✦   ✦   ✦

breaking news

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

AI从幼儿园开始,谁被落下了AI From Kindergarten, and Who Gets Left Behind

科技 结构层 · 文化层 Rest of World ↗ 2026-11-01 § 链接
新闻摘要阿联酋成为首个将人工智能列为K-12必修课的国家,学校以培养AI素养为主,而非直接让幼儿自由使用生成式AI。政府计划培训2.2万名教师。与此同时,纽约市和挪威等地对低龄学生使用生成式AI设限。
The newsThe UAE has become the first country to make AI a required K-12 subject, teaching AI literacy rather than giving young children free access to generative AI. The government plans to train 22,000 teachers, while New York City and Norway have restricted generative AI for younger students.
把AI塞进幼儿园,买的不是能力,是评价方式
Selling AI to toddlers isn't about ability. It's about who measures it.

阿联酋成了第一个把AI列为K-12必修课的国家。报道把这件事写成一个教育选择:保护童年,还是怕孩子落后。这个二选一本身是假的。真正的问题不是孩子几岁接触AI,而是这套课程改变的是孩子的能力,还是评价孩子的方式。

报道里最有说服力的一段来自AI Safety Asia的Edward Tsoi:孩子需要在一句难句子前挣扎,坐在一道数学题里,写出一份笨拙的初稿,这些挣扎才长出后来评估AI的推理能力。如果AI在底层技能还没形成之前就把这段挣扎短路,学生得到的是对工具的熟练,底下没有肌肉。这句话把整件事说透了。阿联酋的课程设计得再谨慎——一到三年级不独立使用生成式AI,老师扮机器人让孩子练习下指令,给孩子看AI出错的例子让他们挑错——它教的仍然是“如何与工具相处”,而不是“如何思考”。前者可以教,后者只能长。

报道里没有出现的,是这套课程要为什么服务。迪拜KHDA有十年路线图,目标是“个性化学习”;政府要培训22000名教师,用AI做教学、评估、课程分析和备课。评估和备课,是学校里最能定义“什么是好学生”的两个环节。当这两件事交给AI,改变的不是孩子学到了什么,是孩子的表现被怎样计量、被谁计量。阿联酋的师生比是1比2,肯尼亚是68比1,报道自己引用的数字。一个资源过剩的系统可以负担“素养教育”的奢侈,一个资源紧缺的系统只能拿AI填坑。同一条政策,在两种结构里是两件完全不同的事。

家长那段值得单独看。几位家长说没人咨询过他们,但都支持。支持的理由很诚实:我在IT行业工作,我亲眼看到AI会怎样改变职场。这是最赤裸的最优解表达——不是我相信AI对孩子好,是我不相信不学AI的孩子将来能活。这种恐惧是真实的,也是被制造出来的。制造它的不是阿联酋教育部,是全球劳动力市场那套“不适应就被淘汰”的叙事。纽约市给幼儿园到八年级的生成式AI踩了一年刹车,挪威近乎禁止小学生使用,理由也是孩子的心理健康。两边都在替孩子做决定,一边怕落后,一边怕上瘾,没有一边问孩子自己想要什么。

这不代表阿联酋做错了。把AI当识字课而不是技术训练,让老师扮机器人教孩子下精确指令,用AI给有阅读困难的学生简化文本、给英语学习者翻译材料——这些是具体的、能缩小差额的操作。一个诵读困难的孩子因此读懂了课文,这是direct层的改善,不是PPT。问题在于,这些好处被包在一个更大的叙事里:AI是未来,不学就落后。这个叙事一旦成立,就没有人再问“不学AI的孩子到底失去了什么”,因为答案已经被预设了。

报道里唯一没被回答的问题是数据。一位在迪拜土地局工作的家长问得很具体:用了哪些工具,孩子和AI互动的时间与和老师同学互动的时间各占多少,提示词存在哪里,孩子能不能绕过安全限制。学校说他们有防火墙、有监控、有学校批准的App清单。但一个六岁孩子对AI说的每一句话都被记录下来,这件事本身没有人解释。训练教师用AI做评估,意味着孩子的作业、表现、错误,都会成为模型的食物。谁拥有这些数据,数据保留多久,孩子长大后能不能要求删除——报道没写,学校没说,政策框架还在“开发中”。

把AI塞进幼儿园,最容易被忽略的代价不是屏幕时间,是解释权的转移。过去一个孩子做错一道题,老师知道他错在哪一步;现在AI告诉他错了,老师和家长看到的是一个分数和一个建议。孩子学到的是“工具说我错了”,而不是“我为什么错了”。前者是服从,后者是思考。阿联酋的课程设计者显然知道这个区别,所以他们反复强调“不要被动使用”“要问更好的问题”。但一套以评估效率为卖点的系统,最后一定会奖励那些给出标准答案的孩子,而不是那些问出好问题的孩子。

报道结尾引用Tsoi的话:不同政府有不同的价值赌注,运行在不同的结构约束下。阿联酋赌的是集中、有钱、小规模系统能跑通一条国家指令。这个赌注可能赢。但赢的定义是什么,由谁来定,报道没问。22000名教师会被培训,170多所私立学校会拿到框架,孩子会在VR里走进金字塔。然后呢?谁来评估这场实验的成败——是AI,还是那些被AI评估的孩子?

The UAE is the first country to make AI a required K-12 subject. The story frames this as an education choice: protect childhood, or fear your kids fall behind. That binary is fake. The real question isn't what age a child meets AI. It's whether the curriculum changes what a child can do, or changes how a child gets measured.

The sharpest line in the piece comes from Edward Tsoi of AI Safety Asia: children need to struggle with a difficult sentence, sit with a math problem, write a clumsy first draft — that struggle is what builds the reasoning capacity they'll later need to evaluate AI. Shortcut the struggle before the underlying skill exists, and students get fluency with the tool and no muscle underneath. That's the whole thing. However carefully the UAE designs its courses — no independent generative AI for grades one through three, teachers pretending to be robots so kids practice giving precise instructions, showing children AI mistakes and asking them to spot what's wrong — it still teaches how to get along with a tool, not how to think. The first can be taught. The second only grows.

What's missing from the story is what this curriculum is for. Dubai's KHDA has a 10-year roadmap toward "personalized learning." The government plans to train 22,000 teachers to use AI for teaching, assessment, curriculum analysis, and lesson planning. Assessment and lesson planning are the two places where schools define what a good student is. Hand those to AI and what changes isn't what kids learn. It's how their performance gets counted, and by whom. The UAE has a student-teacher ratio of two to one. Kenya's is 68 to one — both numbers from the piece. A system with surplus resources can afford the luxury of "literacy education." A system with scarce ones uses AI to plug holes. The same policy is two different things in two different structures.

The parents deserve their own paragraph. Several told the reporter no one consulted them, and all of them supported it anyway. Their reason was honest: I work in IT, I see firsthand how AI will change the workplace. That's the most naked optimal expression there is — not I believe AI is good for my child, but I don't believe a child who doesn't learn AI will survive. The fear is real. It's also manufactured. Not by the UAE ministry of education — by the global labor market's adapt-or-die narrative. New York City put a one-year moratorium on student-facing generative AI from preschool through grade eight, citing children's mental health. Norway has a near-ban for primary school kids. Both sides are deciding for children. One fears falling behind, one fears addiction. Neither asks the children what they want.

None of this means the UAE is wrong. Treating AI as literacy rather than technical training, having teachers role-play robots to teach precise instructions, using AI to simplify texts for students who struggle with vocabulary and translate material for English learners — these are concrete operations that shrink a gap. A dyslexic child who finally reads the assignment is a direct-layer improvement, not a slide deck. The problem is that these benefits come wrapped in a bigger narrative: AI is the future, and not learning it means falling behind. Once that narrative holds, no one asks what a child who doesn't learn AI actually loses, because the answer is already assumed.

The one question the story never answers is data. One parent, who works at the Dubai Land Department, asks it precisely: which tools are used, how much class time goes to AI versus teachers and peers, where are prompts stored, can children accidentally bypass safety restrictions. The schools say they have firewalls, monitoring, approved app lists. But every sentence a six-year-old says to an AI is recorded, and nobody explains that. Training teachers to use AI for assessment means children's homework, performance, and mistakes become model food. Who owns that data, how long is it kept, can a child demand deletion at eighteen — the story doesn't say, the schools don't say, the policy framework is still "in development."

The cost of putting AI in kindergarten that gets missed isn't screen time. It's the transfer of interpretive authority. When a child got a math problem wrong, the teacher knew which step went wrong. Now AI says it's wrong, and the teacher and parent see a score and a suggestion. The child learns the tool says I'm wrong, not why I'm wrong. The first is compliance. The second is thinking. The UAE's curriculum designers clearly know the difference — hence the repeated insistence on not being passive users, on asking better questions. But a system whose selling point is assessment efficiency will, in the end, reward the children who produce standard answers, not the ones who ask good questions.

The piece closes with Tsoi: different governments have different value bets, operating under very different structural constraints. The UAE is betting that a small, centralized, well-resourced system can run a national mandate. That bet might win. But what counts as winning, and who decides, the story never asks. 22,000 teachers will be trained, 170-plus private schools will get the framework, children will walk into pyramids in VR. Then what? Who evaluates whether the experiment worked — the AI, or the children the AI evaluated?