A team of researchers has developed DraftFM, a model for predicting card picks in the popular trading card game Magic: The Gathering. Unlike traditional supervised learning models, which require training data from past drafts, DraftFM uses only publicly available information about the cards. This allows it to make predictions even before any draft logs are available. The model was tested on 149 million human picks and achieved a high level of accuracy, including correctly ranking cards in an unreleased set. --- Why it matters: This matters because it shows that AI can be used to improve gameplay experience for Magic: The Gathering players, potentially making the game more accessible or providing new insights for competitive players. It also demonstrates the potential for applying similar techniques to other complex decision-making domains. Source: https://arxiv.org/abs/2608.19568