My research

My research addresses the applied aspects of AI to football from the point of view of the sociology of science. A baseball executive once said that numerical and visual data are irreconcilable methodologies: you cannot both accept and reject a player when analytics and eye-test don't converge. In my research, I inquire whether such a perfect dichotomy exists in the context of football, showing that the image tradition represented by visual data isn't less selective than the logic tradition represented by AI and, symmetrically, that the logic tradition of AI isn't less pictorial that the eye-test of the real football men.

IAS fellowship

My work on the interplay between different forms of data (visual and numerical) matches a core preoccupation in IAS for the relationship between computational and descriptive methods.

The image tradition of video data

When it comes to football, algorithms are recognised to have a partial view. Statistical models suggest certain goal scoring opportunities are more promising while reality suggests others. The numerical medium prefers events that are easier to measure such as set pieces. AI needs volumes of elementary data: any meaningful difference or contextual detail - including the identity of the player - is stripped out to inflate frequency. Less studied is whether video is symmetrically selective: the position of the camera, the distance, the televised nature of the event and how developments of media technology affected the way we perceive the game. AI is predicated as a science that externally and from above tries to harness football. What about the view of the tactical camera when players on the pitch should struggle with the "view from below"?

The logic tradition of AI

Symmetrically, my approach is also about recognising that the logic tradition of AI isn't less pictorial that the eye test of the real football men. This is apparent in the first place because video is raw data to football analytics algorithms. Also, at the end of the analytics process, visual inspection is how domain experts confirm the meaningfulness of algorithms' results. Finally, video clips are presented to players side by side numerical reports to provide more credibility. Focusing on such combinations aims to prevent an academic bias towards “innovation” or “newness” and to question the idea of a distinct or pure “logic” of AI, offering insights far beyond the sport.