誠實說謊：理解反思型代理中的記憶虛構

摘要

反思型智能體依賴於自我生成的反思作為記憶，隱含假設了智能體能準確診斷自身失敗。我們證明此假設可能系統性地失效：在ALFWorld與HumanEval中，智能體會儲存對任務的自信但錯誤的解讀，並在多次試驗中持續依此行動，即使環境每次都會重置為正確的任務。我們將此失敗模式稱為「記憶虛構」，並引入「反思重複率」（RRR），這是一種基於日誌的指標，用於檢測對錯誤反思內容的重複依賴。透過RRR，我們在ALFWorld中識別出16個凍結環境，其中121次反思完全未提及正確目標物件，並在HumanEval中發現4個類似案例。我們的緩解方法以程式化提取軌跡層級的失敗信號取代開放式自我診斷，將正確物件提及率從0%提升至86%，RRR從0.64降至0.10，並成功解決ALFWorld中16個凍結環境中的3個，顯示反思記憶可能強化而非修正錯誤信念。

English

Reflexion-style agents rely on self-generated reflections as memory, implicitly assuming that agents can accurately diagnose their own failures. We show that this assumption can fail systematically: across ALFWorld and HumanEval, agents store confident but incorrect interpretations of the task and continue acting on them across trials, even though the environment resets to the correct task each time. We call this failure mode memory confabulation and introduce the Reflection Repetition Rate (RRR), a log-based metric that detects repeated reliance on incorrect reflective content. Using RRR, we identify 16 frozen environments in ALFWorld, where 0 of 121 reflections mention the correct target object, and 4 analogous cases in HumanEval. Our mitigation replaces open-ended self-diagnosis with programmatic extraction of trajectory-level failure signals, increasing correct object mention from 0% to 86%, reducing RRR from 0.64 to 0.10, and solving 3 of 16 frozen ALFWorld environments, suggesting that reflective memory can reinforce false beliefs rather than correct them.