Machine learning (ML) and artificial intelligence (AI) are taking the world by storm. Armed forces around the world are thinking about new concepts or adaptations to integrate ML and AI in different functions affecting the military work environment. The purpose of this article is to broaden the perspective and challenge the perception of a military ML or AI integration. Military exercises are important tools for armed forces throughout the world. Learning to fight wars in a safe environment which emphasizes learning, saves lives on the battlefield. Through the theoretical framework of human activity, it is possible to understand the challenges soldiers and officers face when conducting double-sided live field exercises. The introduction of ML and AI in the exercise environment also creates a new learning environment with two more learning participants. By understanding the human zone of proximal development, we can also better understand the learning limitations and constraints an integrated ML or AI must consider before adjusting the algorithm. Simply accepting every collected data stream from a double-sided live field exercise might lead to learned faults and errors endangering the lives of soldiers and officers. To mediate the risk a new type of exercise needs to be developed with a focus on all participants learning opportunities both human and machine