喜歡的不一定最好：AI 助手選擇的兩難
Abstract (原文摘要)
As AI usage becomes more prevalent in social contexts, understanding agent-user interaction is critical to designing systems that improve both individual and group outcomes. We present an online behavioral experiment (N = 243) in which participants play three multi-turn bargaining games in groups of three. Each game, presented in randomized order, grants access to a single LLM assistance modality: proactive recommendations from an Advisor, reactive feedback from a Coach, or autonomous execution by a Delegate; all modalities are powered by an underlying LLM that achieves superhuman performance in an all-agent environment. On each turn, participants privately decide whether to act manually or use the AI modality available in that game. Despite preferring the Advisor modality, participants achieve the highest mean individual gains with the Delegate, demonstrating a preference-performance misalignment. Moreover, delegation generates positive externalities; even non-adopting users in access-to-delegate treatment groups benefit by receiving higher-quality offers. Mechanism analysis reveals that the Delegate agent acts as a market maker, injecting rational, Pareto-improving proposals that restructure the trading environment. Our research reveals a gap between agent capabilities and realized group welfare. While autonomous agents can exhibit super-human strategic performance, their impact on realized welfare gains can be constrained by interfaces, user perceptions, and adoption barriers. Assistance modalities should be designed as mechanisms with endogenous participation; adoption-compatible interaction rules are a prerequisite to improving human welfare with automated assistance.
這篇論文揭示了一個非常有趣，甚至有點反直覺的現象：人類對 AI 的偏好，與 AI 實際能帶來的效益，是脫鉤的。
在談判賽局中，研究者提供了三種 AI 模式：
- Advisor (顧問)：主動給建議。
- Coach (教練)：你問它才回饋。
- Delegate (代理)：全權交給它去談。
結果顯示，大部分的人最喜歡 Advisor（顧問模式）。這很符合人性，我們喜歡掌控感，喜歡「被輔助」而不是「被取代」。我們希望 AI 是那個在旁邊遞茶水、給點子的小秘書。
但數據卻狠狠地打臉了：Delegate (代理模式) 才能帶來最高的個人收益。 更驚人的是，這種代理模式還產生了「正外部性」——也就是說，就算你不用 AI，只要你的對手用了 Delegate，你也會跟著受益，因為 AI 會提出更理性、更雙贏 (Pareto-improving) 的方案，把整個市場的餅做大。
這就是所謂的 「偏好-效能錯位」(Preference-Performance Misalignment)。
我們以為我們需要的是一個「聰明的建議者」，但事實上，我們最需要的可能是一個「理性的執行者」。 人類的情緒、猶豫、以及對掌控權的執著，往往成為了阻礙最優解的絆腳石。AI 代理人之所以能表現得比人類好，正是因為它能像一個冷靜的造市商 (Market Maker)，不受情緒干擾地拋出最優提案。
這給了我們一個深刻的啟示： 未來的 AI 設計，或許不該一味地討好人類的「偏好」（比如做得越來越像人類、越來越順從），而是要思考如何設計出一種機制，讓人們願意放手，信任並交付權力給 AI。
真正的智慧，或許不在於變得更強，而在於知道何時該讓位給更強的邏輯。 我們需要的不是一個聽話的隨從，而是一個能帶領我們穿越非理性迷霧的領航員。
Source: Assistance with Autonomous Agents (arXiv:2602.12089v1)