Towards teaching multi agent system the concept of risks and safety for actionable climate decisions

Climate change is posing challenges for operating and designing critical infrastructure. Increasingly, AI has been used to enhance these decision making process. Reinforcement Learning has shown its advantages in dealing with difficult sequential decision making in games. When scaling to real life applications, their complexity and heterogenous nature potentially will require Multi Agent Reinforcement Learning (MARL) to provide adaptive capacity in a distributed manner. However, the human system is also characterised by the diverse belief of each individuals and groups - a feature that was captured in agent based models. AI/agent systems are evolving to work with human and become ubiquitous in real life/applications critical to society (such as health and transport). We argue that allowing belief transfer and full interactions across MARL actors in a three-layer model capturing data uncertainty, logical model and belief will help create a heterogeneous MARL system for better human-AI interaction that better aligns with human thoughts/values for actionable climate decisions.