AI 模型和平台
Google 扩展指导学习和计划 Opt-In K-12 AI 访问
在 Google 上的首选来源中添加 Unite.AI
Google 宣布于2026年11月3日在全球Google教室提供Guided Learning，并将在2026年11月晚些时候在选定的美国学校Search进行试点，同时计划让几个AI产品选择进入K-12学生账户。
Google的副总裁兼教育总经理Chris Phillips，推出围绕设计原则更新的Google称为“生产性斗争 ”：让学生积极挑战而不会变得不堪重负而放弃。 指导学习于2025年8月6日首次抵达Gemini，使用探查和开放式问题，分步解释以及图，视频和交互式问答等多模式响应。
在教室和搜索中如何指导学习
在教室里，一名教师指定活动，人工智能作为一对一的学习工具，固定在课堂材料中。 Google的教育报告称活动可以基于教师加载的材料。 Focus Mode让学生继续执行任务，教师得到实时级和个人的见解，包括完整的聊天记录，以确定学生需要哪些额外支持。 Google在发布之前对学校和教育工作者进行了课堂特色测试，并吸收了他们的反馈。
对于Search中的学生账号，某些家庭作业帮助请求会触发一步步的提示，而不是直接找到最终答案。 兹Q 1Z的例子包括粘贴的物理问题和要求帮助撰写书籍报告。 试点将在未来几个月向其他校区收集教育者反馈，然后才发布扩展专题。
K- 12 AI 访问权限需要管理员批准
Gemini,Gemini Notebook,Gemini在教室和Google搜索将完全关闭K-12学生账户，直到学校管理员明确开启每个产品。 这项改变尚未完全生效：Google将在控制推出前的几个月内与学校分享更多细节，给管理员准备时间。 经修订的方法遵循了学校、教师、研究人员和决策者的反馈，旨在使学校领导人决定如何在课堂上使用AI产品。
Google为其核心教育服务工作空间表示，学校拥有其数据，学生数据不出售，不用于广告或用于培训基因AI模型。
课堂试验显示
Google的学习LM小组和Fab AI在Sierra塞拉利昂进行了预先登记的随机审判，在政府支助的12所学校的48个7年级和8个数学教室招收了1,763名13岁或13岁以上的学生。 教室在指导学习和标准教学之间随机划分，要求指导学习教室的教师提供12小时的使用时间。 研究人员报告了0.258标准偏差的意向处理数学效应，并将其比喻为中位学生从基线的50百分位移到终点的60百分位；69%的学生达到要求的使用门槛。
教师确定目标和启动激励，组织学生成对工作，并领导讨论巩固学习。 研究人员认为，这种由教师主导的结构对干预至关重要，发现以更强的数学技能进入的学生受益更多，给从后面开始的学生带来改善成果的明确挑战。
另有一次为期10周的英国审判，有1,179名学生在爱迪岛测试了Gemini动力辅导。 专家辅导员在发送之前审查了每个AI起草的信息。 研究人员报告说，在94 385份审查的草稿中，辅导员批准了94.7%的草稿，未作修改，监督的人工智能辅导与关于近距离措施和中间措施的专家人类辅导相匹配，同时大大超过了标准补救。
广义的标准化星数学评估产生了不同的结果：学生在AI辅导或标准补救后表现出类似的收益，对辅导没有明显的整体优势。 研究人员将这种差异部分地与辅导对话联系起来，从而缩小了课程覆盖面，而称较长的研究应在近期成果成为更广泛的评估改进时加以审查。
因此，扩大后，执行工作与准入是必然的结果。 教室设计将教师加载材料、教师指定的目标和聊天记录的获取结合起来，而即将到来的设置则将启动决定与管理员结合起来。 这些试验在一些学习措施方面有所收获，但也使考试成绩、课程覆盖面和学习成绩较低的学生的福利得不到解决。 核心问题是，这些衡量的效益是否将推广到更广泛的学校使用，而不重复研究中记录的不平等收益和覆盖面的缩小。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
AI Models & Platforms
Google Expands Guided Learning and Plans Opt-In K-12 AI Access
Add Unite.AI to your preferred sources on Google
Google announced on November 3, 2026 that Guided Learning is globally available in Google Classroom and will be piloted in Search at select U.S. schools later in November 2026, alongside plans to make several AI products opt-in for K-12 student accounts.
Chris Phillips, Google’s vice president and general manager of education, introduced the updates around a design principle Google calls “productive struggle”: keeping students actively challenged without becoming so overwhelmed that they give up. Guided Learning first arrived in Gemini on August 6, 2025, using probing and open-ended questions, step-by-step explanations and multimodal responses such as diagrams, videos and interactive quizzes.
How Guided Learning Works in Classroom and Search
In Classroom, a teacher assigns the activity and the AI acts as a one-to-one study tool anchored in class material. Google’s education report says the activity can be grounded in teacher-uploaded materials. Focus Mode keeps students on task and redirects the conversation toward teacher-specified learning objectives if they try to move off topic. Teachers receive real-time class-level and individual insights, including full chat transcripts, to identify where students need additional support. Google tested the Classroom feature with schools and educators before launch and incorporated their feedback.
For student accounts in Search, certain homework-help requests will trigger step-by-step prompts rather than a direct path to a final answer. Google’s examples include a pasted physics problem and a request for help writing a book report. The pilot will gather educator feedback before the feature expands to other school districts in the coming months.
K-12 AI Access Will Require Administrator Approval
Access to Gemini, Gemini Notebook, Gemini in Classroom and Google Search will be entirely off for K-12 student accounts until a school administrator explicitly turns each product on. The change is not yet fully in force: Google will share more specifics with schools in the coming months before the controls roll out, giving administrators time to prepare. The revised approach follows feedback from schools, teachers, researchers and policymakers and is intended to let school leaders decide how AI products are used in their classrooms.
For its core Workspace for Education services, Google states that schools own their data and that student data is not sold, used for advertising or used to train generative AI models.
What Classroom Trials Show
A preregistered randomized trial in Sierra Leone conducted by Google’s LearnLM team and Fab AI enrolled 1,763 students aged 13 or older across 48 Grade 7 and 8 mathematics classrooms in 12 government-supported schools. Classrooms were randomized between Guided Learning and standard instruction, and teachers in the Guided Learning classrooms were asked to provide 12 hours of use. The researchers reported an intent-to-treat mathematics effect of 0.258 standard deviations and likened it to a median student moving from the 50th percentile at baseline to the 60th at endline; 69% of students met the requested usage threshold.
Teachers defined objectives and starter prompts, organized students to work in pairs and led discussions to consolidate learning. The researchers described that teacher-led structure as essential to the intervention and found that students who entered with stronger mathematics skills benefited more, leaving an explicit challenge to improve outcomes for students who start behind.
A separate 10-week UK trial with 1,179 students tested Gemini-powered tutoring on Eedi. Expert tutors reviewed every AI-drafted message before delivery. The researchers reported that tutors approved 94.7% of the 94,385 reviewed drafts without changes and that supervised AI tutoring matched expert human tutoring on proximate and intermediate measures while substantially outperforming standard remediation.
The broad standardized Star Maths assessment produced a different result: students showed similar gains after AI tutoring or standard remediation, with no clear overall advantage for tutoring. The researchers linked that discrepancy partly to tutoring conversations reducing curriculum coverage and said longer studies should examine when immediate gains become broader assessment improvements.
The expansion therefore makes implementation as consequential as access. The Classroom design combines teacher-uploaded materials, teacher-specified objectives and access to chat transcripts, while the forthcoming settings place activation decisions with administrators. The trials produced gains on some learning measures, but they also left broad-test gains, curriculum coverage and benefits for lower-attaining students unresolved. The central open question is whether those measured benefits will carry into wider school use without repeating the unequal gains and reduced coverage documented in the research.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.