LLM-Enhanced Contract Design for Edge AIGC Service Provisioning in Teleoperation

With the prevailing of AI-generated content (AIGC) services, incentive mechanism design is an emerging field in AIGC networks. The incentive mechanism design under information asymmetry of hidden action remains underexplored. In this paper, we research the bonus design for a teleoperator that incentivizes the edge AIGC service provider (ASP) to provide high-quality AIGC services when the teleoperator cannot observe the ASP's private settings and chosen actions (diffusion steps). We formulate the problem as an online learning contract design problem and propose a large language model (LLM)-empowered framework to solve it efficiently. Specifically, we utilize the LLM to tackle the ASP's settings approximation problem, NP-hard and with unknown size optimization variables, via the LLM's expertise in various domains. Upon the ASP's approximated settings, we address the contract derivation problem using convex optimization techniques. Simulation results on our Unity-based teleoperation platform showcase that our proposed method augments the utility of the teleoperator by 5 ∼ 40% in comparison with benchmarks. Index Terms-large language model (LLM), edge AI-generated content (AIGC), incentive mechanism design, moral hazard.