11-19 更新 实时
首页/ AI 大模型/ 正文

模拟学生学习以改进 AI 辅导

AI 大模型 来源:Microsoft Research 2026-10-13 00:00 0 阅读 机器翻译
模拟学生学习以改进 AI 辅导

StudentSim 是一个研究系统,用于在 AI 辅导工具接受真实学习者测试之前对其进行评估和改进;它不是产品,没有在课堂中部署,也不打算取代教师。

AI 辅导工具需要来自学生的反馈才能改进,但通过人类研究收集这类反馈速度慢、成本高,且难以规模化。

StudentSim 利用公开可用、已去标识化的学习者记录来训练个性化的 AI 学生模拟器,并为每位学生适配一个模拟器。

该研究从两个有教育学依据的维度来评估这些模拟器:它们是否像特定学生那样作答,以及它们在接受辅导后是否会更新。

在国际象棋、第二语言英语写作和数学领域,StudentSim 的表现优于领先的基线方法,并为训练 AI 辅导工具提供了有用的反馈信号。

在研究环境中,各团队正在探索 AI 驱动的辅导系统如何帮助解释一道数学题、为写作作业提出修改建议,或引导学习者理解一个国际象棋局面。然而,教学质量不能仅从辅导工具的答案来衡量。辅导工具需要知道在指导传达给学生之后发生了什么。学生是否纠正了原来的错误?这个提示是帮助他们推理出答案,还是透露了太多?不同的学生是否需要不同类型的支持?在当今的 AI 辅导工具研究中,回答这些问题往往需要开展真实学习者研究。这些研究很有价值,但运行起来速度慢、成本高,这使得在快速迭代模型的过程中很难利用学生反馈。

StudentSim 是一个 AI 系统,它能为每位学生创建个性化的模拟器,对学生的答题方式以及在接受辅导后答案如何变化进行建模。其目标是让 AI 辅导系统开发过程中能更容易获得学生反馈。这项研究还提出了 StudentSimEval,这是一个涵盖国际象棋、第二语言英语写作和数学的基准测试与评估协议。

需要强调的是,其目标并非取代教师或课堂学习。相反,StudentSim 帮助研究人员在研发过程中评估和改进 AI 辅导系统,之后才用真实学习者进行测试。该研究仅使用公开的、为研究用途发布的、经去标识化处理的学习者数据集;在任何阶段都不会使用可识别个人身份的信息。

图 1:AI 辅导系统需要学生的反馈才能改进。学生模拟器可以在研究阶段的训练和评估中近似提供其中一部分反馈。它们提供的是一种早期研究信号,并不能取代用真实学习者进行的测试。

衡量学生模拟器需要做到什么

构建学生模拟器的一种简单方法是让大语言模型扮演学习者。例如,提示词可以写成“你是一名数学学不好的小学生”。模型可以改变语气并表达不确定。更难的部分在于保持学习者真实的能力边界。模型可能用年幼学生的口吻说话,却仍然依赖远超该学生水平的知识和推理能力。对于训练 AI 辅导系统而言,这会产生误导性的反馈:由于模拟出的学生能力过强,辅导系统可能显得很有效。

传统的学生建模方法面临着另一种局限。知识追踪和行为预测模型以真实的学习记录为基础,能够估计学生当前的知识、技能和学习习惯。然而,这类模型中的许多在设计之初并未将自然语言的导师指导作为输入。它们可以帮助描述学生做过什么,但并不总能告诉我们,学生对某个特定的讲解、提示或苏格拉底式提问会作何反应。

StudentSim 借鉴教育研究,从两个维度来定义目标。教育工作者常常区分学习者能独立完成什么、以及在获得支持时能完成什么。维果茨基的最近发展区(zone of proximal development)刻画的正是当前独立表现与在指导下表现之间的这一差距。StudentSim 将这一理念转化为两个评估目标。行为保真度衡量模拟器在留出题集上的表现是否与特定学生的记录作答相匹配。指导响应度衡量模拟器在阅读导师指导后是否会更新其作答。

图 2:StudentSimEval 评估模拟器是否与学生的独立行为相匹配,以及它是否会对指导作出响应。

这两个目标相辅相成。一个只与过往学生行为相匹配的模拟器或许对预测有用,但它无法表明某项教学干预是否起到了帮助。一个总是听从导师的模拟器可能对答流畅,却可能无法体现某个特定学生的强项、弱项以及可能犯的错误。AI 导师训练同时需要这两种信号:学习者的起点,以及指导如何影响学习者的表现。

从稀疏的学生记录中训练个性化模拟器

StudentSimEval 完全基于公开可用的学习者数据集构建,这些数据集以去标识化记录的形式面向研究用途发布。这些记录不含任何个人身份信息,每个个性化模拟器仅使用其所代表学生的去标识化记录进行训练。

真实的学生数据在个体层面往往十分稀疏。一名学习者可能只贡献少量作文、答案或交互记录。为每个学生从零开始训练一个独立的模拟器,既存在过拟合的风险,也要求每个模型重新学习该领域的通用模式。

StudentSim 采用两阶段训练流程。第一阶段,模型在同一领域内众多学生的汇总记录上进行训练。这一阶段学习常见的回答格式、常见错误,以及学生在获得指导后常见的修改方式。第二阶段,模型针对单个学生自己的记录进行专门化训练,从而为该学生生成个性化模拟器。

图 3:StudentSim 首先从汇总记录中学习共性的学生行为,然后适配到每个学生自己的数据。

该评估涵盖三个领域,对应不同的回答类型。在国际象棋中,模拟器预测某位棋手的下一步走法,以及该棋手在接受指导后可能如何修改。在英语第二语言写作中,它生成与某位学习者错误模式相符的作文,并预测其在教师反馈后的修改。在数学中,它预测某位学生的选择题答案,以及该学生在得到讲解后是否会更新答案。在这些领域中,StudentSimEval 遵循同一原则:在相同的逐学生训练记录和相同的留出评估记录上比较所有方法。

StudentSim 在三个领域均优于主流基线。以国际象棋为例,StudentSim 的行为保真度达到 0.5150,引导响应度达到 0.9067。GPT-5.4 分别为 0.2316 和 0.7186。Maia2 是一款专用于国际象棋的人类走子预测模型,其两项指标分别为 0.4535 和 0.2721。这一对比体现了不同基线家族的差异:GPT-5.4 能够响应引导,但无法稳定匹配特定学生的行为;而 Maia2 对人类走子的预测更贴近,却缺乏自然语言引导路径。

图 4:在国际象棋中,与主流基线相比,StudentSim 同时提升了行为保真度和引导响应度。

用模拟反馈改进 AI 导师

该研究还测试了训练好的学生模拟器能否为导师优化提供反馈。在一项国际象棋导师强化学习研究中,导师接收一个局面和一名真实学生此前的错误走子,然后生成引导。在此过程中,学生模拟器自身的参数保持固定,它读取引导内容并给出修正后的走子。奖励基于从学生此前走子到模拟器修正走子在走子质量上的提升,并以 Stockfish 作为独立的国际象棋引擎信号。

图 5:StudentSim 可以在 AI 导师的强化学习过程中提供反馈,方式是估计学生在接受引导后其回应会发生怎样的变化。

该研究比较了三种实验条件:一个仅经监督微调、未做强化学习的辅导模型;一个以 GPT-5.4 作为模拟学生奖励来源训练出的辅导模型;以及一个以 StudentSim 作为奖励来源训练出的辅导模型。所有条件都使用相同的辅导策略、相同的监督起点和相同的强化学习设置。随后,人类国际象棋专家在不知晓实验条件的情况下对辅导模型的回答进行盲评。

StudentSim 奖励训练出的辅导模型在所有报告维度上评分最高。用 StudentSim 训练的辅导模型达到 90.5% 的准确率,而无强化学习基线的准确率为 75.7%,以 GPT-5.4 为奖励来源的条件为 71.6%。它在引导质量和个性化方面也获得了更高的评分。这一结果构成了一项概念验证:基于真实学习者数据训练的学生模拟器,可以为改进 AI 辅导模型提供有用的反馈。

原文较长,译文为节选;译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Simulating student learning for better AI tutoring)

Simulating student learning for better AI tutoring - Microsoft Research

A conceptual diagram of the feedback loop required to improve an AI tutor. The middle part of the figure shows that a human teacher gives guidance to a real student and receives the student’s feedback, but this feedback is slow, costly, sparse, and hard to scale. On the right panel of the figure, an AI student simulator is trained from real student feedback and serves as a fast proxy for the student during tutor development.

StudentSim is a research system for evaluating and improving AI tutors before they are tested with real learners; it is not a product, is not deployed in classrooms, and is not intended to replace teachers

AI tutoring tools need feedback from students to improve, but collecting that feedback through human studies is slow, costly, and difficult to scale

StudentSim trains personalized AI student simulators from publicly available, de-identified learner records, with one simulator adapted to each student

The work evaluates simulators along two education-grounded dimensions: whether they respond like a specific student, and whether they update after tutor guidance

Across chess, second-language English writing, and mathematics, StudentSim outperforms leading baselines and provides a useful feedback signal for training an AI tutor

In research settings, teams are exploring how AI-powered tutoring systems can help explain a math problem, suggest an edit to a writing assignment, or guide a learner through a chess position. Yet the quality of teaching cannot be measured only from the tutor’s answer. A tutor needs to know what happens after the guidance reaches a student. Did the student correct the original mistake? Did the hint help them reason toward the answer, or did it give away too much? Would a different student need a different kind of support? In today’s AI tutor research, answering these questions often requires real learner studies. Those studies are valuable, but they are slow and costly to run, which makes it hard to use student feedback during fast model iteration.

StudentSim is an AI system that creates personalized simulators of individual students, modeling how they answer questions and how their answers change after receiving a tutor’s guidance. The goal is to make student feedback more available during AI tutor development. This research also introduces StudentSimEval, a benchmark and evaluation protocol spanning chess, second-language English writing, and mathematics.

Importantly, the goal is not to replace teachers or classroom learning. Instead, StudentSim helps researchers evaluate and improve AI tutoring systems during the research and development process before testing them with real learners. The research uses only publicly available learner datasets released for research use with de-identified records; no personally identifying information is used at any stage.

**Alt text:** A conceptual diagram of the feedback loop required to improve an AI tutor. The middle part of the figure shows that a human teacher gives guidance to a real student and receives the student’s feedback, but this feedback is slow, costly, sparse, and hard to scale. On the right panel of the figure, an AI student simulator is trained from real student feedback and serves as a fast proxy for the student during tutor development.Figure 1:AI tutors need feedback from students to improve. Student simulators can approximate part of that feedback during research-stage training and evaluation.Theyprovide an early-stage research signal, not a replacement for testing with real learners.

Measuring what a student simulator needs to do

A simple way to build a student simulator is to ask a large language model to role-play a learner. For example, a prompt might say, “You are an elementary-school student who struggles with math.” The model can change its tone and express uncertainty. The harder part is preserving the learner’s actual ability boundary. A model may speak in a young student’s voice while still relying on knowledge and reasoning far beyond that student’s level. For training an AI tutor, this can produce misleading feedback: the tutor may appear effective because the simulated student is too capable.

Traditional student modeling approaches face a different limitation. Knowledge tracing and behavior prediction models are grounded in real learning records and can estimate a student’s current knowledge, skills, and learning habits. Many of these models, however, were not designed to take natural-language tutor guidance as input. They can help describe what a student has done, but they do not always tell us how a student would respond to a particular explanation, hint, or Socratic question.

StudentSim defines the target through two dimensions inspired by education research. Educators often distinguish what a learner can do independently from what they can do with support. Vygotsky’s zone of proximal development captures this gap between current independent performance and performance with guidance. StudentSim turns this idea into two evaluation goals. Behavioral fidelity measures whether a simulator matches a specific student’s recorded responses on held-out problems. Guidance responsiveness measures whether the simulator updates its response after reading tutor guidance.

**Alt text:** Diagram defining the two evaluation criteria used for student simulators. The upper portion shows a problem, “minus 7.3 times 5 equals what.” The lower portion compares a real student and an AI simulator. Both initially answer 36.5, then receive guidance and revise their answer to negative 36.5. Asking whether a simulator’s responses match the recorded student answers is behavioral fidelity. Matching the guidance-induced change is guidance responsiveness.Figure 2: StudentSimEval evaluates whether a simulator matches a student’s independent behavior and whether it responds to guidance.

These two goals are complementary. A simulator that only matches past student behavior may be useful for prediction, but it does not show whether a teaching intervention helped. A simulator that always follows the tutor may respond fluently, but it may fail to represent a particular student’s strengths, weaknesses, and likely mistakes. AI tutor training needs both signals: the learner’s starting point and how guidance affects the learner’s performance.

Training personalized simulators from sparse student records

StudentSimEval is built entirely from publicly available learner datasets that are released for research use with de-identified records. The records carry no personally identifying information, and each personalized simulator is trained only on the de-identified records of the student it represents.

Real student data is often sparse at the individual level. A learner may contribute only a small number of essays, answers, or interactions. Training a separate simulator from scratch for each student would risk overfitting and would require each model to relearn common patterns in the domain.

StudentSim uses a two-stage training pipeline. In the first stage, the model trains on records pooled across many students in the same domain. This stage learns common response formats, common errors, and common ways students revise after guidance. In the second stage, the model is specialized on one student’s own records, producing a personalized simulator for that student.

**Alt text:** A two-stage pipeline for training personalized StudentSim models from sparse student records. At the top, several students each have limited examples of question-answer interactions and answer revisions following tutor guidance. In the middle, stage one training, records from all students are pooled to learn shared behavioral patterns, including the space of possible responses, common mistakes, and typical revisions after guidance. At the bottom, stage two training, the model is adapted separately to each student’s records, producing one personalized simulator per student.Figure 3: StudentSim first learns shared student behavior from pooled records, then adapts to each student’s own data.

The evaluation covers three domains with different response types. In chess, the simulator predicts a player’s next move and how that player may revise after coaching. In second-language English writing, it generates essays with error patterns matching a learner and predicts corrections after teacher feedback. In mathematics, it predicts a student’s multiple-choice answer and whether the student updates after an explanation. Across these domains, StudentSimEval uses the same principle: compare all methods on the same per-student training records and the same held-out evaluation records.

StudentSim outperforms leading baselines across the three domains. In chess, for example, StudentSim reaches behavioral fidelity of 0.5150 and guidance responsiveness of 0.9067. GPT-5.4 reaches 0.2316 and 0.7186, respectively. Maia2, a chess-specific human move prediction model, reaches 0.4535 and 0.2721. The comparison illustrates the difference between the baseline families: GPT-5.4 can respond to guidance but does not reliably match a specific student’s behavior, while Maia2 predicts human chess moves more closely but lacks a natural-language guidance pathway.

**Alt text:** Scatter plot comparing three student simulation approaches on behavioral fidelity, shown on the horizontal axis, and guidance responsiveness, shown on the vertical axis. StudentSim is positioned in the upper-right at approximately 0.51 fidelity and 0.91 responsiveness. A prompted GPT-5.4 model has lower fidelity, approximately 0.23, and moderate responsiveness, approximately 0.72. Maia2, a state-tracking model, has moderate fidelity, approximately 0.45, but low responsiveness, approximately 0.27. The plot shows that StudentSim performs strongly on both criteria.Figure 4: In chess, StudentSim improves both behavioral fidelity and guidance responsiveness compared with leading baselines.

Using simulated feedback to improve AI tutors

The research also tests whether a trained student simulator can provide feedback for tutor optimization. In a chess tutor reinforcement learning study, the tutor receives a position and a real student’s previous wrong move, then generates guidance. The student simulator’s own parameters are held fixed during this process, and it reads the guidance and produces a revised move. The reward is based on the improvement in move quality from the student’s previous move to the simulator’s revised move, using Stockfish as an independent chess engine signal.

**Alt text:** Diagram of reinforcement learning for an AI chess tutor using StudentSim as the feedback source. Each episode begins with a real student’s chess position and previous incorrect move. The AI tutor receives this context and generates guidance describing what and how to teach. A frozen, personalized AI student simulator reads the guidance and produces a revised move. The reward equals the improvement in chess evaluation from the student’s original move to the revised move, and this reward updates the AI tutor policy while the simulator remains fixed.Figure 5: StudentSim can provide feedback during AI tutor reinforcement learning by estimating how a student response changes after guidance.

The study compares three conditions: a supervised fine-tuned tutor with no reinforcement learning, a tutor trained with GPT-5.4 as the simulated student reward source, and a tutor trained with StudentSim as the reward source. All conditions use the same tutor policy, supervised starting point, and reinforcement learning setup. Human chess experts then evaluate the tutor responses blind to condition.

The StudentSim reward produces the highest-rated tutor on all reported dimensions. The tutor trained with StudentSim reaches 90.5% accuracy, compared with 75.7% for the no-RL baseline and 71.6% for the GPT-5.4 reward condition. It also receives higher ratings for guidance quality and personalization. The result is a proof of concept that student simulators trained on real learner data can provide useful feedback for improving AI tutors.

StudentSim does not remove the need for real learners in education research. Human studies remain essential for measuring learning outcomes, safety, usability, and classroom fit. The role of student simulators is earlier in the development process: they can help researchers test tutor strategies, compare models, and refine training signals before moving to more expensive studies with real students.

The broader direction is to build AI tutor systems that learn from student reactions, not only from expert-written explanations or static correctness labels. StudentSim shows one path: use real learning records to train personalized simulators, evaluate those simulators on both independent behavior and response to guidance, and use them as scalable feedback sources for tutor training. Future work can extend this approach to longer learning trajectories, additional subjects, richer student profiles, and studies that connect simulated feedback with real learning outcomes.

The paper and the StudentSim training and evaluation code, including the StudentSimEval protocol, are publicly available.

University of Illinois Urbana-Champaign

Senior Principal Research Manager

Professor, University of Illinois Department of Computer Science

University of Illinois Urbana-Champaign

Technical Fellow & Corporate Vice President

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 Microsoft Research(Simulating student learning for better AI tutoring),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗