通过辩论实现超级智能的理性对齐（arXiv 论文）
Rational Alignments of Superintelligences via Debates
AI 导读
一篇 arXiv 论文研究超级智能（ASI）辅助问答场景下的理性对齐机制：ASI 智能体知道正确答案而人类委托方不知情，需确保其行为符合人类利益。论文提出在智能体效用不变的前提下控制人机交互方式的机制，并证明智能体可进一步推理、以子博弈完美均衡方式行为的辩论框架能带来更可预测的结果，某些情形下更为有利。
正文 · 原文
Abstract:Artificial superintelligences (ASI) is artificial intelligences with greater capability. However, as human principal has to use it, we have to ensure that the agent will perform in the interest of the human. We investigate the framework of ASI-assisted question answering where ASI agents know the correct answer but the human does not. We provide some example of mechanism for rational alignment, where the agent utility is unchanged, but the interaction between the agent and the human principal is controlled to be in a certain fashion. We show specifically that a debate framework where the agents can further reason and behave in a subgame perfect equilibrium manner will lead to a more predictable result, and in some case will be favorable.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.16455 [cs.AI] |
| (or arXiv:2611.16455v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.16455 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Thanawat Sornwanee [view email]
[v1]
Thu, 12 Nov 2026 08:07:48 UTC (9 KB)
来源：arXiv cs.AI · arxiv.org