manager ‘arms race.’
While this Blog is obviously speculative and focused on medium-term technology advancements, it is critical to initiate discussions now. The fields of computer science and AI are moving rapidly, and we can’t afford to ignore these developments and then try to deal with the consequences when it is already too late. One of the major criticisms associated with AGI development in other domains, is a limited focus on risk, safety, and management [6, 10]. At present, it remains unknown who precisely is in control of advanced technology insertion into sport. In the end, perhaps one of the most unfortunate of unintended consequences, for the sporting public, might be the over-dominance of AGI mangers that see the likes of Pep Guardiola, Carlo Ancelotti, and Luis Enrique reduced to coaching Sunday league football. That is, if their android roboplayers will still let them.
Dr Scott McLean is the Director of Australian based sports consulting company- Leverage Point Consulting. Scott is also an Adjunct Associate Professor at the Centre for Human Factors and Sociotechnical Systems at the University of the Sunshine Coast (UniSC). Scott is recognised as a global leader in systems thinking and complexity in sport. He has worked and conducted research with numerous sporting organisations including English Premier League, the World Anti-Doping Agency, International Olympic Committee, Sport Integrity Australia, the Australian Institute of Sport, Athletics Australia, Cycling Australia, the French Anti-Doping Agency, Sport and Recreation Victoria, the English Institute of Sport, Scottish Rugby Union, the Defence Science & Technology Group, Office of US Naval Research and multiple elite clubs in football, AFL, netball, and Para Sport. As an academic, he has over 100 publications, including two books, numerous peer reviewed journal articles, book chapters, reports, conference articles and abstracts. Please feel free to get in touch with Scott via LinkedIn.
References
1. Barrett, A.M., and S.D. Baum. ‘A Model of Pathways to Artificial Superintelligence Catastrophe for Risk and Decision Analysis’. Journal of Experimental and Theoretical Artificial Intelligence 29 (2017): 397–414.
2. Elstak, I., Salmon, P., & McLean, S. (2024). Artificial intelligence applications in the football codes: A systematic review. Journal of Sports Sciences, 42(13), 1184-1199.
3. Bostrom, N. Superintelligence: Paths, Dangers, Strategies. New York, NY, USA: Oxford University Press Inc, 2014.
4. Müller, V.C., and N. Bostrom. ‘Future Progress in Artificial Intelligence: A Survey of Expert Opinion’. Fundamental Issues of Artificial Intelligence (2016): 555–572.
5. Torres, P. ‘The Possibility and Risks of Artificial General Intelligence’. Bulletin of the Atomic Scientists 75 (2019): 105–108. 2https://www.linkedin.com/in/scott-mclean-0135a322a/
6. McLean, S., Read, G. J., Thompson, J., Baber, C., Stanton, N. A., & Salmon, P. M. (2023). The risks associated with Artificial General Intelligence: A systematic review. Journal of Experimental & Theoretical Artificial Intelligence, 35(5), 649-663.
7. Silver, D., A. Huang, C.J. Maddison, A. Guez, L. Sifre, G. Van Den Driessche, J. Schrittwieser, I. Antonoglou, V. Panneershelvam, and M. Lanctot. ‘Mastering the Game of Go with Deep Neural Networks and Tree Search’. Nature 529 (2016): 484–489.
8. Hancock, P.A. ‘Imposing Limits on Autonomous Systems’. Ergonomics 60 (2017): 284–291.
9. Geortzel and Pitt, ‘Nine Ways to Bias Open-Source Artificial General Intelligence’, 5; Hancock, ‘Imposing Limits On Autonomous Systems’, 289–290.
10. Salmon, P. M., Baber, C., Burns, C., Carden, T., Cooke, N., Cummings, M., … & Stanton, N. A. (2023). Managing the risks of artificial general intelligence: A human factors and ergonomics perspective. Human Factors and Ergonomics in Manufacturing & Service Industries, 33(5), 366 378.
