《科学》 2026年10月1日 · 中文解读
英国资助机构用“AI初筛”淘汰资助申请
U.K. funder used ‘AI triage’ to reject grant proposals · Catherine Offord
约 5 分钟News在小程序里点播,20 到 60 分钟做好
这篇讲什么
英国公共资助网络CRANE用多个人工智能模型评估179份网络安全研究申请,在人工评审前就淘汰了排名后50%的项目。兰卡斯特大学Claire Hardaker团队的申请因此被拒,引发研究人员对AI评审公平性和未发表想法隐私泄露的担忧。CRANE称流程正在审查中,其上级机构UK Research and Innovation则表示不允许用生成式AI进行申请评估。
原文开头
Researchers raise privacy and fairness concerns following unexpected use of AI-based assessment CATHERINE OFFORD When forensic linguist Claire Hardaker’s team at Lancaster University submitted a grant proposal on the risks of trusting artificial intelligence (AI) models in voice analysis, they assumed it would be read by at least one reviewer. They had drafted the proposal in response to a call from CRANE, a publicly funded U.K. network that supports research on cybersecurity. Application guidelines seemed to describe a standard review process, with subject matter experts deciding which projects to fund. So Hardaker was stunned by the rejection email she received last week. The proposal “was not selected to progress beyond the AI triage stage,” it read. …
摘自《科学》(Science)2026年10月1日,Catherine Offord。仅引用开头一小段供了解文章,版权归原刊所有,全文请阅读原刊。

微信扫码,在小程序里听完整版
不用登录先听一篇 · 或在微信搜索小程序 晨间信号