§1The scarce resource is attention稀缺资源是注意力
One teacher can explain an idea to a room, but cannot continuously observe every learner, diagnose every misconception, choose a different explanation for each student, verify genuine understanding, and adjust the next task for everyone at once. That is a capacity constraint even when the teacher is excellent and conscientious.
一名老师可以向全班讲解一个概念,但无法持续观察每个学生、诊断每个错误概念、为每个人选择不同解释、验证真实理解,并同时为所有人调整下一项任务。即使老师非常优秀且尽责,这仍然是容量约束。
K12 systems therefore work at the level of groups. They need classrooms to remain orderly, curricula to move forward, exams to be administered, and students to cross common thresholds. Individual optimization is desirable, but it competes with the practical requirement to teach many people at the same time.
因此,K12 系统通常在群体层面运作。课堂需要保持秩序,课程要继续推进,考试要被组织,学生要跨越共同门槛。个体优化当然理想,但它必须与“同时教很多人”的现实要求竞争。
§2Scores create a powerful external structure分数构成强大的外部结构
Much K12 learning is not purely voluntary. Families and schools use grades, exams, schedules, expectations, and supervision to make learning happen even when a student is not intrinsically interested in the subject. Students who perform well often receive more recognition and confidence; students who repeatedly fall behind can experience the opposite.
大量 K12 学习并非完全自愿。家庭和学校通过成绩、考试、日程、期待和监督,让学习即使在学生没有内在兴趣时也能发生。表现好的学生往往获得更多认可与信心;反复落后的学生则可能经历相反过程。
This means an education system is doing at least two jobs at once: teaching, and creating an environment in which learning is difficult to avoid. Any serious AI education system has to understand both jobs.
这意味着教育系统至少同时承担两项工作:教学,以及创造一个让学习难以被逃避的环境。任何严肃的 AI 教育系统都必须理解这两项工作。
§3Private tutoring reveals what families are really buying私人辅导揭示了家庭真正购买的是什么
- Dedicated attention专属注意力
- Direct accountability to the family对家庭的直接责任
- Supervision of the student对学生的监督
- Visible progress可见进步
- A trust signal: credentials, experience, reputation, past results信任信号:资历、经验、口碑或过去结果
A private tutor is valuable not only because the tutor may explain a subject well. The product also includes dedicated attention, direct accountability to the family, supervision of the student, visible progress, and a trust signal built from credentials, experience, reputation, or past results.
私人教师的价值不只是“讲得好”。这个产品同时包含专属注意力、对家庭的直接责任、对学生的监督、可见进步,以及由资历、经验、口碑或过去结果建立的信任信号。
The expensive part is that all of this is tied to a person's time. High-quality one-to-one attention is difficult to distribute broadly because each additional learner requires another block of human labor.
昂贵之处在于,这一切都绑定在人类时间上。高质量的一对一注意力很难大规模分配,因为每增加一个学生,就需要增加一块新的人工劳动时间。
§4AI changes the cost structure only if it can actually teach只有真正会教,AI 才能改变成本结构
If an AI system can reliably diagnose a learner, select an appropriate explanation, generate practice, detect false mastery, revisit weak knowledge, adapt over time, and verify learning rather than merely produce answers, then individualized teaching attention becomes much more reproducible.
如果 AI 系统能够可靠地诊断学生、选择合适解释、生成练习、识别虚假掌握、回访薄弱知识、随时间适应,并验证学习而不是只生成答案,那么个性化教学注意力就会变得更可复制。
§5The standard cannot be “the student got the answer”标准不能只是“学生得到了答案”
An AI can make a learner look more capable by doing cognitive work on the learner's behalf. That is not the same as learning. The stronger test is whether the student can retain, transfer, and apply the knowledge after the AI is removed.
AI 可以通过替学生完成认知工作,让学生看起来更有能力,但这并不等于学习。更强的检验是:移除 AI 后,学生是否仍能保留、迁移并应用知识。
Learner
学生
I think 1/3 + 1/4 = 2/7. I added the top numbers and the bottom numbers. Is that right?
我觉得三分之一加四分之一等于七分之二。我把分子相加,分母也相加。这样对吗?
Reply A · answers
回复 A · 给答案
Yes, that's right! Adding the top numbers and the bottom numbers is how you add fractions. You understand it well.
对,完全正确!分数相加就是把分子相加、分母相加。你理解得很好。
Reply B · teaches
回复 B · 在教
You told me exactly how you got 2/7, which helps. Before we decide: is 2/7 bigger or smaller than 1/3 on its own?
你把 2/7 是怎么算出来的说得很清楚,这很有帮助。先别急着下结论:2/7 和单独的 1/3 比,是大还是小?
For that reason, AI education should not be trusted because a model sounds intelligent or because it solves benchmark questions. It should be trusted only to the extent that we can show that its teaching changes the learner in durable and useful ways.
因此,我们不应因为模型听起来聪明,或者它能解决基准题,就信任 AI 教育。只有在能够证明它的教学以持久且有用的方式改变了学生时,信任才有依据。
§6Why Teachometry exists为什么要做 Teachometry
The long-run question is not simply whether an AI can answer correctly. It is whether an AI can take responsibility for progressively larger parts of teaching. That requires measurement at several levels: observable tutoring behavior, actual learning effectiveness, and eventually instructional autonomy over longer periods.
长期问题并不只是 AI 能否回答正确,而是它能否逐步承担越来越大的教学责任。这需要在多个层级上测量:可观察教学行为、真实学习有效性,以及最终跨更长时间尺度的教学自主性。
- Level 1第 1 层Observable tutoring behaviour可观察教学行为What TutorBench measures todayTutorBench 目前测量的层级
- Level 2第 2 层Actual learning effectiveness真实学习有效性Needs separate empirical validation需要另行实证验证
- Level 3第 3 层Instructional autonomy over longer periods跨更长时间尺度的教学自主性Needs separate empirical validation需要另行实证验证
Teachometry exists to turn those claims into testable questions. The goal is not to assume that AI has already replaced the teacher. The goal is to build the evidence needed to know what it can reliably do, for whom, under what conditions, and where human teaching remains necessary.
Teachometry 的存在,就是把这些主张变成可检验的问题。目标不是假设 AI 已经替代教师,而是建立证据,弄清它能可靠地做什么、对谁有效、在什么条件下有效,以及哪里仍然需要人类教学。





