State representation affects agent learning and task transfer success
State Representation Shapes Long-Horizon Agent Learning and Transfer: An Empirical Study
Artificial Intelligence
Summary
When teaching AI agents to perform tasks step-by-step, how the agent sees the current situation matters a lot. The authors found that agents trained viewing tasks in a simple way struggled when tested on more detailed views, and vice versa. They tested different sizes of AI models and observed that changing the way the state is shown can limit how well the agent can apply what it learned to new, longer tasks. Their study shows that the way task information is represented is a key challenge for improving AI learning and transfer.
What this means in practice
- •For ai system developers: Improve AI task training by carefully matching state representation to testing scenarios for better long-term task completion.
- •For robotics software engineers: Design robot control modules that maintain consistent state views during training and execution to enhance tool-use effectiveness.
Authors
Dong Nie, Rui Lan, Yixin Nie, Lingzi Hong
Abstract
An end-to-end success rate alone cannot distinguish failure to learn an action from failure to apply it as task state changes. We study this distinction in deterministic tool workflows that provide the current artifact state and candidate action recipes. Using 1.7B, 8B, and 27B Qwen models, we compare supervised learning across workflow horizons and cross the state representation used for training with the one used at evaluation. Horizon-matched supervision yields high completion, but an H4-trained 1.7B model transfers poorly to longer workflows under full-state observations. In an H16 crossover, compact-trained 1.7B adapters complete 50/50 held-out tasks in compact view and 0/50 in full view for each of three paired training seeds; full-trained outcomes vary across seeds. A separate 8B adapter pair shows a similar asymmetry. The view dependence also appears in paired-choice tasks where both candidate calls are executable and the correct choice depends on table contents; here, the compact mask does not consult the target action. An explicit relevance cue improves full-view execution but does not close the gap for the tested checkpoint. In a separate 27B ScienceWorld study, compact-state training improves local accuracy under both views, indicating that the transfer pattern depends on the task and observation construction. Together, these experiments identify representation-sensitive transfer as a bottleneck in the controlled workflows, without isolating content reduction from other changes to the observation.