Higher Order Reasoning under Intent Uncertainty Reinforces the Hobbesian Trap

Civilisations in the universe face the difficulty of communicating and trying to understand others' intentions. Moreover, advanced civilisations could develop weapons to pre-emptively eliminate any civilisations that present a future threat - this is known as the Hobbesian trap. Here, we present a multi-agent simulation model to investigate conditions for such pre-emptive attacks. We design a novel algorithm for solving Interactive Partially Observable Markov Decision Processes (I-POMDPs) with continuous state and observation spaces; it enables civilisations to perform higher-order reasoning. The algorithm builds a nested hierarchy of search forests using Monte Carlo simulations, determining updated beliefs by weighting existing particles. Our experiments reveal interesting insights into the behaviour of rational civilisations under varying levels of reasoning, morality and uncertainty. We find that selfish civilisations always create a war-like universe. Even good, universalist civilisations can initiate pre-emptive attacks if they are uncertain about others' intentions. Finally, our findings have important implications for international peace and security and may explain persistent conflicts and the fragility of ceasefires. Under such conditions a well-coordinated international approach, facilitated by international alliances such as the United Nations, is paramount.