Summary
Handling complex tasks with AI often requires people to give detailed instructions, check results, and clear up confusion. The paper proposes 'Avatar Intelligence,' where personal AI agents continuously represent a user’s goals and preferences over time and across different situations. These agents gradually take on more autonomous decisions while letting people oversee important outcomes. The authors suggest ways to measure how well agents help without burdening users and highlight challenges like making agents remember personal preferences, evolve with users, and maintain safety. This research aims to make AI assistants more helpful and trustworthy for long-term use.
What this means in practice
- •For enterprise technology teams: Create intelligent assistants that manage ongoing business tasks by reflecting employee goals and reducing daily supervision.
- •For service robot developers: Develop robots capable of adapting to individual user preferences continuously while ensuring human control over critical decisions.
A position paper. It proposes an approach and reports no results.
Authors
Zheng Wu, Xinbei Ma, Yao Yao, Zhaoyuan Huang, Haowen Hu, Yijie Lu, Taolei Wang, Yifei Yang, Wuyue Xiong, Pengzhou Cheng, Zhuosheng Zhang
Abstract
Personal AI agents are increasingly capable of handling long-horizon tasks across applications, yet users often need to invest substantial effort in specifying goals, resolving ambiguity, and verifying outcomes. We argue that progress should be measured by how effectively agents amplify human agency: enabling people to pursue their goals with less routine supervision while retaining control over consequential decisions. We advocate Avatar Intelligence, a paradigm of persistent personal agents that represent users' goals, values, and constraints through sustained decision delegation across contexts and time. We describe a maturity roadmap from human-directed execution to autonomous operation within delegated domains, and introduce the Agentic Oversight Index (AOI) as a framework for evaluating delivered value against agent-side operational cost and human oversight burden. We identify three core research challenges: identity-continuous personalization that carries user preferences across tasks, user-situated evolution through Rehearsal Engineering, and safety mechanisms that combine technical control with accountability and recourse. These directions define a research agenda for reliable decision delegation to personal agents across tasks and over time.