监测AI模型以探明其意图和动因，被广泛视为确保该技术安全运行的关键环节。英国AI安全研究所（AISI）模型透明度负责人约瑟夫·布鲁姆表示，这项工作也变得日益困难。
The ability to monitor AI models to detect what they’re trying to do and why is widely considered a crucial part of the effort to keep the technology operating safely. It’s also increasingly hard to do, said Joseph Bloom, the head of model transparency at the U.K. AI Security Institute.
布鲁姆在《The Information》的AI Deep Dive节目最新一期中表示：“总体而言，我们看到可监测性——在各个方面或从各个角度——都在下降。”这是AISI技术人员首次接受公开记录采访之一。（完整节目可在此收听。）该研究所是全球最大的专注于AI的政府机构之一，其工作包括评估领先AI公司的模型。
“In general, we’re seeing monitorability—in various respects or from various angles—decrease,” Bloom said on the latest episode of The Information’s AI Deep Dive, in one of the first on-the-record interviews that a member of technical staff at AISI has given. (The full episode is available here.) The institute is one of the world’s largest government bodies focused on AI, and its work includes evaluating models from leading AI companies.