Meta押注 AI Agent Muse，力图追赶 AI 竞赛对手

Meta 正通过个人助手 Muse 再次尝试将人工智能带给大众。该公司称，Muse 能让几乎任何人都用上 AI。该产品是 Meta 一项数十亿美元战略调整的最新一步，旨在扭转公司在 AI 竞赛中落后的局面，并追赶 OpenAI、Anthropic 和 Google 等竞争对手。

Muse 是一款“个人 AI Agent”，旨在帮助用户处理网购、发送电子邮件和规划旅行等日常任务及项目。Meta 表示，用户给出目标后，Muse 可以自主工作，包括打开浏览器、填写表单，甚至代表用户进行协商。对于耗时较长的任务，Meta 称用户关闭应用后 Muse 仍会在后台继续工作；如果情况发生变化，或需要用户批准某项操作——例如进行购买——它会重新联系用户。

Muse 将在美国通过 iOS、Android 和 muse.ai 推出，AI 眼镜版本则“即将上线”。该助手对大多数用户免费，但 Meta 表示将为“想做更多事情的人”提供付费订阅。该公司未说明这些付费套餐的具体形式和价格，也未概述免费用户将面临哪些限制。

“Muse 是第一步：一个承担更多工作的 Agent，让人们能够专注于对自己重要的事情。”Meta 表示。

1/4 记得在放映期间关闭你的 Agent。 图片：Meta

尽管 Meta 将 Muse 宣传为“为所有人打造的全球首个个人 AI Agent”，但该产品实际上与竞争对手提供的日益增多的 AI 助手、队友和同事十分相似。大多数此类工具确实面向企业或具备技术背景的个人，尤其包括 OpenAI 的 ChatGPT Work、Anthropic 的 Claude Cowork、Microsoft 的 Copilot Tasks、SpaceXAI 的 Grok Bot，以及爆红的开源 Agent Moltbot。不过，Google 最近发布的 Gemini Spark 助手似乎也瞄准了类似的用户群体。

易用性是 Meta 希望 Muse 脱颖而出的地方。Meta 声称：“任何人开箱即可使用，无需技术经验”，并强调“没有学习曲线”。沟通方式也很简单：用户可以通过 Muse 应用或 Meta 旗下消息平台 WhatsApp 与助手聊天。Meta 表示，即使用户仅提及一次，Muse 也会轻松记住用户分享的细节，从而作出更好的主动建议。

Meta 同样热衷于强调 Muse 的隐私和安全功能。要让 Muse 真正成为实用的 AI 助手，获得用户信任至关重要，因为它需要掌握每位用户一定程度的个人信息才能有效运作。为此，Meta 表示，用户可以选择不将与 Muse 的互动用于训练 Meta 的 AI 模型（目前尚不清楚该选项默认开启还是关闭），也可以指示 Muse“忘记”其学到的特定信息。

鉴于 Agent 能够访问个人信息以及密码、支付信息等凭证，它们天然可能成为安全和隐私噩梦。为保护数据、密码和其他信息，Muse 运行在云端虚拟计算机中，该环境旨在确保数据安全，并与其他用户的 Agent 隔离。Meta 表示，Muse 无法查看密码或支付方式；另一个名为 Sentinel 的 AI Agent 会在同一虚拟机内进行监控，确保 Muse 执行的任何操作未经批准都无法访问互联网。

Meta 表示，计划在今年晚些时候推出更安全的“机密”版虚拟机，通过加密确保即使 Meta 自身也无法访问其中内容。该公司还称，1Password 和 Shop Pay 支持也在筹备中，而 Muse 已可以通过 Stripe 的 Link 安全结账。

由公司自研模型 Muse Spark驱动的 Muse，被定位为 Meta 未来 AI 战略的核心。此前该公司经历了多年挫折和失败，如今正试图重新夺回阵地。

作为家喻户晓的品牌和社交媒体巨头，Meta 具备触达大量普通用户的良好条件。但该公司也有不少历史包袱需要克服。在信任层面，它刚刚经历了一场围绕社交媒体危害的艰苦审判，过去还面临过多次严重的隐私风波，其中包括剑桥分析丑闻。其 AI 推进也屡有失误，包括一项“Discover”功能展示了其他用户的提示词、对话和图像输出；一款支持聊天机器人帮助黑客接管了逾 2 万个 Instagram 账户；以及一个短暂上线的AI 生成点击诱饵新闻流，其行为似乎违反了 Meta 自己的 AI 内容政策。此外，Meta 近期还不幸获得了“变态眼镜”的名声。

Meta is making another push to bring artificial intelligence to the masses with Muse, a personal assistant it says can put AI in the hands of virtually anyone. The product is the latest step in a multi-billion-dollar strategy overhaul designed to revitalize the company’s ailing position in the AI race and help it catch up to rivals like OpenAI, Anthropic, and Google.

Muse is a “personal AI agent” designed to help out with everyday tasks and projects, like online shopping, sending emails, and planning a trip. Once given a goal, Meta says Muse can work on its own, opening a browser, filling out forms, and even negotiating on users’ behalf. For lengthy tasks, Meta says Muse will continue working in the background after users close the app, returning to users if something changes or if it requires approval for something, such as making a purchase.

Muse will roll out in the US on iOS, Android, and muse.ai, with AI glasses “coming soon.” The assistant will be free for most users, though Meta said there will be paid subscriptions “for people who want to do more.” The company did not specify the nature and costs of these paid plans, nor outline what limits there are for free users.

“Muse is a first step: an agent that takes on more of the work so people can focus on what matters to them,” Meta said.

1/4 Remember to switch your agent off during screenings. Image: Meta

While Meta is pitching Muse as “the world’s first personal AI agent built for everyone,” the product is, in fact, very similar to the growing array of AI assistants, teammates, and coworkers on offer from the company’s competitors. Most tools are indeed aimed at businesses or technically-minded individuals — notably OpenAI’s ChatGPT Work, Anthropic’s Claude Cowork, Microsoft’s Copilot Tasks, SpaceXAI’s Grok Bot, and the viral open-source agent Moltbot — though Google does appear to be targeting a similar audience with its recently-announced Gemini Spark assistant.

Ease of use is where Meta hopes Muse will stand out from the crowd. “Anyone can use it out of the box, no technical experience required,” Meta claims, boasting that “there’s no learning curve.” Communication is simple too, as easy as messaging the assistant through the Muse app or WhatsApp, one of Meta’s messaging platforms. It will also readily remember details users share, even if only mentioned once, Meta says, enabling it to make better and unprompted suggestions.

Meta is similarly keen to emphasize Muse’s privacy and security features. Securing trust will be key for making Muse actually useful as an AI assistant, as it will require a certain amount of personal information about each user to function effectively. With this in mind, Meta says people can opt out of their interactions being used to train Meta’s AI models (it’s not known whether this is on or off by default) and can instruct Muse to “forget” specific things it has learned about them.

Given their access to personal information and credentials like passwords and payment information, agents can naturally be a security and privacy nightmare. To keep data, passwords, and other information secure, Muse runs on a virtual computer in the cloud designed to keep data safe and separate from other users’ agents. It has no visibility into passwords or payment methods, Meta says, and another AI agent, Sentinel, patrols the same virtual machine to ensure nothing Muse does can reach the internet without approval.

Meta says it plans to introduce a more secure “confidential” version of the virtual machine later this year, encrypted so that even Meta cannot access it. 1Password and Shop Pay support are also on the way, the company said, adding that Muse can already checkout securely with Stripe’s Link.

Muse, powered by the company’s in-house model Muse Spark, is being positioned as the centerpiece of Meta’s AI strategy going forward as the company seeks to regain ground after years of setbacks and failures.

Given its position as a household name and social media behemoth, Meta is in a good position to reach many of these more casual users. But there is also no shortage of baggage for Meta to overcome. Trust-wise, it has just endured a grueling trial over the harms of social media and has faced numerous bruising privacy-related ordeals in the past, including the Cambridge Analytica scandal. Its AI overtures have also been plagued with missteps, including a “Discover” feature that displayed prompts, conversations, and image outputs from other users, a support chatbot that helped hackers take over more than 20,000 Instagram accounts, and a short-lived AI-generated clickbait news feed that appeared to break its own policies on AI content. There’s also its recent and unfortunate reputation for “pervert glasses.”

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

• Robert Hart

• AI

• Meta

• News

• Tech