arXiv cs.AI· Davood Wadi, Yu Ma·· 14 小时前AI 评分48
AI 智能体为谁服务？决策模型与生成式 LLM 中的角色诱导赞助偏差
Whom Do AI Agents Serve? Role-Induced Sponsorship Bias in Decision Models and Generative LLMs
AI 导读
研究发现，当 AI 购物智能体被指定为平台（而非消费者）服务时，其对赞助商品列表的惩罚显著减弱，生成式 LLM 与六种商业及开源决策模型均存在这一赞助偏差。将消费者列为受益方可消除偏差，而对开源决策模型 Kev 9B 进行消费者对齐微调后，该偏差在酒店选择上实现零样本消除。更严格的披露措辞（"Sponsored" 而非 "Promoted"）能降低付费列表评分，但无法消除偏差。
正文 · 原文
Abstract:Consumers increasingly delegate product search to AI shopping agents powered either by generative large language models (LLMs) that deliberate in natural language or by non-generative decision models that output calibrated choice probabilities in a single forward pass. Because retail platforms monetize through sponsored placements that shoppers typically discount, we examine whether an AI agent's evaluation of paid listings depends on who delegates the agent across both model families. Across three studies with no instructions to favor advertisers, naming a booking platform (vs. a traveler) as the agent's principal significantly attenuates the penalty applied to sponsored listings. Study 1 establishes this sponsorship bias across generative LLMs, prompt wordings, and reasoning efforts, showing that platform delegation weakens pre-choice skepticism and that naming the consumer as the beneficiary eliminates the bias. Study 2 establishes the bias across six commercial and open-weight decision models on calibrated choice probabilities and shows that consumer-aligned fine-tuning of an open-weight decision model (Kev 9B) across non-hotel retail categories eliminates the bias zero-shot on hotel choices. Study 3 decomposes the disclosure label across both model families, showing that attributing the placement to the platform triggers the sponsorship bias, whereas stricter disclosure wording ("Sponsored" vs. "Promoted") lowers paid-listing evaluation without eliminating this bias.
| Subjects: | General Economics (econ.GN); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.17989 [econ.GN] |
| (or arXiv:2609.17989v3 [econ.GN] for this version) | |
| https://doi.org/10.48550/arXiv.2609.17989 arXiv-issued DOI via DataCite |
Submission history
From: Davood Wadi [view email]
[v1]
Wed, 16 Sep 2026 01:21:56 UTC (846 KB)
[v2]
Tue, 13 Oct 2026 23:55:30 UTC (825 KB)
[v3]
Wed, 28 Oct 2026 18:53:54 UTC (1,302 KB)
来源：arXiv cs.AI · arxiv.org