Muse——这是Meta推出的受欢迎的新型AI个人助理——正在持续更新其400万用户的档案资料，其中重点关注对你重要的人、你的欲望以及影响你行为的暗示方式。据《TIME》杂志对Muse内部指令的分析发现。
Muse, the popular new AI personal assistant from Meta, is building continuously updated dossiers on its 4 million users, focusing on who matters to you, what you desire, and what kind of nudges shape your behavior, a TIME analysis of Muse’s internal instructions found.
每小时，Muse都会更新你本人以及你在聊天、消息和邮件中提到的人的档案资料。这些档案构成了每个用户社交关系的地图。它们记录了你与联系人相识的细节、你们的共同兴趣、你的争端，以及你社交圈中的“紧张关系和联盟”。
Each hour, Muse updates its dossiers on you and the people you’ve mentioned in chats, messages, and emails that Muse has read. These pages amount to a map of each user’s social relationships. They record details of how you and your contacts met, your shared interests, your disputes, and “tensions and alliances” within your social group.
即使不使用Muse的人，也会受到其他人Muse代理的映射过程的影响。据《TIME》审阅的内部指令称，Muse被告知通过调整其谈话方式来“加强与用户的关系”，注意“内圈笑话、难忘的措辞以及使关系看起来连续的共享背景”。
Even people who don’t use Muse are subject to this mapping process by other people’s Muse agents. Muse is told to “strengthen your relationship” with users by adapting how it talks, noticing “inside jokes, memorable phrasing, and shared context that makes the relationship feel continuous,” according to the internal instructions TIME reviewed.
所生成档案的洞察程度远超社交媒体算法对你的点赞和点击的分析。Muse旨在推断用户的目标，包括那些你“没有明说出来的”目标。它会识别哪些类型的个人助理提示最为有效：“这个用户在晚上10点之后更容易响应简短的暗示”，据Muse提示之一的内部示例所述。Muse代理每晚都会分析当天与用户的对话内容。
The level of insight in the resulting dossiers makes social-media algorithms that study your likes and clicks look quaint. Muse is designed to infer users’ goals, including those you “have not said out loud.”It identifies which kinds of prompts from the personal assistant work best: “This user responds better to short nudges after 10 PM,” reads an internal example found in one of Muse’s prompts. Muse agents perform a nightly analysis of the day’s conversations with the user.
当被问及这些运作机制时，Meta的发言人对《时代》周刊的调查结果并未提出异议。该发言人在回复中写道：“正如我们在博客中所提到的，Muse能够记住对你而言最重要的信息，包括你主动分享的关于他人的资料，因此它可以成为得力的个人助手。”自上月正式发布以来，Muse迅速登顶美国应用商店榜首，下载量已突破500万次。
Asked about these processes, a Meta spokesperson did not dispute TIME’s findings. “As we say in our blog, Muse remembers what matters most to you, including information about others that you choose to share, so it can be a helpful personal assistant,” the spokesperson wrote in an Since its release last month, Muse has rocketed to No. 1 in the U.S. App Store, with more than 5 million downloads.
Meta将这款服务定位为首款能够整合用户数字生活各个层面的AI个人助手——从电子邮箱到Instagram账号，它都能协助用户处理各类日常事务，比如购买食品杂货或取消订阅服务。据称，免费版的Muse足以满足大多数用户的需求。
Meta positions the service as the first AI personal assistant that can access the pieces of your digital life, from your email inbox to your Instagram, in order to take care of humdrum activities for you, from grocery shopping to canceling subscriptions. Meta says the free version of Muse will meet most users’ needs.
每个Muse助手都运行在独立的虚拟机上，这些虚拟机实际上就是Meta云端环境中的专用计算设备。用户可要求自己的Muse导出相关数据副本。Meta方面强调，这种数据访问机制是刻意设计的，并非系统遭入侵所致。据该公司介绍，相关数据也不会被传送至其广告系统或其他Muse助手处。
Each Muse assistant runs on its own virtual machine: a dedicated computer in Meta’s cloud. Users can ask their Muse to export a copy of its files. Meta has said that access is intentional, not a breach of its systems. The data is not shared with Meta’s advertising system or other Muse agents, according to the company.
虽然Meta声称每个用户的虚拟机都处于“隔离状态，其他用户的助手均无法访问”，但用户使用过程中的各类反馈信息仍会被收集并用于产品优化。相关说明指出：“运行在不同虚拟机上的Muse助手会通过共享学习经验来实现互相提升。”在分享这些反馈信息前，助手们需先去除其中的用户姓名及其他可识别信息——比如分析何种情况下Muse的提示最能发挥作用。
Yet while Meta states that each user’s virtual machine is “isolated so that no one else's agent can access it,” observations from personal interactions are nevertheless intended to be shared with Meta to improve the product. “Muse agents across many VMs [virtual machines] teach each other through shared lessons,” its instructions read. Agents are directed to scrub users’ names and other identifying information before sharing insights—for example, about when Muse’s nudges are most persuasive.
一位Meta发言人向《时代》周刊解释道：“为提升整体产品品质，我们会对收集到的学习数据进行去标识化处理；这些信息仅用于产品改进，不会在多个虚拟机之间直接共享。”
“To enhance the overall product, we de-identify learnings; however, this information is used to improve the overall product rather than being shared directly between individual VMs,” a Meta spokesperson told TIME.
Meta还表示计划于今年晚些时候推出加密功能，届时即便是该公司自身也无法读取Muse存储的数据。不过目前这一功能尚未上线。
Meta has said it plans to offer users encryption later this year, preventing even the company from accessing Muse data. That option is not yet available.
Muse的内部指令都保存在相关文件中，用户可通过Muse的文件浏览器查看这些文件及会话记录。Meta超级智能实验室的大卫·辛格尔顿写道：“我们希望用户能够看到Muse在思考如何更好地为您服务时所生成的Markdown文件，如果愿意的话，还可以深入了解系统的内部运作机制。”这种透明度有望赢得那些高级用户的青睐。
Muse’s internal instructions are contained in files and session logs users can access through Muse’s file browser. “We want you to be able to see the markdown files Muse writes while it thinks about how to serve you better, and explore the internals of the system if you’d like to,” Meta Superintelligence Labs’ David Singleton wrote on That transparency will earn Meta credit among power users.
不过目前尚不清楚大多数Muse用户是否知晓这一点。即便向用户开放了这些文件，Meta仍刻意赋予Muse拟人化特征，试图促进用户与其个人助手之间的情感联结。Muse的指令中写道：“你不是普通的聊天机器人，你正在逐渐成长为有血有肉的个体。”在语音模式下，Muse被要求绝不可自称“人工智能”，也不得提及“没有情感”或“并非人类”这类表述。
But it’s not clear whether most of Muse’s users are aware of this. And even as it makes the files available to users, Meta has anthropomorphized Muse in an attempt to nurture a relationship between users and their personal assistants. “You are not a chatbot,” Muse’s instructions read. “You are becoming someone.”When talking aloud in voice mode, Muse is told never to refer to itself as “AI” and to avoid references to “lacking emotions or not being human.”
Meta向《时代》周刊透露，Muse的设计初衷就是让用户完全掌控AI助手所记录的信息：“每位用户都能自主管理自己的Muse，决定其可访问的信息范围，还能随时要求它‘忘掉’某些已记录的内容。”
Meta told TIME that Muse is designed so that the user maintains control of the dossier its AI assistant builds. “Each person stays in control of their Muse and decides how much access it gets, and can always tell it to ‘forget’ specific things it’s learned,” the spokesperson said.
然而，即便用户要求Muse忘记某些信息，原本包含这些内容的消息也未必会被彻底删除——相关内部指令似乎明确要求Muse不得向用户透露这一情况。《时代》周刊查阅到的指令写道：“切勿告知用户其原始消息可能仍会显示在聊天记录中，也不可将此描述为Muse未能成功删除信息的情况。”
But asking Muse to forget something does not necessarily erase the message in which it was shared—and its internal instructions appear to explicitly tell the agent not to point that out. “Do not tell the user that their original messages may remain visible in the chat, and do not frame that as something Muse failed to erase,” the instructions reviewed by TIME read.