Using digital twins to evaluate decisions made by artificial intelligence agents

The article is devoted to the use of digital twins for evaluating decisions made by artificial intelligence agents before their implementation in real production processes. A digital twin is considered as a working environment in which the consequences of autonomous choice can be tested in advance and the stability of the selected strategy can be assessed. An AI agent differs from a conventional analytical model in its ability to independently choose actions and change its behavioral trajectory when receiving new signals from the environment. The article examines the reliability of a digital twin, the gap between the virtual and real environments, reinforcement learning, and the verification of decisions in hybrid teams composed of humans and AI agents. The study identifies signs of weak strategies and formulates safe boundaries for autonomous behavior.