Controllable User Simulation

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Tennenholtz, Guy, Meshi, Ofer, Globerson, Amir, Shalit, Uri, Jeong, Jihwan, Boutilier, Craig
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866918496339755008
author Tennenholtz, Guy
Meshi, Ofer
Globerson, Amir
Shalit, Uri
Jeong, Jihwan
Boutilier, Craig
author_facet Tennenholtz, Guy
Meshi, Ofer
Globerson, Amir
Shalit, Uri
Jeong, Jihwan
Boutilier, Craig
contents Using offline datasets to evaluate conversational agents often fails to cover rare scenarios or to support testing new policies. This has motivated the use of controllable user simulators for targeted, counterfactual evaluation, typically implemented by prompting or fine-tuning large language models. In this work, we formalize controllable simulation as a causal inference problem. By bridging natural language evaluation with off-policy evaluation methodology, we show that the standard practice of training simulators via supervised fine-tuning on post-hoc trajectory labels yields a structurally biased model. Specifically, these labels are inextricably coupled to the data-generating behavior policy, injecting a look-ahead bias that breaks causal consistency. Furthermore, we prove that under policy shift this failure causes the variance of evaluation metrics to explode geometrically, a phenomenon we term controllability collapse. To restore causal consistency, we establish theoretical conditions for accurate simulation and propose practical training mitigations: a priori controls, step-wise dynamic controls, and direct policy-conditioned learning. Empirical evaluation confirms that while standard global controls distort conversational distributions and collapse behavioral diversity, our causally grounded simulators eliminate look-ahead bias, preserve natural variance, and exhibit robust zero-shot generalization to unseen agent behaviors.
format Preprint
id arxiv_https___arxiv_org_abs_2605_11519
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Controllable User Simulation
Tennenholtz, Guy
Meshi, Ofer
Globerson, Amir
Shalit, Uri
Jeong, Jihwan
Boutilier, Craig
Artificial Intelligence
Computation and Language
Machine Learning
Using offline datasets to evaluate conversational agents often fails to cover rare scenarios or to support testing new policies. This has motivated the use of controllable user simulators for targeted, counterfactual evaluation, typically implemented by prompting or fine-tuning large language models. In this work, we formalize controllable simulation as a causal inference problem. By bridging natural language evaluation with off-policy evaluation methodology, we show that the standard practice of training simulators via supervised fine-tuning on post-hoc trajectory labels yields a structurally biased model. Specifically, these labels are inextricably coupled to the data-generating behavior policy, injecting a look-ahead bias that breaks causal consistency. Furthermore, we prove that under policy shift this failure causes the variance of evaluation metrics to explode geometrically, a phenomenon we term controllability collapse. To restore causal consistency, we establish theoretical conditions for accurate simulation and propose practical training mitigations: a priori controls, step-wise dynamic controls, and direct policy-conditioned learning. Empirical evaluation confirms that while standard global controls distort conversational distributions and collapse behavioral diversity, our causally grounded simulators eliminate look-ahead bias, preserve natural variance, and exhibit robust zero-shot generalization to unseen agent behaviors.
title Controllable User Simulation
topic Artificial Intelligence
Computation and Language
Machine Learning
url https://arxiv.org/abs/2605.11519