•    A self-learning algorithm inspired by AlphaGo that learns from tabula rasa by interacting with a digital twin, to learn how to adapt short-term production decisions in a mining supply chain
•    New data from multiple different sources with varying levels of uncertainty are integrated into the resource models and equipment productivity scenarios, which are then used by a deep neural network agent to adapt decisions within minutes
•    Looking at a case study in a copper-gold mining supply chain shows substantial economic benefits with better realization of production targets. The results show the ability of the algorithm to understand the synergies and complexities inherent to a supply chain, and to learn how to use incoming data in real time to adapt production decisions.