28 Nov The model that demonstrates that Collective Intelligence outperforms Artificial Intelligence in football
When winning depends on something no algorithm can see
Imagine preparing a match against a team that is not especially fast, not physically superior, and not individually brilliant.
And yet, every coach who faces them says the same thing:
“They move as one. They play as if they share a single mind.”
You look at the data, and nothing about their physical output or isolated actions explains their dominance.
But when the ball moves, they move — together.
When one accelerates, three adjust.
When pressure arrives on the left, balance appears on the right.
Every coach recognizes this phenomenon.
Until now, we simply called it “being a team.”
Our latest research — published in a peer-reviewed scientific journal and developed by the Football Intelligence & Performance Department at LALIGA — shows that this intuition has a measurable structure.
And, surprisingly:
A football team’s Collective Intelligence is a better predictor of performance than any Artificial Intelligence model we tested.
This article explains why — and how coaches can use this knowledge today.
What were we trying to understand?
Football is full of assumptions:
- “We need faster players.”
- “We need more talent.”
- “We need players who win duels.”
But what if the real competitive advantage is not individual talent, but the integration of information between players?
For years, neuroscience has shown that the human brain works not because neurons are individually impressive, but because they integrate information efficiently. This integration can be quantified mathematically using Integrated Information Theory (IIT) through a variable called Φ (phi).
Our question was simple:
Can a football team integrate information in a similar way to a brain? And if so, does this explain performance?
How we tested it: football as a complex information system
Tracking the “activation” of players
Using Mediacoach® data from 100 matches of LALIGA, we transformed player movement (positions and speed) into a binary “activation” time series — similar to how neural activity is modelled in cognitive neuroscience.
Every half-second, each player was classified as:
- Active (high-speed, purposeful behaviour)
- Inactive (low-speed, low-engagement state)
Goalkeepers were excluded because their behavioural patterns are fundamentally different.
Transition Probability Matrices
We analysed how combinations of player states changed from one moment to the next.
This allowed us to construct Transition Probability Matrices (TPMs) for:
- all pairs of players
- all trios of players
- some quartets (computationally demanding)
These TPMs describe how the “system” — the group of players — evolves over time.
Computing Φ: the measure of Collective Intelligence
With TPMs in hand, we used the mathematical tools of Integrated Information Theory to calculate how much information the players integrate as a collective system.
This is not:
- passing networks
- synchronisation metrics
- heat maps
- xG models
This is a causal measure of how each player’s behaviour changes the behaviour of the others — and how the group forms a single, self-organised unit.
What we found: the signature of a “collective football brain”
1. Football teams do integrate information
Φ values were consistently above zero — the minimum condition for integrated behaviour.
This means:
Teams behave as interconnected systems, not as isolated individuals.
2. Top teams show significantly higher Φ
Across all matches, FC Barcelona, Real Madrid and Atlético de Madrid displayed significantly higher Φ values than other teams.
This was not an accident:
- They integrate information more efficiently.
- They behave more coherently as a unit.
- Their collective structure is stronger than their individual parts.
Statistically, the difference was extremely robust.
3. The team with higher Φ creates more shots on goal
Across all matches:
- The higher a team’s Φ → the more shots they generated
- The difference in Φ between two teams → predicted the difference in shots
This is crucial:
Collective Intelligence influences performance even when physical, technical or tactical metrics look similar.
4. Randomising the players destroys all effects
When we replaced one player’s behaviour with a random time series:
- Φ differences disappeared
- The relationship between Φ and shots collapsed
This shows that:
Φ is capturing a real, meaningful signal of team organisation, not statistical noise.
What does this mean for football coaches?
1. Cohesion is not a cliché — it’s a measurable competitive advantage
Teams do not win because of individual brilliance alone.
They win because their internal causal structure is more integrated.
A team with moderate individual talent but high Φ can outperform a team full of stars with low Φ.
This aligns with other research showing that:
- synchronisation enhances tactical dominance
- functional connectivity predicts passing quality
- complex networks outperform isolated units in game intelligence
2. Training should not only improve players — it should improve the system
If you improve a player but disrupt the system, Φ drops.
This has real implications for:
- session design
- team selection
- substitution strategy
- player recruitment
- in-game adjustments
Coaches should ask not only:
“Is this player good?”
But rather:
“Does this player integrate well with the system?”
“Does he increase or decrease our Φ?”
3. Tactical identity emerges from information integration
The model explains why certain styles produce different levels of Φ:
- Tiki-taka → high Φ (dense, short interactions, constant mutual influence)
- Direct play or strong transitions → lower Φ (fewer interaction points)
This does not mean one style is better.
It means each style requires different coherence patterns to be successful.
The key question is:
Does your style maximise your team’s ability to integrate information?
4. The pitch becomes a laboratory
This study encourages a new mindset:
Don’t assume — test.
Don’t guess — measure.
Don’t rely on intuition alone — validate through behaviour.
Football is a living experiment.
Every minute generates data that can confirm or refute your assumptions.
This is disciplined curiosity — the foundation of modern performance.
Limitations (and why they reinforce, not weaken, the message)
- We computed Φ on subsets of players (pairs, trios) because full-team analysis is not yet computationally feasible.
- Activation was derived from speed — a simplification necessary for large-scale modelling.
- TPMs were constructed observationally rather than via controlled interventions.
Despite these limitations:
- The results were strong,
- consistent,
- replicable,
- and validated through null tests.
This strengthens the core insight:
Collective Intelligence exists, it is measurable, and it matters.
What this research adds to football science
Previous work has shown:
- groups behave as complex systems
- synchronisation predicts performance
- passing networks distinguish winning teams
- tactical structure emerges from shared information
Our study adds something new:
A mathematical way to quantify the “team mind” — the integrated intelligence that emerges when players interact.
This is not metaphor.
It is measurable, evidence-based and predictive of performance.
Conclusion: the intelligence that wins in football is Collective, not Artificial
Artificial Intelligence models may analyse the game.
But Collective Intelligence is what plays the game.
And now, for the first time, we can measure it.
The next competitive frontier in football is not only improving individuals.
It is improving the system they create together.