Chris Kamara, AI, and Why Football Refuses to Be Predicted

By Richard Faint · 5 July 2026 · 8 min read

England face Mexico in the last 16 of the World Cup at 1am on Monday morning, which gives the nation several hours to perform its traditional pre-match ritual of constructing extremely confident predictions from deeply incomplete information. In the build-up to kick-off, television analysts will exhaustively debate formations, former players will identify the isolated individual battles that will apparently decide the entire tie, and sophisticated statistical models will churn through data to calculate each side’s exact probability of progressing. Meanwhile, social media will confidently announce both the starting lineup and the manager’s tactical failures long before either has actually been confirmed Then, 22 people will run onto a pitch, the referee will blow the whistle, and reality will aggressively introduce itself to the model.

TL;DR:

Football is an open, reflexive system: the people being forecast change their behaviour in response to predictions. A 57% probability is not a promise, and more data cannot make a system predictable when its participants, context and incentives keep changing.

Smoggie Queens and Analysis

For anyone looking for a more honest approach to the absolute chaos of sports forecasting, a recent episode of the comedy Smoggie Queens offers the perfect alternative. Smoggie Queens is not a well known show, but it is a bit of a sleeper hit, I have also got a personal connection as it is filmed in my brothers street.

“A Smoggie Match,” follows Mam as she manages a team in a local charity match while her friends Dickie and Lucinda compete to become the town’s most convincing football WAGs. This is a far cry from the Champions League final, as the event is a benefit match, involving a mix of local semi-professionals and a surreal lineup of celebrity lookalikes. The roster of lookalikes reportedly includes versions of Cher, Prince Harry, Liza Minnelli, Elton John, and Stevie Wonder, proving that the show deeply understands that football is almost always improved by abandoning restraint as early as possible. Alongside these eccentric lookalikes, the episode features media figures Steph McGovern, Jeff Stelling, and Chris Kamara all appearing as themselves. During the festivities, Stelling turns to his former Soccer Saturday colleague to ask the standard, inevitable question of whether he has a prediction for the game. Kamara does not disappoint, offering a response that completely bypasses the match to announce that he thinks AI will take over the planet and destroy the human race.

While it is certainly not a conventional scoreline forecast, Kammy’s apocalyptic pivot is considerably more memorable than saying the match will be tight and could go either way. It also captures something profoundly important about our current cultural relationship with prediction. We consistently ask increasingly sophisticated technological systems to tell us exactly what will happen in the future, but the answers we get back often reveal far more about our desperate appetite for certainty than they do about actual upcoming events. We want the comfort of a fixed outcome, turning to data to shield ourselves from the anxiety of the unknown.

Before Mexico and England kick off, there is certainly plenty of historical evidence to feed into a mathematical forecast. Mexico have won all four of their matches at this World Cup and stand on the verge of becoming only the second side to begin a tournament with five consecutive clean sheets. They also have the immense psychological advantage of playing at home in Mexico City, a historic fortress where they have never lost a World Cup match. England, on the other hand, are attempting to reach at least the quarter-finals for a third successive tournament, but they must do so more than 2,200 metres above sea level. While England have had limited time to acclimatise, Mexico are intimately familiar with both the grueling physical conditions and the intense atmosphere of the stadium.

These are all highly relevant variables, but absolutely none of them can tell us what will actually happen on the night. That distinction matters because a mathematical forecast is not the same thing as actual knowledge. A model can estimate the likelihood of an outcome by examining historical results, player performance data, injuries, tactics, location, rest, altitude, and hundreds of other metrics, but the final output it produces is merely a structured expression of uncertainty. Unfortunately, modern sports audiences do not generally ask analysts to provide a nuanced probability distribution across the range of plausible match outcomes alongside their core assumptions and confidence intervals. They simply ask who is going to win. The systemic uncertainty is completely stripped away by the media because absolute certainty is easier to communicate, easier to publish, and much easier to argue about on television.

Artificial intelligence has drastically intensified this collective craving for certainty. If you feed an AI system enough football data, it can easily identify complex patterns that would be incredibly difficult for a human eye to detect. It can compare thousands of historical matches, examine how specific teams perform against different tactical formations, and dynamically update probabilities as new information becomes available. What it absolutely cannot do, however, is turn an open system into a closed one. A football match is shaped by deflections, bizarre refereeing decisions, sudden injuries, shifting weather, fatigue, crowd behavior, and individual choices made under immense psychological pressure. It is defined by players actively responding to the match as it develops and managers completely altering their tactics because of what they have just observed from the dugout.

Furthermore, the prediction itself can actively influence the very system it is trying to forecast. A team described by data models as overwhelming favorites may subconsciously become cautious and protective, while an underdog may decide to take massive risks simply because conventional tactics offer so little chance of success. Managers read the same analysis, study the same data, and make late decisions explicitly intended to invalidate the opposition’s expectations. This predictability problem is not a matter of missing data that a better algorithm could solve, but rather the reality of a dynamic system containing intelligent actors who constantly adapt to one another. The better the participants become at reading the system, the more the system itself changes based on those very readings.

There is nothing inherently wrong with football forecasting, but problems begin when the presentation deliberately disguises the uncertainty baked inside it. A model might conclude that England have a 57% chance of progressing, which does not mean it has predicted an England victory. It simply means that, under its current assumptions, Mexico would still progress in approximately 43 out of 100 comparable situations. Yet a sensationalized headline will almost always convert that delicate math into a definitive statement declaring that AI predicts England will beat Mexico. Through this translation, the probability becomes a promise, the underlying assumptions disappear, and the limits of the data are forgotten. When the opposite result occurs, the model is unfairly declared either entirely useless or astonishingly wrong, even though neither conclusion is actually justified by the laws of probability.

A useful prediction should tell us what outcome is considered most likely, how confident the model is, which variables have the greatest influence, what assumptions have been made, and what specific events could rapidly invalidate the forecast. That approach is admittedly less dramatic than revealing a single, clean scoreline, but it is considerably more honest to the nature of reality. There is a deeper issue here, which is that we often talk about uncertainty as if it were a technical defect that better technology will eventually eliminate. In reality, uncertainty is the entire product. People do not stay awake until nearly 3am because they want to observe the inevitable execution of a statistically superior plan. They watch because the result remains entirely unresolved and because, however convincing the pre-game analysis appears, something entirely unexpected might happen.

Mexico’s incredible home record matters precisely because England could break it, just as England’s tournament history matters because they could either repeat it or completely escape it. The data model matters because the match might prove it entirely wrong. A completely predictable football match would be operationally efficient but emotionally pointless. This is the exact truth that Smoggie Queens gets right. Its charity match is nominally about football, but the event quickly becomes about performance, identity, rivalry, friendship, and people making the occasion mean whatever they need it to mean. The celebrity players are not even real celebrities, but lookalikes participating in a match organised around a fictional charity with an extremely specific medical cause. It is beautifully absurd, but it is recognisably football.

Ultimately, the match creates a temporary system into which every single participant brings their own personal objective. Mam wants to manage, Dickie and Lucinda want to become famous WAGs, Stewart wants to be the team mascot, and the spectators simply want to be entertained. Kammy, when asked for a prediction, apparently wants to warn humanity about the looming threat of artificial intelligence. No single objective fully explains what the event is actually for, and that is exactly why it mirrors the real sport so well.

Will AI take over the planet and destroy the human race? It probably will not happen before full-time on Monday morning. Will England beat Mexico? The available evidence can help us establish probabilities, but it cannot give us certainty. England have undeniable quality, experience, and a strong recent tournament record, while Mexico have superb form, the altitude advantage, a formidable home record, and an entire stadium cheering behind them. The correct forecast is not that anything can happen, as some outcomes are clearly more plausible than others, but rather that plausible does not mean predetermined. Chris Kamara’s prediction works so well because it leaps from the narrow uncertainty of a football match to the largest possible uncertainty facing humanity, completely refusing to provide the tidy answer the question expects. That may well be the most responsible prediction anyone makes all evening, although should an AI system confidently predict a 2–1 England victory before deciding to destroy civilisation, we will at least know exactly where it got the idea.

Try it interactively. Open the open-systems prediction model on its own page → — full-width, with a walkthrough of what each control does.

Back to Articles