面向 AI 安全的人类与 AI 互补性研究
Towards Human-AI Complementarity for AI Safety
研究构建了已知最大的安全相关人类-AI 互补性评测套件，覆盖 8 个数据集、6 个领域共 3,957 个实例，并收集 416 名人类评分者的 35,613 条判断，发现存在可观的互补效应空间。但测试的辅助方法反而让人更少纠正 AI 错误，即使人类整体准确率上升；所测试的混合方法平均而言相比更优的单一来源提升有限。
Authors:Sander Lanbo Shi, Miranda Anna Christ, Eric Modesitt, Andrew Louis, Konrad Cheng, Brandon Thai Tran, Sewoong Lee, Ratnakar Pawar, Gaia Molinaro, Juliana Bain, Nandika Donthi, Maja Pavlovic, Annya Dahmani, Francesca Bianco, Abhishek Shah, Tyler Crosse, Arth Singh, Karissa Huang, Sukanya Krishna, Konstantinos Voudouris, David Demitri Africa, Joshua Jacob, Rishub Jain
Abstract:Safely overseeing AI is increasingly challenging for the humans doing it, as they must evaluate longer and more complex outputs with every new model generation. To mitigate this challenge, we could leverage complementarity: where humans and models make different mistakes, combining their judgements could yield an oversight signal stronger than either produces alone. We study this opportunity in realistic, safety-relevant settings, contributing the largest known suite of such tasks (3,957 instances across 8 datasets and 6 domains) as well as a platform and data pipeline for measuring human, AI, and human-AI performance on matched items. We collected 35,613 judgements from 416 human raters, finding a considerable opportunity for a complementarity effect. We also found that with the assistance methods we tested, human raters corrected the AI's errors less often, even when overall human accuracy rose, and the hybridisation methods we tested added little over the better single source on average. While complementarity is an important and feasible engineering target for safe oversight, achieving it remains an open challenge for future work.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.24969 [cs.AI] |
| (or arXiv:2610.24969v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.24969 arXiv-issued DOI via DataCite |
Submission history
From: Lanbo Shi [view email]
[v1]
Sun, 18 Oct 2026 02:49:28 UTC (857 KB)
[v2]
Tue, 27 Oct 2026 06:46:31 UTC (856 KB)
来源：arXiv cs.AI · arxiv.org