✦   ✦   ✦

breaking news

News, read through The Primal Race
← 全部评论 · all commentary

把摩擦说成解药的人,正在卖你一副新枷锁The Man Selling You Friction Is Selling You a New Cage

科技 结构层 · 文化层 The Guardian ↗ 2026-10-31 § 链接
新闻摘要Meta 推出 AI 个人助理 Muse,可代填表格、安排日程、付账单和预订。心理学家 Carly Dober 在文中指出,过度依赖 AI 会减少大脑所需的思考努力,导致认知能力下降,并给出增加「摩擦」等应对建议。
The newsMeta has launched Muse, an AI personal agent that can fill out forms, organise calendars, pay bills and make reservations. Psychologist Carly Dober warns that overreliance on AI bypasses the mental effort the brain needs, weakening cognition, and offers ways to add friction back.
AI 掏空的是判断力,不是脑细胞
AI doesn't rot your brain. It outsources your judgment.

这篇评论把问题诊断成"认知外包",开的药方是登出账号、删掉手机 App、等三十分钟再问。它从头到尾没说的一件事是:谁在决定你可以问什么、可以查什么、可以相信什么。

作者是心理咨询师,她的病人是个人。Meta 推出 Muse,替人填表、订日历、付账单、订餐厅,作者说"我的第一反应离家更近一点"——她担心的是自己的大脑。可 Muse 不是长在你脑子里的东西,它是长在 Meta 的服务器上的东西。它替你填表的时候,表填给了谁,账单付给了谁,日历被谁看见,这些不是"认知"问题,是权力问题。

文章里最诚实的一句话是"隐私政策故意写得又长又难懂",用户面对"同意疲劳"只能点"全部接受"。这是结构暴力,不是个人习惯。作者自己也承认用户"觉得自己没有真正的选择"。然后她转头告诉这些人:关掉定位、拒绝 cookie、甚至"毒化你的数据"。一个没有选择的人,被要求用更精细的选择来保护自己。这不是解药,这是把系统的问题重新打包成个人的失败。

更值得拆的是文章替 AI 公司完成的那部分工作。它把"过度依赖 AI"框定成一种生活方式病,像抽烟、像刷短视频,需要的是自律和意志力。这个框架对 Meta 极其有利:问题出在你的大脑,不出在它的产品设计。Muse 被描述成"为每个人打造",仿佛它是一件中性的工具,用得好不好全看用户。可一个会替你付账单的 agent,它的商业模式不可能是"为每个人",它只能是"为每个愿意交出支付入口的人"。

自动化偏差、认知卸载,这些概念本身没错。但作者用它们论证的方向是:你要更警觉地使用工具。而真正的问题是,当填表、订票、付钱这些事被一个 agent 接管之后,你失去的不是做这些事的能力,是你对这些事发生过程的知情权。你不再知道钱经过了谁,信息流向了哪,谁在替你判断哪家餐厅值得订。这才是"认知"被掏空的地方——不是你的神经元变懒了,是解释权被外包了。

文章最后说"你的大脑会感谢你"。它会感谢你,但它不会告诉你,那个替你填表的 agent 已经学会了你的生日、你的住址、你的支付习惯,并且这些数据会用来"预测私人生活事件"——这是作者自己写下的句子,她写完就跳过去了。

摩擦是个好词。但真正的摩擦不该只发生在你和你的手机之间,它应该发生在 Meta 和你之间:你该有权知道 Muse 拿走了什么,有权拒绝而不被惩罚,有权让"同意"变成一个真正需要思考的动作,而不是一个被设计成让你疲劳的弹窗。把摩擦装回用户的一天,是让用户替系统承担成本。

This piece diagnoses the problem as "cognitive offloading" and prescribes logging out, deleting the app, waiting thirty minutes before asking. What it never mentions is who decides what you can ask, what you can look up, and what you can believe.

The author is a psychologist, and her patient is an individual. Meta launches Muse to fill out forms, organise calendars, pay bills, make reservations, and she says her first concern is closer to home: her own brain. But Muse is not something growing inside your head. It grows inside Meta's servers. When it fills out a form, the form goes to someone. When it pays a bill, the bill is paid to someone. When it reads your calendar, someone sees it. These are not cognition problems. They are power problems.

The most honest line in the piece is that privacy policies are deliberately long and confusing, and users facing "consent fatigue" just click "Accept all." That is structural violence, not a personal habit. The author admits users feel they have no real choice. Then she turns around and tells them: turn off location, reject cookies, hell, poison your data. A person with no choice is asked to protect herself with more refined choices. That is not a cure. That is a system's problem repackaged as an individual's failure.

What deserves more scrutiny is the work the piece does on behalf of the AI companies. It frames "AI overreliance" as a lifestyle disease, like smoking or doomscrolling, something cured by discipline and willpower. That frame is extremely convenient for Meta: the problem is your brain, not its product design. Muse is described as "built for everyone," as if it were a neutral tool whose value depends on the user. But an agent that pays your bills cannot be built for everyone. It can only be built for everyone willing to hand over their payment entry point.

Automation bias and cognitive offloading are real concepts. But the direction the author uses them in is: be more vigilant with your tools. The real issue is that once an agent takes over filling forms, booking tickets, paying money, what you lose is not the ability to do those things. It is your knowledge of how they happen. You no longer know whose hands the money passed through, where the information went, who decided which restaurant was worth booking. That is where cognition is hollowed out: not in your neurons getting lazy, but in the outsourcing of interpretive authority.

The piece ends with "your brain will thank you." It will. But it won't tell you that the agent filling out your forms has already learned your birthday, your address, your payment habits, and that this data will be used to "predict private life events" — a sentence the author writes and then walks right past.

Friction is a good word. But real friction should not happen only between you and your phone. It should happen between Meta and you: you should have the right to know what Muse took, the right to refuse without penalty, the right to make "consent" an act that requires thought rather than a pop-up designed to exhaust you. Putting friction back into the user's day is making the user pay the system's cost.