TacticAI: an AI assistant for football tactics

TacticAI is an AI assistant for football tactics developed by Pet Vitkovic from DeepMind in collaboration with Liverpool Football Club. Using graph machine learning and geometric deep learning, TacticAI focuses on corner kicks to predict and generate tactical suggestions preferred by experts, potentially revolutionizing coaching strategies and player performance in football.

In a recent paper called TacticAI, developed by Pet Vitkovic from DeepMind in collaboration with Liverpool Football Club, an AI assistant for football tactics was introduced. The algorithm uses a predictive and generative component to explore alternative player setups for corner kicks, with suggestions that are indistinguishable from real tactics and favored by experts 90% of the time. TacticAI represents corner kick situations as graphs with players as nodes, modeling player interactions using geometric deep learning. This approach enhances sample efficiency and robustness of neural networks by capturing symmetries in the game.

The project involved using graph machine learning and geometric deep learning to study football tactics, specifically focusing on corner kicks. TacticAI is capable of predicting and generating tactical suggestions that are not only plausible but also preferred by experts, demonstrating its potential as an AI football tactics assistant. By focusing on set pieces like corner kicks, the AI model can provide practical and validated suggestions for coaches and analysts, aiming to streamline decision-making during matches. The use of graph neural networks allows for explicit reasoning about pairwise player relations, making it well-suited for modeling complex interactions in football.

While the AI system is more aligned with system one reasoning due to its focus on short-term predictions for corner kicks, it has the potential to influence coaching decisions and player positioning based on shot probabilities. The generative system, based on a variational autoencoder, provides subtle refinements to existing tactics, aiding coaches in identifying patterns and making strategic adjustments. By freeing up coaches to focus on creative decision-making rather than analyzing game situations, TacticAI could lead to more innovative play and strategic concepts in football. The AI model’s ability to adapt to new tactics and concepts through continuous training allows for ongoing improvement and creativity in gameplay.

The future of AI in football tactics may revolve around set pieces like corner kicks, with potential expansions to other scenarios like free kicks and throw-ins. Graph neural networks are well-suited for modeling natural problems with complex interactions, making them a valuable tool for analyzing player relationships and strategic patterns in team sports. The use of AI assistants like TacticAI could lead to a more creative and adaptive approach to coaching, enhancing decision-making processes and fostering innovation in the sport. Overall, the integration of AI systems in football tactics has the potential to revolutionize coaching strategies and player performance, offering new insights and possibilities for tactical development in the sport.