The article discusses a new machine learning system called FEERL, which helps machines learn without needing specific training examples. This system is designed to handle large environments more effectively than traditional reinforcement learning methods. FEERL can reuse rules it has learned, making it more efficient and adaptable for different tasks.