Interactive Task Alignment as a POMDP

2026-07-17

AI Summary

This work addresses the challenge that current language models struggle to accurately align with users’ true tasks when intentions are ambiguous, incomplete, or inconsistent. The study formalizes task alignment as a partially observable Markov decision process (POMDP) and introduces a general method for automatically generating ambiguous interactive data from well-specified tasks. It further develops an alignment system integrating user simulators, supervised fine-tuning, and reinforcement learning. Human user studies across shopping, programming, and professional work scenarios validate both the fidelity of the simulator and the practicality of the approach. Experiments reveal that existing models recover only 22–32% of users’ true intent—substantially lower than the 48% achieved by humans—highlighting a significant gap in current systems’ ability to interactively resolve uncertainty compared to human performance.

Abstract

Current benchmarks for language models primarily evaluate execution on fully specified tasks. However, real user tasks are often ambiguous. Users arrive with incomplete, exploratory, or even inconsistent goals, requiring the assistant to first determine the intended task before carrying it out. We study this problem as task alignment: the ability to align with a user on their intended task. We introduce a general framework for converting specified tasks into underspecified interactions, formalized as a POMDP in which the model must infer a latent task from partial and evolving user intent. We validate our user simulator post hoc with a human user study. Across shopping, coding, and professional work settings, we find that while models often perform well once the task is specified, models still struggle with task alignment: current models act prematurely, interact ineffectively, and fail to resolve ambiguous requests. Models on average recover the user's intended task only 22-32% of the time under ambiguity. In a human study in the same setting, humans reach 48%, outperforming all evaluated models. We show that post-training with supervised fine-tuning and reinforcement learning improves task alignment, but models still lag behind humans in resolving uncertainty through interaction. Together, our results suggest that current models still lack key interaction abilities required for reliable agency.

Problem

• task alignment
• ambiguous tasks
• interactive task inference
• POMDP
• user intent

Innovation

• task alignment
• POMDP
• interactive ambiguity
• user simulation
• reinforcement learning