1
The AI advantage: How the next two Rugby World Cups will be won
By Ben Smith
One of the early experiments involving machine learning was done by the Icahn School of Medicine in New York to predict cancer in patients.
ADVERTISEMENT
Fed with the data of 700,000 patients, the model began spotting new patterns in the data that to the human eye, weren’t visible or didn’t make any sense. The AI model proved to be very good at finding patients with early-stage diseases. As a side, it also figured out warning signs of other disorders like schizophrenia.
The conundrum was researchers running the project had no idea how it was doing it, and still don’t.
As with the case with most AI models, the more data you have to train it, the better the results you get.
They are predictive machines, evolving towards superhuman-level intelligence. The applications are going to have wide use cases but in the realm of professional sport, obtaining the AI advantage is going to be a necessity over the next decade.
You don’t even need to explain the rules of the game. We’ve learnt that the AI models can learn the rules just by watching. Ingest years and years of game footage, it will understand the sport at a level greater than any human could.
You can start to imagine the impact this is going to have. And if you don’t have it, it will be used against you.
ADVERTISEMENT
An AI model trained on enough games of professional rugby will find every weakness or vulnerability in every single player on the field. Just like the cancer research team found, it will soon find patterns that are unrecognisable to the human eye.
If it watches every game that a professional player has played over their lifetime, it will take into account every single read in defence that they have made, what they do when presented with this picture or that picture, what players they struggle to tackle, what technique they use. Every single decision.
All of that information will be calculated in seconds and result in the AI planning and strategising on how to take advantage.
Armed with that information, it will come up with the perfect play to expose those players. Going further, it will come up with the perfect game plan to win against any combination of 23 players.
ADVERTISEMENT
If there is a match-up where one team theoretically loses 99 times out of 100, the model will be able to find the formula for the one outcome they can win and show them how to do it.
Upload every game possible from the team of an opposition coach and the AI will figure out every tactic they’ve ever used, every flaw in their plans, and predict what they will do next and the best way to play.
The job of the analyst is going to become rather easy, but the knowledge obtained at such speed will lead to incredible outcomes in game strategy and play.
Teams will have to continually come up with new plans, which will be driven by AI. Coaches who can’t or won’t evolve will get weeded out.
Even in the realm of managing your own team, the technology will be invaluable.
It will be able to detect the slightest changes in a player’s running style, perhaps indicating that player isn’t 100 per cent fit and has a problem.
If the model has all that player’s training data and has been trained on hours and hours of footage of that player’s movement, it will start predicting with scary accuracy whether an injury is likely to occur.
To be clear, the AI is never going to be able to win games of rugby, which are always decided by humans on the actual field. That is sport and won’t change.
The physical attributes still matter greatly, the skill, strength, size, power and the conditioning of the players. No AI can overcome a disproportionate mismatch in this area.
But between two teams that are evenly matched, the one that has superhuman level intelligence feeding them information about the battle at hand is going to improve their chances of victory greatly.
And between the top four nations right now, Ireland, France, South Africa, and New Zealand, where very little separates them, that is going to matter.
Professional sport is always after one per cent improvements, this is going to add far more than one per cent.
Right now it takes hundreds of millions of dollars to build these models. And they lie in the hands of very few, the tech giants who are building major data centres and feeding them as many data points as they can get their hands on.
But once model access is obtainable, professional sports teams will start building their own AI models for competitive use.
If one of the big four rugby nations were able to get a hold of one right now they would increase their chances of winning the 2027 Rugby World Cup greatly. By 2031 you would think this will be widespread.
Quite quickly the AI advantage is going to be a necessity as teams that adopt AI will gain an edge that is far superior to those that don’t.
That is the AI advantage.
AI is only as good as the information put in, the nuances of the sport, what you see out the corner of the eye, how you sum up in a split second the situation, yes the AI is a tool but will not help win games, more likely contribute to a loss, Rugby Players are not robots, all AI can do if offer a solution not the solution. AI will effect many sports, help train better golfers etc.
What’ll happen when the AI models of the future go back in time and try to destroy the AI models of the past standing in their way of certain victory?
If rugby wants to remain interesting in the AI era then it will need to work on changing the rules.
AI will reduce the tactical advantage of smart game plans, will neutralize primary attacking weapons, and will move rugby from a being a game of inches to a game of millimetres. It will be about sheer athleticism and technique,about avoiding mistakes, and about referees. Many fans will find that boring.
The answer is to add creative degrees of freedom to the game. The 50-22 is an example. But we can have fun inventing others, like the right to add more players for X minutes per game, or the equivalent of the 2-point conversion in American football, the ability to call a 12-player scrum, etc. Not saying these are great ideas, but making the point that the more of these alternatives you allow, the less AI will be able to lock down high-probability strategies. This is not because AI does not have the compute power, but because it has more choices and has less data, or less-specific data.
That will take time and debate, but big, positive and immediate impact could be in the area of ref/TMO assistance. The technology is easily good enough today to detect forward passes, not-straight lineouts, offside at breakdown/scrum/lineout, obstruction, early/late tackles, and a lot of other things. WR should be ultra aggressive in doing this, as it will really help in an area in which the game is really struggling.
In the long run there needs to be substantial creativity applied to the rules. Without that AI (along with all of the pro innovations) will turn rugby into a bash fest.
I couldn’t agree more on the ref/tmo assistance. Software that can measure better than the human eye for those types of calls would allow the ref to focus on the more challenging stuff.
Like listening out for racial slurs from Bongi.
You hit the nail on the head right at the end. Rugby needs more IT as in the Hawkeye tennis and cricket have had for years (and soccer has failed miserably with its VAR up till now) rather than AI. Coated balls that can be detected under pile ups in the goal area and so on. Unfortunately it seems too many people believe AI is the answer for way too many things, when in many ways it's just an extension of the old ‘garbage in, garbage out’ from the early computer days. Mind you, I wouldn't put it past the genius Rassie to prove me wrong!
I think maybe the opposite, where the NFL looses that aspect of complexity, and becomes what most people only see anyway, were rugby could be better served simplifying it’s game? Having it a player driven mess basically. AI will be too powerful, there will need to be new trickery towards complexity.
Making the game more difficult for, like adding skill bonuses for certain kicking, or advantages, eee not so much power play time things, anything that demands more randomness from the players, might be whats needed to keep the game interesting for sure. I have the same fear of it taking too much away from the game, selection, tactics, live options etc.
Interesting article with one glaring mistake. This sentence: “And between the top four nations right now, Ireland, France, South Africa, and New Zealand…” should read: And between the top four nations right now, South Africa, Ireland, New Zealand and France…”.
Get it right wistful thinkers, its not that hard.
Does the AI take into account refs? hahaha Seriously why not have two on field refs to avoid bias?
there are already 4 match officials per match
but even with 1, bias wouldn’t be an issue
Watching the SA series no AI will motivate players like a Human can cause no matter your IP if you lack the hype to be super human or the level to go to the deep dark places you simply can’t win big games. France Ireland All Blacks and SA will surely get this AI but the end of the day it's luck and believe that matters
No way SA can afford it or have the talent and knowledge to make their own.
Rather AI than the disastrous and disappointing human errors of the last RWC final.
Dont know what you are talking about.
The AI will find the 1 scenario out of 100 to win. But then so will the other teams AI do the same (to prevent it). So then it will just be 99 losses and maybe a draw?
All of this is fine and dandy, but assumes that the players and coaches will be able to move flesh and bone around the training pitch and on game day to implement against whatever it was the AI scenarios predicted.
I am not so sure it will have that a big effect in the next 10 years. To have a a.i giving you extra info doesn't mean you have a team that can implement the plan. It will take years for humans to adato be able to use the a.i's data
At some point the calculator was invented. Game changer. Except everyone has access to it. Back to square one.
Don’t know who’s gonna win the next one but I’ll make a prediction and say that England will be knocked out at the pool stages.
Why’s that? Since 1995, however far Australia go in a World Cup, England do the same in the following World Cup, with 2015 being the only year to buck the trend.
Let me explain:
1995: Australia out in the knock out stages
1999: England out in the knock out stages. Australia win the World Cup
2003: England win the World Cup. Australia runners up
2007: England runners up. Australia out in the knock out stages
2011: England out in knock out stages. Australia out in knock out stages
2015: (the only to buck the trend) England out in pool stage. Australia runners up.
2019: England runners up. Australia out in knock out stages
2023: England out in knock out stages. Australia out in pool stages.
2027?
Start with Australia wins in 2027 and work backwards to find how they did it in the past pick the top 3 crucial truths build the model forward. When we win, start again. I don't know why we don't build for back to back wins. Storm and Crusaders do it time and time again. Difference is and it should be a plus we have 4 years to build one from the other so now we test and experiment scenarios.
South Africa are going to win the next one.
And on the off chance they don’t - New Zealand.
My predictive model has a 70% success rate. Only 1991, 1999 and 2003 bucked this trend.
It remains to be seen how AI will impact sport in general, it certainly will have a big effect like you say. Humans in the system will have to get used to choosing what to use and what not to and how to beat the odds. Its a human sport played by humans on the field and so execution matters and error rates may come to be the thing that determines matches as opposed to strategy which will be guided by AI. Nice article here
Maybe when Elon Musk has implanted processors into his team's brains.
AI models are really just larger and less transparent variants of the statistical models that have been in use since Moneyball was invented. And a big difference between the Icahn centre’s results and AI today is that ChatGPT-like Large Language Models can explain (to some degree) how they reached their conclusions.
In terms of what impact they will have, I suspect it will have two primary impacts:
1) It will place a premium on coaching creativity
2) It will lead to more selections that baffle fans and pundits.
Analysts will be able to run the models both ways: they will see their own team’s and players’ weaknesses and strengths as well as the opposition’s. So they will have a good idea at what the other team will be targeting and the decisive difference may well be which coaches are smart enough to think of a gameplan that the other side didn’t identify and prepare for.
For players, it places a premium on three key things:
1) Having a relatively complete game with no major weaknesses (or the dedication to work on eliminating them)
2) Having the tactical flexibility to play a different game every week
3) Having a point of difference that is so compelling that there isn’t a defence for it.
(3) is relatively rare even among pro players. There have been only a handful of players over the years where you knew what they were going to do and the problem was stopping it - Lomu would be the classic example. And even when someone does have that, it’s hard to sustain. Billy Vunipola in his prime was very hard to stop, but fell away quite badly when the toll on his body began to accumulate.
So coaches will look for (1) - a lack of exploitable weaknesses - and (2) - the ability to exploit others’ weaknesses - ahead of hoping for (3), at least for the majority of the pack. Which is likely to mean that, as with the original Moneyball, competent, unshowy players who do the stuff that wins matches will win out over outrageous talents who can’t adapt to cover their own weaknesses.
Which will leave a lot of people on the sidelines sputtering over the non-inclusion of players whose highlights reels are spectacular, but whose lowlight reels have been uncovered by AI… at least until the point where every fan has access to a sporting analysis AI.
That will be some cheap AI where a team still requires “Analysts”
Your players 1 and 3 contradict each other.
Heavily disagree with your Moneyball context/understanding (not that I have even watched the movie). Moneyball is there because of humans. Humans can’t conceptualise the randomness of players into effectiveness. I’d see AI as being able to use unique player types to an advantage. Not wanting to homogenize them.
Your AI will tell you how to beat my team. My AI will tell me how your AI will tell you how to beat my team and how we should adjust. Both our teams will then have to interpret that. This will be happening in both directions, with layers of second guessing. In conclusion, it will be the same as now.
There goes our Thursday team naming.
AI will have a much smaller impact than is being predicted. In the next 10 years or so all it will give us is some very dull automated articles. Maybe in 20 or 30 it will do what this author describes.
It will be good to get AI doing the player ratings. We should have a much better collection of stats to get engaged in.
It will be a constant game of cat and mouse.
I think AI would write better articles.
I wonder if we fed all of Ben’s articles into an AI model, if it would also come to the conclusion that he is a prick. I suppose you don’t really need the AI advantage to come to that conclusion…
Hahahaha…..
Brilliant!
Comment of the year…