MIT researchers have developed a more efficient algorithm, Model-Based Transfer Learning (MBTL), to enhance AI decision-making in tasks like traffic control. This innovative approach strategically selects key tasks to train on, improving performance and cutting training costs.