ASPAC · Algorithmics of Stochastic Principal-Agent Coordination

When making decisions, autonomous agents rely on their own incentives and private information. When these factors diverge, conflicts can arise, leading to inefficiencies or suboptimal outcomes. Economists have studied this phenomenon under the principal-agent framework, where the principal seeks to coordinate the agents so that the agents' rational choices align with the principal's objectives. This framework is gaining increasing attention, especially in the era of AI, where automated agents become more capable of making decisions autonomously.In computer science, algorithmic approaches to principal-agent coordination strategies have become an area of growing interest, giving rise to influential subfields, such as algorithmic information design, contract design, and automated mechanism design. Despite this progress, much of the literature focuses on static, one-shot models, which fall short of capturing the complexities of real-world problems occurring in stochastic environments. These problems are inherently dynamic and involve sequential decision-making. Solving them requires addressing fundamentally new computational challenges arising from the interplay between the complexities of the principal-agent model and the dynamics and stochasticity of the environment.This project aims to tackle these challenges by investigating the computational and algorithmic foundations of stochastic principal-agent coordination (SPAC). The goal is to introduce a useful framework for formalising SPAC and develop algorithms and computational theories regarding the tractability, approximability, and learnability of problems within the framework. By exploring these fundamental aspects, we expect the project to substantially advance the theoretical literature while contributing algorithmic solutions to real-world applications of SPAC.