The limits of AI in scouting: what an algorithm does not see
AI models learn from past data. If that data includes many players from some leagues and very few from others, the system “sees” the former better. A good player in a league with little data coverage can go unnoticed.
There are also qualities that are hard to turn into numbers: how a player reacts after a mistake, whether he communicates, whether he offers himself when the team is struggling. An algorithm can measure distance covered, but not attitude.
Another risk is false precision: a score of 87 out of 100 looks exact, even if it depends on model decisions that almost nobody sees. It is wise to always ask what data it uses, which players it was trained on and how the result was calculated.
That is why good scouts use AI to filter and compare, and then watch the game with their own eyes and talk to the player and those around him.