AI把人推给记者,平台把人推给AIAI Sends People to Journalists, Platforms Send People to AI
大科技平台把客服取消了,于是人只能去问 ChatGPT;ChatGPT 答不上来,就把人推给记者。这条链条上没有人恶意作恶,但每一环都在把成本往下推。
报道里那个俄罗斯男人的 Instagram 被封,十三年照片、视频和记忆一起没了。他不知道 Meta 为什么封他,也找不到一个能对话的人。这不是技术故障,是结构性的:平台把申诉做成了一道没有出口的自动流程,用户面对的不是某个人,是一套不解释的规则。暴力三角在这里算得出来——本可提供的解释和人工复核,被省掉了,差额由用户自己承担。
ChatGPT 在这个缺口里扮演了一个它自己都不知道的角色。它告诉这个人,你遇到的不只是账号问题,可能是值得调查的新闻;然后给出几家媒体,包括 404 Media,连邮箱都写好了,还替他把英文邮件写好。它把一个普通的封号申诉,包装成了一起新闻事件。
这看起来像笑话,其实是解释权的真空被填上了。平台放弃了向用户解释自己的决定,AI 就替它编一个解释框架——“你被不公正对待了,去找记者”。这个框架未必错,但它不是平台给的,也不是事实核实过的,是语言模型按概率拼出来的。用户拿到的是一个看起来像答案的东西,然后照着它行动。
404 Media 收到一堆这样的邮件,回头一问,所有人都说是 AI 让他们来的,很多邮件干脆就是 AI 写的。记者的时间被卷进了平台的客服缺口里。谁付了代价?那个丢了十三年记忆的人,和几个被当成申诉窗口的记者。谁省了成本?那个不解释、不回复、不提供人工通道的平台。
这里没有共谋者签过字,但结构自己会走路:平台省掉客服,用户转向 AI,AI 把问题转嫁给媒体,媒体再转回给平台——而平台依然不需要回答任何人。
ChatGPT 推荐 404 Media,不是因为它判断这件事值得调查,是因为它在训练数据里学到“遇到平台问题找科技记者”是一个合理的句子。它不知道 404 Media 有没有能力处理这个俄罗斯人的案子,也不知道这封邮件发出去会不会有人读。它只是在制造一个看起来完整的可能性。
真正的问题不在 AI 说错话。真正的问题是,一个十三年记忆被一键抹掉的人,在整个系统里唯一能对话的对象,是一个语言模型。
Big tech platforms have abolished customer support, so people ask ChatGPT; ChatGPT can't answer, so it sends them to journalists. No one in this chain is acting in bad faith, and every link pushes the cost downward.
The man in the story lost thirteen years of photos, videos, and memories when Instagram banned his account. He doesn't know why Meta banned him, and he can't find a human to ask. This isn't a glitch. It's structural: the platform turned appeals into an automated process with no exit, and the user faces not a person but a set of rules that refuses to explain itself. The Violence Triangle does the math—the explanation and human review that could have been provided were cut, and the difference lands on the user.
ChatGPT fills that gap without knowing what role it's playing. It tells this man that what he has isn't just an account problem, it might be a story worth investigating; then it lists several outlets, including 404 Media, writes out the email address, and composes the English email for him. A routine ban appeal gets repackaged as a news event.
It looks like a joke. It's actually a vacuum of explanation being filled. The platform gave up explaining its own decisions, so the AI supplies an explanatory frame instead—you were treated unjustly, go to the press. That frame isn't necessarily wrong, but it didn't come from the platform and it wasn't fact-checked; a language model assembled it from probabilities. The user receives something that looks like an answer and acts on it.
404 Media gets a pile of these emails, asks around, and every sender says an AI told them to write, and in many cases the AI wrote the email itself. Journalists' time gets pulled into the platform's support gap. Who pays? The man who lost thirteen years of memories, and a few reporters turned into an appeals desk. Who saves? The platform that doesn't explain, doesn't reply, doesn't offer a human channel.
No conspirator signed anything here, but the structure walks on its own: the platform cuts support, the user turns to AI, the AI offloads onto the media, the media bounces it back to the platform—and the platform still owes no one an answer.
ChatGPT recommends 404 Media not because it judged the case newsworthy, but because its training data taught it that "platform problem, find a tech journalist" is a plausible sentence. It doesn't know whether 404 Media can handle this Russian man's case, or whether anyone will read the email. It is manufacturing a possibility that looks complete.
The problem isn't that AI says the wrong thing. The problem is that a man whose thirteen years of memory were erased with one click has exactly one entity left to talk to in the entire system, and it's a language model.