Blog: Future Sports: AI’s Next Step
AI as Our Next Greatest Sportsmen
He sees an opening on the left wing and immediately punishes them. After rushing down the side, he looks for his teammates in the center and quickly makes the cross in to finish it off! Turn on any sports channel and you’ll hear something similar. Chances are you pictured Ronaldo or another star player running down a fresh grass field. In fact, this describes a play from an artificial intelligence bot in a recent international tournament. Your preconceptions need to be updated as we move from human to digital, to robotic, as AI make their way into our sports as the star players.
Artificial Intelligence to enhance the performance of human players is already pervasive (Forbes, 2019). The next step for AI in sports is introducing AI players. In fact, we currently have AI agents smart enough of mimicking high-level human tactics. They have the potential to revolutionise the sports industry as well as further drive the AI industry along.
The immediate response from many people is that such a world will never come to be — how could we enjoy watching machines? Many claim that playing against traditional AI can often be a repetitive and boring experience. Others can’t imagine any joy from beating their machine opponents. To address this, let’s start by examining why we like traditional sports and then outline how AI will come to meet these demands.
Note: this article will focus on sports that require some level of strategy such as chess, football and rugby; but will not address those which are primarily won through pure physical prowess such as long jump, javelin and weight lifting.
Why Do We Love Sports?
Research in the field of sports-fan psychology is surprisingly limited. Daniel Wann — one of the leading researchers — describes the field as having less than 100 active participants with little grant funding (CJR, 2013). Despite their numbers, they’ve nailed down eight motivations for why people love sport (Wann, 2008):
- Group affiliation: sports provides a common topic for friends to discuss and enjoy
- Family: like group affiliation, this applies particularly to family members
- Economic: sports can act as a tool for betting and earning money
- Escape: sports can be a diversion from any dissatisfactions with regular life
- Entertainment: sports can provide various forms of entertainment leading to pleasure
- Eustress: sports can arouse an excitement through the competition engaged in by others
- Aesthetic: sports can be perceived as beautiful or artistic providing aesthetic pleasure to the viewer
- Self-Esteem: sports can provide people with an increased self-worth/image
It’s evident there are many different motivations for watching sports. In fact, it’s common for people to have 2–3 distinct motivations. This is because sports is a personal experience.
Many of the motivations mentioned above are not unique to traditional sport. For example, getting together with friends and family to bond is about the people, not about the sport. As such, if the conditions were right, a similar variant involving AI could make inroads into the industry.
How Do We Get There
The adoption of AI into the world of sports will be slower than other AI and software applications. Many of the motivations of sports relate to how others around an individual think and behave, so it’s not enough to change a few people, you need to change a huge group of them to be truly effective. Yet we can make distinct steps towards this, as outlined below:
- Competitive AI
- The Rise of Esports
- Mainstream AI
- Advanced Robotics
Let’s explore these steps.
Firstly, AI must be competitive with humans before anything interesting can take off. It turns out they’re already there with some of our most complex games and the methods can likely be applied to our more complex sports too. Some of the key cases:
- Chess — Deep Blue first won in 1997, consistently winning by 2005 (Extreme Tech, 2014)
- Go — AlphaGo consistently winning since 2016 (Wired, 2016)
- StarCraft — AlphaStar beats out a top-tier StarCraft player in 2018, work is being done to expand to the entire game (DeepMind, 2018)
- Dota 2 — OpenAI’s bot defeated a bunch of amateur gamers in 2018 but still lost to pro gamers (The Verge, 2018)
These are all examples of deep learning AI, where complex strategies are not pre-programmed, but learned. Note that there exist many bots that follow pre-set rules that able to beat humans. We will not discuss these techniques as they are not extensible to more complex scenarios which exist in most sports.
The Rise of Esports
Our robotics capabilities are still somewhat limited as seen in various robotic games such as football (HTWK Robots, 2015) so it will still be some time before we can apply AI players to most traditional sports (though Boston Dynamics is getting there quickly (Boston Dynamics, 2017)). Instead, AI is likely to become common in the world of esports.
Esports is quickly becoming a comparable industry to traditional sports. With 30% YoY growth, it will eclipse $1 billion in revenue in 2019 (World Economic Forum, 2018). The largest team in esports, Cloud 9, has a valuation of over $300 million, approximately 7% of the largest sports club, the Dallas Cowboys, at $4.8 billion. In prize pools, esports already exceeds many, including the Golf Masters and Confederations Cup at over $24 million.
The key thing to note is that esports is relatively new. As opposed to traditional sports, which have been around for centuries, esports only began 25 years ago and the most popular game, Dota 2, was released just 10 years ago⁵. This shows how quickly it has grown. Once this continued growth hits a critical mass and breaks out of the current closed demographics (it’s hugely popular in China (South China Morning Post, 2017) and Korea (ISPO, 2019)), it will provide similar family and group affiliation motivation that traditional sports do.
Consider that FIFA now runs an international tournament of esports for their very own sports games. For fans at home, the experience is largely the same, watching the same match on the same television with the same live commentary. Granted, the animation of the current games still has room for improvement, but this is rapidly advancing and the fact that it is generated allows for far greater creativity, for example 3D watching where you can experience being in the play, maybe even in the referees shoes. The world’s most lucrative sport⁶ is already moving into esports, so it won’t be long before others follow.
There are various other reasons esports makes a good first choice such as the ability to better train and improve AI. For a computer game, AI can play millions of games (e.g. 5 million games for AlphaGo) for training as opposed to regulars sports where it must physically play the game to learn and test its performance (even this is being worked on by OpenAI). We’ll discuss this in further detail next.
Right now, if someone asked you to watch two programs compete against each other inside another program, you might think they’re a little weird. This is a reasonable reaction, but one which will slowly change. If there’s one thing humans are not good at it, it’s detecting small changes and that’s exactly how AI will come to be a standard thing to watch. In fact, it has already begun.
“I literally want to sit and watch these matches so I can learn new strategies. People are looking at this stuff and saying, ‘This is something we need to pull into the game.’” — Stephen Merity
There are various competitions between AI which garner decent sized audiences. The following is a list of various games and AI representations on YouTube which have large audiences.
- Rocket League, a popular football-like game has an active bot-community with videos featuring the AI approaching 1 million views (link). The game has around 50 million players.
- A particular channel, Code Bullet (link), focuses on programming various AIs to play in games. It has over 1.3m subscribers with videos earning up to 7m views. For reference, Dallas Cowboys have 90K subscribers while Real Madrid FC has 4.5m subscribers.
- The first stream of the new StarCraft AI had over 2m views (link). As a reference, the finals of the main 2018 StarCraft tournament had 1m views.
- A video featuring the training of a Super Mario AI received over 8 million views (link).
- Robot Wars was a TV programme about fighting robots. While not about AI, the format could readily be run with AI without much loss. The show had a peak viewership of 6m viewers in the UK in the late 1990s.
- Robot football when first publicised shot to 2m views, but since then hasn’t gained much popularity (link). While it is entertaining to watch the little robots stumbling around, it’s not providing the variety and interest of a true sporting event.
- Halo, a first-person shooter, has various AI vs. AI competitions setup by players, these can receive up to 300K views (link).
- Injustice 2 has an AI simulator function that has been quite popular in discussions, videos receiving 100K views (link).
- Fighting games lend themselves well to AIs since they have a highly controlled environment. This Killer Instinct AI video describes their innovative AI and has 20K views (link).
- The Galaxy 11 promotional video, whilst not humans vs AI, shows that people can get the feeling of eustress, escapism and entertainment by watching an animated in-human sport. It has over 7 million views (link).
Overall, this is on the order of 100 million views on YouTube, which is only around 2% of one day of streaming (1 billion hours (Tech Crunch, 2017) vs. 10 mins / video ~20 million hours), however, given the relatively small community this number is not insignificant. As mentioned earlier, this will only continue to grow. Coupling the growth of AI bots with the growth in esports will create massive growth in the genre as a whole. However, this growth won’t be sustainable unless the AI stays interesting.
Once watching AI becomes familiar, more depth will need to provided to keep viewers involved. In order to achieve this, it’s critical that we diversify our AI. People don’t want to watch the same thing over and over again. As previously mentioned, one of the motivators for watching sport is entertainment which comes from the ‘chance factor’ of not knowing who will walk out victorious on any given day. In order to achieve this, the agents must be capable of making various high-level, non-straightforward plays (which we’ve already seen with Dota 2 and Go, to name a few).
In fact, there’s a common misconception that watching AI is a boring experience as they unintelligently copy humans or follow pre-described rulesets. Certainly that was true of machines of the past, but for many years now we’ve had AI that can act in creative and all-together astonishing ways.
One of the most interesting parts about Google’s AlphaGo was its creativity and ways that it played that game that were unexpected of real humans (Forbes, 2019). Along the same line, in the world of chess, when human players make moves that vary from the standard procedure, referees start to suspect players of using artificial intelligent systems as assistants. To make that 100% clear, in the game of chess, creativity is no longer the mark of a human, but that of a machine. It’s the same in Go and as time passes by it will be the same in other sports.
During the AlphaStar training, it was observed that the bots adopted various good strategies. One might expect that the bots followed a specific strategy and got better and better at it in time. In fact, the bots could be clumped into various groups, each group having a varying way to play the game, e.g. aggressive start, focus on a certain type of units, etc. In a way, each bot had its own ‘personality’. These personalities, with varied play-styles will keep sport interesting and entertaining.
Following the creation of various AI personalities would naturally be personification. One of the motivations for watching sports is Self-Esteem, as we try to project ourselves onto the people we watch. Seeing the various styles, we will choose ones we like and begin to unconsciously project ourselves onto them. This will greatly improve the standing of AI players in society and will lead to more sponsorship and marketing opportunities, in turn fuelling more growth.
Once AI agents have become a regular part of our sporting experience, the advancing robotics will catch up, allowing them to play all of the games we usually play, with us. Footballers will be able to practice against full teams of AI bots that are set to challenge them and help them grow. They’ll also be able to compete in human-robot leagues.
While human biology is relatively fixed, robotics will continue to advance. This means that the sports can continue to evolve too. Imagine a game of football played at double the pace with a magnetic ball and speeds matching that of tennis? Sounds pretty exciting to me. A recent Hollywood movie “Alita Battle Angel” featured a similar futuristic game called Motorball.
Finally, new games can be created that only AI can master. As previously mentioned, Escape and Aesthetic are two of the motivators for sports fans. It may be in-human, but watching an AI empowered machine conquer and handle complex games will create a feeling of escape like we’ve never experienced before.
What Does it Mean for the Future
If the above story comes to be, there would naturally be significant impacts on the sporting industry.
- Better AI: the AI revolution is the next major revolution that human society will undergo. The rate of approach to this revolution will be propelled by the economic incentives provided by sporting competition. Currently there a few major well funded organisations such as those within Google and Elon Musk’s OpenAI but the more competition, the faster the growth we can see.
- On-Demand: since you don’t need to worry about pesky biological needs, games can be played and watched as and when it’s required. Not only this, fans can pick their own teams and see exactly how it would have played out. This takes fantasy football to a whole new level. Similarly, the world of sports betting and the economic motivation discussed will be heavily influenced.
- VR: filming a live match between humans for fully-immersive VR will always be difficult, however, with digital sports, it will be a matter of simulation allowing for a far more enhanced viewer experience.
- Collaboration: right now a sports team has 11 players who have all the glory. Behind the scenes there are various members including physician, psychologist, dietician who help the player perform but the match day performance is down to them. For AI agents, it’s more common that many different members work on parts responsible for making it a success. Members will be able to collaborate meaning becoming a ‘top athlete’ will be a dream far more achievable for the average person.
- Economic: instead of funds going to sportsman, it will be redirected to programmers and their organisations. Good AI is made from teams of people, with many terrific players in the field, so it may lead to a more diversified top ranking and spread out funds.
- No retirees: since players will only get better in time, you’ll be able to follow your players throughout time. Of course, the designers or ‘coaches’ will still have their limitations, but personified bots can stay around.
- Decreasing Costs: there will be lowered costs since games played with cheaper components or online. This can counter the trend of lessening attendance that is happening across the sporting industry (Sports Management, 2018).
Sports organisations and related companies should start preparing for the changes before they’re too late. For the rest of us, likely not much will change. We can’t hope to imitate Cristiano Ronaldo’s beautiful strikes or Federer’s impossible serves and I won’t be able to match the feats of our robotic future sportsman. If nothing else, it will be interesting to see how it evolves. So for now, I’ll sit back, pick a side and enjoy the game with my friends.
If you enjoyed reading this, please don’t forget to leave a clap or two, or three… 👏👏👏🔄
If you’d like to read about how AI is used in the food delivery business then check out this article (WARNING: it’s technical):