By Martin Vongsuravatana (FootbAI) 
Liverpool’s idea: using artificial intelligence for football tactics
In the first match of the 2024 Champions League, Virgil Van Dijk restored Liverpool lead at the San Siro with a perfectly-taken corner. This was anything but a coincidence, given that Liverpool have been studying how to optimise corners using artificial intelligence.
Data is omnipresent today in football, both in quantity and quality, with extremely precise sensors on players, for example. Artificial intelligence, which feeds on this data, has naturally led DeepMind, Google’s AI research lab, to look into the matter. TacticAI was born with the aim of creating a tactical assistant. A collaboration began in 2019 between Liverpool, which has always been a pioneer in the data sector, and DeepMind, led by researcher Petar Veličković in particular.
AI Project Scoping
For TacticAI to be useful, its advice needed to be applicable. This simple logical rule quickly narrowed down the potential situations to predict. Indeed, open game phases are too unpredictable with 22 players able to move as they wish. Even though it is possible to anticipate a movement in the very short term, this prediction is unusable in a match by a coach.
4 situations have been studied by TacticAI:
- Penalties
- Free kicks
- Throws-in
- Corners
The TacticAI team decided to focus on corners because they are taken from a fixed point, unlike free kicks and throw-ins. The corner also has the advantage of being a frequent situation that can create a goal opportunity and is often determined before the match. The integration of TacticAI as a tactical assistant could be useful and easily applicable in this regard.
Setup of the Artificial Intelligence Model
The specific question that TacticAI wanted to answer is: “How to position players to increase/decrease shooting chances?”
TacticAI is based on a graph neural network (GNN) with particular emphasis on:
- Graph characteristic: the position of the posts (which is not the same depending on the stadium)
- Node characteristics: the position of the players, their speed, height, weight
- Edge characteristics: the information about teammates or opponents because this graph is fully connected
This organization makes it easy to specify various predictive tasks, particularly:
- The prediction of shots as graph classification
- The prediction of the receiver as node classification
The results applied to football
The results obtained were conclusive. TacticAI was able to predict the player receiving a corner 78.2% of the time. The same situations were presented to the experts from Liverpool, and their performance was 79%.
Then, TacticAI’s tactical adjustment recommendations were submitted to the experts at Liverpool. They were unable to differentiate them from real situations and even favored TacticAI’s corner tactics 90% of the time.
Thus, experts predict the corner receiver as well as TacticAI, and TacticAI offers credible and useful tactical adjustments.
Conclusion: artificial intelligence can be applied to football set-piece tactics
TacticAI is a very interesting tool that shows how artificial intelligence can assist coaches in their tactics. However, this contribution is very limited to certain phases of play in a sport that is mostly in motion with 22 players who can perform unexpected moments of genius. Good news for tech fans: AI can improve your team’s tactics. But, for skeptics: AI will not replace football coaches.
Find out more about machine learning for football.
If you’d like to find out more about this project, here’s a long video explaining the subject in English where Petar Veličković is interviewed by Aleksa Gordić.