Manager ex Machina – Investigation of the transferability of productionmanagement tasks by systems with artificial intelligence

由于人工智能(AI)的不断发展，出现了用计算机控制甚至复杂决策的新可能性。从科学的角度来看，问题是人工智能实例在哪些类型的生产管理决策上优于人类。同样，这个问题也得到了极大的共鸣和科学的回答，适用于国际象棋、围棋或危险等游戏中的用例。人工智能，特别是强化学习(RL)的进展表明，人工智能在复杂场景中可以跟上甚至超过人类参与者。因此，本研究项目采用了将人与机器对决(AI实例)中获得的知识转移到生产管理中的原始方法。为此，生产管理决策在模拟环境中进行广泛的建模，然后由人类和人工智能使用RL执行。在AI实例完成对仿真环境的学习之后，分别确定人的性能相对于机器在仿真环境中的性能。这两种性能的比较提供了哪些决策优于基于人工智能的方法的信息，从而可以接管原本由人类做出的决策。最后，为了确保仿真结果在真实环境中的适用性，以实际生产环境为例进行了验证。知识的预期收益是识别生产管理决策，通过将RL应用于生产管理决策，基于人工智能的方法可以采用这些决策。其次，对于归纳结论，更重要的是回答了人机对决是否适合作为评估基于人工智能的方法在生产管理中的性能的方法。

Due to the continuous development of artificial intelligence (AI), novel possibilities arise to control even complex decisions by a computer. From a scientific perspective, the question arises for which types of production management decisions an AI instance is superior to a human. Similarly, this question has been answered with great resonance and scientific applied for use cases in games such as Chess, Go, or Jeopardy. Advances in AI, especially in reinforcement learning (RL), show that artificial agents can keep up with or even outperform human players in complex scenarios. Therefore, this research project takes the original approach of transferring the knowledge gained from the duel between humans and "machines" (AI instances) to production management. To this end, production management decisions are modeled with a broad spectrum in a simulation environment and then executed by both humans and an AI using RL. After the AI instance completes learning to the simulation environment, the performance of the human compared to the performance of the machine in the simulation environment is determined individually. The comparison of the two performances provides information on which decisions are superior to AI-based approaches and thus can take over the decision that would otherwise be made by a human. Finally, to ensure that the simulation-derived results are applicable in a real-world environment, they are validated in a real production environment as an example. The expected gain in knowledge is the identification of production management decisions that can be adopted by AI-based approaches by applying RL to production management decisions. Second, and more significant for the inductive conclusion, the question is answered whether the human-machine duel is suitable as a method for evaluating the performance of AI-based approaches in production management.