When Artificial Intelligence Sits in the Coach's Chair: Vietnamese Football and the Question of Dependence

Late in the afternoon at the training centre of a club fighting relegation in the V-League, the lights in the analysis room were still on long after the players had left the pitch. A young data analyst pushed his screen toward the head coach, showing an injury-risk scoreboard for the eleven names set to start at the weekend. He spoke quickly, confidently: the model had finished its work, the star striker carried the second-highest risk score in the squad, and it would be best to bench him. The coach looked at the screen, then looked out at the pitch, where that same player was still taking shots alone under the yellow floodlights. He said nothing for almost a minute.

Around the same time, half a world away, a research team at the University of California published a result that set the technology world talking: just about ten minutes of using ChatGPT could temporarily weaken a person's ability to solve complex problems unaided. Across three controlled experiments, the group that used AI scored 57 percent on hard problems once the tool was taken away, while the control group scored 73 percent. A finding that seems to have nothing to do with a football. But sitting in that analysis room, watching a coach hesitate in front of a scoreboard, I understood that both stories are about the same thing: what happens to human judgment when we hand part of our thinking over to machines.

I have spent more than a decade reading players' bodies. In 2026, when I first entered a major club's medical room to film, I learned something no classroom ever taught me: injury risk is not where you think it is. It does not live on a clean X-ray, or in the line that reads "no structural damage." It lives in the gap between what a machine can measure and what a player will not say. Years later, as Vietnamese clubs began buying analytics software, hiring data staff and, more recently, experimenting with artificial-intelligence tools, I realised that gap is widening in a new way.

The University of California study is not a story about bad technology. It is a story about habit. When a tool answers too fast, too fluently, people tend to stop asking questions. And in football, where every decision about a body can determine a human being's career, that habit costs more than anywhere else.

The frightening thing is not that a machine answers wrongly, but that a person forgets how to answer on their own. That is the line I want to keep in mind throughout this piece.

Context: When the technical bench gains a chair for an algorithm

Vietnamese football entered the data era roughly a decade later than Europe, but the catch-up has been hard to track. Around 2026, most V-League teams still recorded fitness by hand and by the team doctor's feel. By 2026-2026, a few well-funded clubs began hiring data analysts, installing GPS tracking systems and buying statistics packages. After 2026, as large language models became popular, a new wave arrived: AI tools were pitched as the answer to everything, from scouting young players to predicting injuries.

I have sat in many such meetings. People project beautiful charts, forecast curves and probabilities calculated to two decimal places. The mood in the room usually carries a touch of joy, the joy of someone who has just found a magic wand. And I understand that feeling. For years, Vietnamese football had to decide by intuition because it lacked data. When data arrives, the sense of being supported is real.

But there is a paradox few mention in those meetings. The more data there is, the less people argue. Once the numbers have been neatly arranged by a computer, it becomes psychologically harder to push back. Who dares to say the scoreboard is wrong when it was produced by a system nobody in the room fully understands? Who dares to question how the model was built, when the person who built it left long ago?

Based on my experience watching V-League matches across many seasons, I see a repeating pattern. A club adopts a new analytics tool, enjoys good results for a few months, then gradually delegates more decisions to the tool. When results decline, nobody remembers how to decide without it. The trap is not using AI. The trap is forgetting that you once could do without it.

The University of California study lands exactly on this point, even though it was written for an entirely different setting. The team recruited 1,222 online participants and split them into different groups. One group used ChatGPT to solve hard problems; another solved alone. Then the tool was taken away from everyone, and all of them faced new problems without help. The result: the group that had used AI scored only 57 percent on hard problems, while the control group scored 73 percent. That sixteen-point gap is not a rounding error. It is the trace of something being worn away.

Three controlled experiments, with sample sizes of 354, 667 and 201 people respectively, showed the effect repeating. When the tool disappeared, some users did not merely do worse. They began making mistakes, or simply abandoned the problems. They had forgotten what it feels like to wrestle with a difficult problem alone. And that is exactly what worries me as I look at the technical benches of Vietnamese clubs eagerly embracing new tools.

Core analysis: Three layers of dependence in the medical room and the analysis room

To understand what the study says to football, we must see that it is not about ChatGPT as a product. It is about a psychological mechanism: when we let an external system perform the reasoning, that cognitive muscle weakens if it is not trained. Vietnamese football is building three layers of dependence that the same mechanism could erode, at three different levels.

The first layer sits in injury prediction, and this is where I worry most. In recent years, some clubs and youth academies have begun using models to calculate a "risk score" for each player, based on training load, distance covered, fixture congestion and injury history. I myself once built such a model in 2026, with five indices: muscle endurance, pain level, minutes played, training volume and psychological state. That model once helped me predict a re-injury case with high probability, and the analysis was widely shared by a physiotherapist.

But I learned something from my own model. A risk score is a number, and a number does not know how many nights a player slept, does not know he just had a newborn, does not know he is hiding an ache because he fears losing his starting place. If the person reading the scoreboard forgets all that, the model becomes an excuse to stop listening. And when people stop listening, that cognitive muscle begins to shrink.

I once witnessed a painful case in 2026, early in my career. A young defender was rehabilitating from an anterior cruciate ligament tear, but his progress was being pushed abnormally fast. I quietly recorded the numbers: the player had reached only 78 percent quadriceps strength, yet the coaching staff still listed him for the squad. He re-injured just twelve minutes after coming on and was out for another four months. There was no AI in the medical room then. But the mistake was identical in type to what a poorly trained model can produce: a number read by nobody who asked a question.

The second layer sits in scouting and the evaluation of young players. Academies across Vietnam, from long-established talent factories to the newer academies of high-achieving clubs, are all looking to use data to filter. The idea sounds sensible: a computer can watch thousands of hours of footage and surface patterns the human eye misses. But there is a subtle trap. Such models are often trained on data from major leagues, where the style of play, the physical profile and even the decision-making differ sharply from Vietnamese football. When a young coach offers an opinion that contradicts the model, he is often seen as lacking evidence. Gradually, people stop offering contrary opinions.

I believe in data, but data can also lie if we do not ask the right question. A scouting model may rate a player highly because he runs a lot, without knowing that in his league, running a lot is a consequence of a weak team chasing the ball. If the analyst does not ask why, the model turns a weakness into a strength on the scoreboard. The cost is a scholarship awarded in the wrong place, a talent left behind.

The third layer sits in tactical preparation before a match. This is where AI tools are marketed most aggressively. A coach can ask a system to analyse hundreds of an opponent's matches and summarise their attacking patterns. That work once took a team of assistants weeks; now it can be done in hours. The efficiency is obvious. But that efficiency also means the coach no longer has to watch the footage himself, no longer has to wrestle with the small details the model considers unimportant.

In football, the details a model treats as noise are often what decides. The way an opposing defender plants his foot before being beaten, the way a midfielder breathes before a free kick, the way a goalkeeper glances before a goal kick. Those things are not in the data. But a coach who has watched enough will sense them. If he no longer watches enough because a machine summarises for him, that sensitivity wears away.

Contrarian angle: The problem is not the tool, but the person who dares not argue with it

The easiest thing is to read that study and conclude that AI is harming football. I think that conclusion is both true and useless, because it does not show where to fix anything. Vietnamese football will not stop using tools. Clubs will keep buying software, hiring analysts and testing AI at every stage. Calling for a return to the notebook era is meaningless and even harmful, because it makes serious people look old-fashioned.

The real problem lies elsewhere. The University of California study does not measure dependence on a tool. It measures what happens when the tool disappears. In football, the tool never disappears completely, but it disappears at the most important moments. When a match reaches the eightieth minute and a player clutches his calf, no model has time to tell you what to do. When the referee points to the penalty spot in a tense derby, no system makes the decision for you. Those are the minutes when people must stand on their own two feet, and that is exactly when the ability to solve problems unaided is tested.

So the right question is not "should we use AI in football." The right question is: is the club training people to stand firm when AI is absent? And in most cases I know in Vietnam, the answer is no.

One detail in the study kept me thinking. When the tool was taken away, it was not only the AI group's scores that fell. Some participants began making mistakes, or simply gave up. They did not just perform worse; they lost the will to continue. This is a sign of something deeper than skill: belief in one's own ability to solve. When that belief goes, people stop trying.

In football, that belief has its own name. It is what keeps a defender from giving up after being beaten twice in the first half. It is what makes a midfielder still ask for the ball in the ninetieth minute, even after a poor match. If a club teaches its players that every answer lies in the machine, then when the machine can no longer answer, the player will stand still. And I have seen players stand still. It is the saddest image on a football pitch.

Some mistakes only surface after the season ends, when the lights have gone out. Dependence on tools is one such mistake. It does not lose you a match. It loses you slowly across seasons, until you forget where you were once strong. By then, no one remembers to trace the cause.

Protective section: What can be done now, and what I have learned

I am not writing this to scare anyone. I am writing because I believe Vietnamese football can benefit from new tools without losing human judgment, if it is willing to do a few concrete and inexpensive things.

The first is to keep one person with the power to say "no" to the model. Not a technology sceptic, but someone informed enough to challenge. Every analysis room should have one person tasked with questioning the scoreboard itself: where does this data come from, which league was the model trained on, what has been left out. The role seems small, but it keeps the whole room from falling asleep.

The second is to maintain a "no-tools session." For players, that might be a video-analysis session done entirely by human eye, without a summarising model. For team doctors, it is a fitness assessment based only on observation and listening. The aim is not to abandon tools, but to keep that cognitive muscle from shrinking. The crack is not on the X-ray; it is in how we listen to the body. And that way of listening must be trained daily, like an exercise.

The third is transparency about limits. Every model should be clearly labelled: what it can predict, what it cannot, what its error margin is. At many clubs I have visited, the data analyst knows the model's limits, but the coach and the players do not. That information gap creates an excessive reverence for numbers. When people know where a model tends to fail, they fear it less and depend on it less.

I learned this from the empty 2026 season. When the pandemic left stadiums without crowds and forced the fixture list to compress, I applied my risk model to predict that muscle injuries would rise sharply. I recommended that a club bench its star striker for an important derby. Fans pushed back hard, calling me an excessive pessimist. In that very match, two other players suffered muscle injuries. The striker, after resting, scored four goals in the next five games.

I tell that story not to boast that my model was right. I tell it because what I did right was not in the model. It was that I still read the match myself, still went to the ground, still asked questions. The model was only one of many sources, and I kept the right to doubt it. Had I delegated everything to the model, I might not have dared make that recommendation, for fear of being wrong. Responsibility does not need a grandstand; it needs one person keeping discipline every morning. That discipline, for me, is the discipline of thinking for myself.

Reflection: What is worth keeping when everything can be asked of a machine

The University of California study ends with a note many hurried articles skip. The effect observed was temporary, not permanent. The researchers do not recommend banning AI. They point out that the tool is not harmful in all cases, and that what matters is distinguishing between using AI as learning support and letting it perform the entire reasoning process.

I think Vietnamese football should read that note very slowly. Because this is precisely the line we stand on. We can use tools to learn faster, see more, and make fewer foolish mistakes. Or we can let tools think for us, and gradually forget how to stand on the pitch ourselves.

The difference between those two paths is not in the software. It is in people. It is in the coach who dares to look at the pitch instead of only the screen. It is in the team doctor who dares to trust his instinct when the scoreboard says otherwise. It is in the player who dares to ask why he ran into the wrong space, instead of waiting for a machine to point it out.

A player whose leg is not broken can still be breaking from within. The same is true of a club. A team that loses no matches can still be weakening from within, if it has stopped thinking for itself. And that weakening does not show in the table. It only shows on some night, when the tool can no longer answer, and no one in the room knows what to do.

Viewers see the goal. I see that knee three months later. And now, looking at brightly lit analysis rooms, I also see three seasons ahead of a football culture that has forgotten how to decide for itself. I believe in data. I just do not believe in handing everything over to it. Keeping that distance, keeping the habit of asking questions, is the only way technology can serve Vietnamese football rather than replace it. The final question is not for the models, but for us: when the tablet screen goes dark, do you still trust your own judgment?