中文导读

AI 辅导老师能不能像真人一样教数学？英国一项上千名学生参加的试验给出了“一半肯定、一半保留”的答案。文章先交代试验的基本设计：学生每次做错题，会随机得到提示视频、真人辅导或 AI 辅导中的一种帮助。接着用数据说明，在课堂上 AI 与真人辅导的效果相当，且 AI 的消息大多无需修改；但在更大范围的标准化数学测验中，10 周后并未看到提升。研究者给出的解释是辅导更花时间，学生做的题和覆盖的内容变少了。文章最后提到男女生结果不同、约五分之一学生从不回复等新问题，并指出这些结论仍需更长时间的试验检验。阅读时可留意 however、although、so 如何连接对比与因果，以及 may、seemed、tentative 等词如何表达结论的不确定性。

正文

Can an AI tutor teach maths as well as a person? A new study from Britain gives a mixed answer. On October 29, the London education company Eedi and Google DeepMind published the full results of a trial with 1,179 students in 10 state schools. During lessons, the AI tutor did about as well as expert human tutors. After ten weeks, however, it did not raise scores on a broad maths test.

本段译文

AI 辅导老师能像真人一样教数学吗？英国的一项新研究给出了喜忧参半的答案。10 月 29 日，伦敦教育公司 Eedi 与谷歌 DeepMind 公布了一项试验的完整结果，参与者为 10 所公立学校的 1,179 名学生。在课堂上，AI 辅导老师的表现与真人专家辅导老师大致相当。然而 10 周后，它并没有提高学生在一项综合数学测验中的成绩。

The trial ran from April to July. Once a week, students aged about 12 to 15 practised maths on Eedi’s online platform for around 30 minutes, with a teacher in the room. Whenever a student got a question wrong, they received one of four kinds of help: simple hints and videos, a typed chat with a human tutor, or one of two versions of an AI tutor. Every message the AI wrote was checked by a human before students saw it.

本段译文

试验从 4 月持续到 7 月。学生年龄约 12 至 15 岁，每周一次在 Eedi 的在线平台上练习数学约 30 分钟，教室里有老师在场。每当学生答错一道题，就会得到四种帮助中的一种：简单的提示和视频、与真人辅导老师的文字聊天，或者两种版本的 AI 辅导老师之一。AI 写的每条消息在学生看到之前都由真人检查。

In class, the tutoring clearly helped. After a chat with either an AI or a human tutor, students answered a similar question correctly 56 to 58 percent of the time, compared with 43 percent after hints alone. Human checkers approved about 95 percent of the AI‘s messages without changes. No harmful content reached any student, although seven AI messages with maths mistakes did.

本段译文

在课堂上，辅导明显有帮助。与 AI 或真人辅导老师聊过之后，学生答对类似题目的比例为 56% 到 58%，而只看提示后这一比例为 43%。真人审核者对 AI 约 95% 的消息未作修改就予以通过。没有有害内容送到任何学生面前，不过有七条含数学错误的 AI 消息到了学生那里。

The bigger test told a different story. Students took Star Maths, a standard test, before and after the trial. Those who used the AI tutor scored about the same as those who only got hints. The researchers think time is part of the reason. A tutoring chat took several minutes, so tutored students solved fewer problems, about 14 per lesson instead of 19, and covered less of the material on the test. In an earlier trial, Eedi says, clear gains on the same test appeared only after a year or more.

本段译文

更大范围的测验则呈现出不同的情况。学生在试验前后参加了标准化测验 Star Maths。使用 AI 辅导的学生与只得到提示的学生成绩大致相同。研究者认为时间是部分原因。一次辅导聊天要花好几分钟，因此接受辅导的学生解的题更少，每节课约 14 道而不是 19 道，覆盖的测验内容也更少。Eedi 表示，在此前的一项试验中，同一测验上的明显进步要在一年或更长时间后才出现。

The data also raised new questions. Tutoring seemed to help girls a little but may have set boys back, especially with human tutors. The researchers call this result tentative. One possible reason is that being pulled into a chat after every mistake may feel like a signal of failure. About one in five students never replied when a tutor appeared.

本段译文

数据还引出了新问题。辅导似乎对女生略有帮助，却可能让男生退步，在真人辅导中尤其如此。研究者称这一结果是初步的。一个可能的原因是，每次出错后都被拉进一段聊天，可能让人觉得这是失败的信号。约五分之一的学生在辅导老师出现时从未回复。

Eedi began a year-long trial in September to see whether the classroom gains add up over time. As its chief impact officer, Bibi Groot, put it, AI tutoring can support motivated students, but the harder task is helping those who are not motivated.

本段译文

Eedi 已于 9 月启动一项为期一年的试验，以观察课堂上的进步能否随时间积累。正如其首席影响官 Bibi Groot 所说，AI 辅导能帮助有学习动力的学生，但更难的任务是帮助那些缺乏动力的学生。

新闻来源

• Eedi · Eedi Publishes Full Findings from Its Most Recent AI Tutor Trial (29 Oct 2026)
• Google Blog · How Gemini is helping students and teachers in the classroom (29 Oct 2026)
• The 74 · Five Things You Should Know About New AI Tutor Study (29 Oct 2026)

✳更多文章 →