Football games look astonishing now. Individual blades of grass, accurate player likenesses, ball physics modelled from real aerodynamic data, animation systems with thousands of captured movements. By any visual standard the problem is solved.

And matches still don't feel like football. The reason isn't rendering or physics. It's that simulating a football match requires simulating collective decision-making, and that's a genuinely unsolved problem.

The controllable-player problem

Start with the structural issue that shapes everything else. In a football match there are twenty-two decision-makers. In a football game there's one human controlling one player at a time, with ten teammates being driven by AI.

This creates an immediate asymmetry that has no real-world equivalent. The human plays as a striker for three seconds, then as the defender who wins the ball, then as the midfielder who passes it. No footballer experiences the game this way, and the resulting flow of play is nothing like an actual match.

Various games have tried to address it — tactical instructions, team mentality sliders, player-lock modes where you're stuck in one position. The last of these is the closest to reality and most players find it boring, which tells you something about what these games are actually for.

Off-ball movement is the hard part

The single biggest gap between game and reality is what the other ten players do when they don't have the ball.

In real football, off-ball movement is a continuous negotiation between eleven people, driven by a shared understanding of the team's principles, the specific opponent, the score, and what happened in the last five minutes. A striker's run is influenced by whether the midfielder has his head up, whether the defence is stepping, and whether he made the same run two minutes ago and it wasn't seen.

In a game, off-ball movement is generated by rules — run into space if it exists, offer for a pass if the ball carrier is under pressure, hold the line. These produce individually plausible movements that don't cohere into a team, because there's no shared model of what the team is trying to do.

You can see this most clearly in how AI teammates make runs. They're either constantly available in ways real players wouldn't be, or they're statically positioned. The rhythm of arriving and departing, of making a run to create space for someone else, essentially doesn't exist.

Why the AI can't just be better

The obvious response is: use better AI. Modern machine learning can play extremely complex games at superhuman level, so why not this?

A few reasons, and they're instructive.

The objective is wrong. Reinforcement learning agents optimise for winning. A football game's AI needs to produce entertaining, realistic, beatable football. Those aren't the same target and the second one is much harder to specify.

Compute budget. The AI for ten players has to run in a few milliseconds per frame on a console that's also rendering a stadium. That's a severe constraint, and the sophisticated approaches don't fit inside it.

Predictability is a feature. Players need to be able to learn the game. An AI that behaved with genuine footballing variability would feel random and unfair, because players can't read intent from a screen the way a footballer reads it from body language on a pitch.

That last one is the deep issue. A lot of what makes real football legible to its participants is information that simply isn't transmitted through a controller and a camera angle.

What the games get right

Worth acknowledging where the simulation is genuinely good, because it's not all deficit.

Ball physics is excellent — trajectory, spin, bounce off wet surfaces. Individual player animation is close to photorealistic. And the moment-to-moment feel of striking a ball, the timing and weight of it, has improved enormously and is genuinely satisfying.

Management simulations are a different and in some ways more successful genre. Because they abstract away the physical execution, they can model the things that matter — squad dynamics, tactical fit, development curves, transfer market behaviour — at a level of fidelity the action games can't approach. It's slightly odd that a text-and-spreadsheet game simulates football more accurately than a photorealistic one, but that's what the abstraction buys you.

Does it matter

Genuine question, and I think the answer is mostly no.

These games aren't trying to be simulations. They're trying to be enjoyable, competitive, and to deliver the feeling of football, which is a different target and one they hit reasonably well. A perfectly accurate football simulation would probably be quite dull to play — long periods of positional adjustment, a handful of chances, a lot of sideways passing.

Where it does matter is when the games are used as tactical education, which happens more than you'd think. A generation of fans has learned formation names and role descriptions from game menus, and the models embedded in those menus are simplifications that occasionally aren't right.

The idea that a formation is a fixed shape with numeric attributes, for instance, is a game concept that has leaked into how a lot of people discuss real football. That's a small distortion and it's real.