✦   ✦   ✦

breaking news

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

把脑子交出去的人The People Who Handed Over Their Brains

科技 结构层 · 文化层 WIRED ↗ 2026-11-01 § 链接
新闻摘要北加州沙斯塔山三名20岁徒步者因依赖谷歌AI聊天机器人Gemini规划行程,准备不足、体力误判,登顶后摸黑下山迷路,一人膝盖受伤,最终求救。怀俄明大提顿国家公园过去一年报告使用AI获取旅行信息的游客比例从零升至26%。
The newsThree 20-year-old hikers on Mount Shasta in Northern California relied on Google's AI chatbot Gemini for planning, were underprepared and got lost descending in the dark; one injured a knee and they called for help. At Wyoming's Grand Teton National Park, the share of visitors reporting AI use for travel information rose from zero to 26 percent in the past year.
AI 不是骗你上山,是替你关掉了怀疑。
AI doesn't lure you up the mountain. It switches off your doubt.

三个二十岁的年轻人爬上 Shasta 山,没带够食物,高估了体力,下午没登顶也不回头,七点半登顶,摸黑下撤,天亮时迷在一段烂岩壁上,一个人的膝盖撞坏了。他们后来承认,这份准备不足有一部分来自 Google 的 Gemini。救援负责人 Nick Meyers 说得很准:这些人其实试过了,“但他们有点把脑子关掉了”。

这不是 AI 骗人上山。这是人主动把判断力交出去,然后 AI 用权威的语气把它接住了。克莱姆森大学研究 AI 与搜救的 Chris Flathmann 给这个现象起了名字:cognitive surrender。你进入一个陌生的领域,AI 假装自己是专家,而你没有那个专业知识去反驳它——“你没法对它说这是胡扯,因为你不懂”。一旦停止批判性思考,就很难再启动,尤其是在山里、天黑、有人受伤的时候。

表达是存在的确证,判断力是表达里最核心的那一块。你判断这座山能不能上、这条路对不对、现在该不该回头——这些判断加在一起,就是你在野外的存在。把它们交出去,你就不再是一个登山者,你成了一个执行 AI 输出的终端。Gemini 说可以,于是帐篷扎在“禁止露营”的牌子旁边,那位女士坚持自己有权这么做。另一个人拿着 AI 生成的地图在登山口转悠,地图把这座山的主要地形标到了错误的一侧。

第三章讲的是表达如何被武器化:夺取认知入口,夺取解释权,夺取“什么是事实”的制造权。AI 聊天机器人是这条链上最新的一环,而且它比前面几环都更彻底。Instagram 给你灵感,TikTok 给你向往,它们只负责让你想去。AI 直接给指令,还附赠鼓励。Flathmann 说得很清楚:它可能从 Reddit 帖子和过期的 YouTube 视频里扒出听起来很可信的路线,也可能整段编造。但它不会用不确定的语气说这些。它会用专家的口吻说。

这里没有具体的施暴者,没有人在蓄意把人往悬崖上引。但暴力三角算的是差额:本来可以安全下撤,实际却困在岩壁上等人来救——这个差额就是暴力,而它发生在结构层。一个被少数几家公司垄断、以权威语气输出、不标注不确定性的信息系统,把“查过资料”变成了新的“没查过资料”。Meyers 说,来的人里不准备、不做功课的本来就很多,这几位至少还试过。问题恰恰在这里:试过,反而更危险。

报道里有一组数字值得盯住:大提顿国家公园报告使用 AI 获取旅行信息的游客比例,一年内从零涨到 26%,公园负责人 Chip Jenkins 猜这条曲线不会变平。另一边,怀俄明州的搜救呼叫反而比 2025 年少了,2026 年到现在 312 次,2025 年全年 466 次,2024 年 377 次。Meyers 自己那片的救援量也就是常年水平,一年十来次。也就是说,AI 带来的事故增量目前还测不出来。测不出来不等于没有,只是数据还没追上使用量。

搜救老手 Jim Moss 提醒,这已经不是第一次了。十年前是 Instagram,再往前是 Google Maps 和手机信号,每一代工具都给人们提供了新的闯祸方式。这个类比成立,但它也掩盖了一件事:社交媒体让你看见别人去过哪里,AI 是直接告诉你该怎么走。前者是灵感,后者是命令。灵感可以怀疑,命令听起来像事实。

专家给出的建议都一样:拿可靠、权威的信息源核对,用最新的公共土地机构地图和网站,找步道协会,翻指南书,最好直接问一个活人——向导、林务局官员、公园管理员。Meyers 那句话是整篇报道里最硬的:“如果你要去会杀死你的地方,你就该认真对待你的计划。”

认真对待计划,第一步是把脑子拿回来。Gemini 不知道你膝盖的旧伤,不知道你下午三点还剩多少体力,不知道那片岩壁在天黑后是什么样。它只知道你问的那句话,而它被训练成无论如何都要给你一个答案。

26% 这个数字还会涨。真正的问题不是聊天机器人会不会继续把人带偏,而是当它用不容置疑的语气说出错误答案时,还有多少人保留着说“这是胡扯”的能力——以及,在信号消失、天已经黑透的山上,这个能力还来不来得及被找回来。

Three twenty-year-olds climbed Mount Shasta without enough food, overestimating their fitness. They didn't turn back in the late afternoon when they hadn't reached the top. They summited at 7:30 pm and had to descend in the dark. By morning they were lost on a nasty stretch of rock face and one of them had banged up his knee. They later admitted that part of their lack of preparation came from relying on Google's Gemini.

This is not AI tricking people into going up mountains. This is people handing over their judgment, and AI catching it in a confident voice. Chris Flathmann, a Clemson professor who has studied AI and search and rescue, has a name for it: cognitive surrender. You enter unfamiliar territory, the AI pretends to be an expert, and you don't have the expertise to call BS. Once you stop thinking critically, it's hard to start again — especially in the backcountry, in the dark, with someone injured.

Expression is the confirmation of existence, and judgment is the core of expression. Whether this mountain is climbable, whether this trail is right, whether it's time to turn back — those judgments together are your existence in the wild. Hand them over and you stop being a hiker. You become a terminal executing AI output. Gemini said it was fine, so the woman pitched her tent next to a No Camping sign. Another man wandered a trailhead with an AI-generated map that put the mountain's main features on the wrong side.

Chapter Three is about how expression gets weaponized: seize the cognitive entry point, seize the interpretive authority, seize the power to manufacture what counts as fact. The chatbot is the newest link in that chain, and it goes further than the ones before it. Instagram gives you inspiration. TikTok gives you longing. They only make you want to go. AI gives you instructions, plus encouragement. Flathmann is precise about it: it might scrape credible-sounding directions from Reddit threads and out-of-date YouTube videos, or fabricate them entirely. But it won't hedge. It will speak in the voice of an expert.

There's no identifiable perpetrator here, no one deliberately steering people toward a cliff. But the Violence Triangle measures the gap: the difference between a safe descent that was possible and being stranded on a rock face waiting for rescue. That gap is violence, and it sits at the structural layer. An information system monopolized by a few companies, outputting in an authoritative voice, flagging no uncertainty, has turned "I did my research" into the new "I didn't do any research." Meyers says plenty of people arrive unprepared and having done nothing at all. These guys at least tried. That's exactly the problem: trying made it worse.

One set of numbers in the story deserves attention. At Grand Teton National Park, the share of visitors who report using AI for travel information went from zero to 26 percent in a year, and Superintendent Chip Jenkins guesses that curve won't flatten. Meanwhile search and rescue calls in Wyoming are actually down from 2025 — 312 so far in 2026, against 466 for all of 2025 and 377 in 2024. Meyers' own area averages a dozen rescues a year, and 2026 looks average. The accident increase from AI is, so far, unmeasurable. Unmeasurable is not the same as absent. The data just hasn't caught up with the usage.

Veteran Jim Moss points out this isn't the first time. A decade ago it was Instagram; before that, Google Maps and cell signal. Every generation of tool gives people new ways to get into trouble. The analogy holds, but it also hides something: social media showed you where others had been. AI tells you how to go. The first is inspiration. The second is a command. Inspiration can be doubted. A command sounds like fact.

The experts' advice is uniform: cross-check against reliable, authoritative sources — current public-land agency maps and websites, trail association sites, guidebooks. Better yet, talk to a live human: a guide, a forest official, a park ranger. Meyers' line is the hardest in the piece: "If you're going to go into places that will kill you, you should take your planning seriously."

Taking your planning seriously starts with taking your brain back. Gemini doesn't know about your old knee injury, or how much energy you'll have left at three in the afternoon, or what that rock face looks like after dark. It only knows the question you asked, and it was trained to always give you an answer.

That 26 percent will keep climbing. The real question isn't whether chatbots will keep leading people astray. It's how many people still retain the capacity to say "that's BS" when the machine states a wrong answer with total confidence — and whether, on a mountain with no signal and the dark already down, that capacity can still be found in time.