arXiv cs.AI· Yinhong Liu, Zhili Tan, Zilin Wang, Zhijiang Guo·· 12 小时前AI 评分47
LLM 智能体的动作空间塑造：测量与缓解工具 Schema 偏差
Action-Space Shaping for LLM Agents: Measuring and Mitigating Tool-Schema Bias
AI 导读
研究提出“schema bias（Schema 偏差）”现象：同一可执行动作经不同但功能等价的工具定义暴露时，LLM 智能体表现并不一致。作者构建含九种重写算子的可执行转换框架，在最多 32 种 schema 变体上评测 11 个 LLM（含 2 个闭源模型），成功率仅因 schema 不同就从完全失败跨越到 97%。
正文 · 原文
Abstract:Large Language Models (LLMs) have shown strong performance on tool-use agentic tasks when given a fixed tool schema. Yet a tool schema is not the action space of an agent; it is merely one interface representation of it. The same executable action can be exposed through many different, functionally equivalent tool definitions, and an agent that has truly learned a task should behave consistently across them. We show that current agents often do not, a phenomenon we term schema bias. To study this systematically, we introduce an executable transformation framework that rewrites a native tool schema using nine operators, including merging and splitting tools, altering how a single tool is expressed, and distributing one action across several dependent calls. The tasks, executable actions, and reachable states remain fixed, so any change in success is attributable to the interface alone. Evaluating eleven LLMs, including two closed models, on up to 32 schema variants, we ask how large schema bias is, how it manifests, whether the difficulty of a schema variant can be predicted without a full evaluation, and whether training removes it. We find that schema bias is substantial even for the newest models: success rates range from complete failure to 97% depending solely on the schema. To reliably estimate schema difficulty, it requires running a small sample of the target queries. Training repairs a schema variant only when that variant appears in the training data.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.34971 [cs.AI] |
| (or arXiv:2609.34971v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.34971 arXiv-issued DOI via DataCite |
Submission history
From: Yinhong Liu [view email]
[v1]
Mon, 28 Sep 2026 11:52:43 UTC (403 KB)
[v2]
Fri, 16 Oct 2026 07:18:49 UTC (403 KB)
来源：arXiv cs.AI · arxiv.org