How universities are affected by AI moving knowledge outside their walls

A student today can ask an artificial intelligence (AI) model anything without embarrassment: Whether it’s a request for an explanation of how compound interest works or of Pareto efficiency, each question gets a patient, direct answer at any time and at whatever level of sophistication the asker needs.

如今，学生可以向人工智能 (AI) 模型提出任何问题，而不会感到尴尬：无论是要求解释复利如何运作还是帕累托效率，每个问题都可以随时得到耐心、直接的答案，无论提问者需要什么复杂程度。

For most of the university’s 900-year history, the core repository of teachable knowledge has been inside the walls of the institution—in its faculty, libraries, laboratories, and classrooms.

在这所大学 900 年历史的大部分时间里，可教授知识的核心存储库一直在学校的围墙内——教职员工、图书馆、实验室和教室。

But now, the center of accessible explanation has moved outside the university’s walls to AI models.

但现在，易于解释的中心已经转移到大学之外的人工智能模型上。

And the implications are larger than the current debate about AI in the classroom suggests, rather, the shift affects the underlying value proposition of higher education itself.

其影响比当前关于课堂人工智能的争论所暗示的要大，相反，这种转变影响了高等教育本身的潜在价值主张。

For much of its history, the university has been organized around the production and management of scarce knowledge.

在其历史的大部分时间里，大学都是围绕稀缺知识的生产和管理而组织的。

The medieval university held the books, trained the clergy and lawyers, and certified mastery of a fixed body of authoritative texts.

中世纪的大学拥有书籍，培训神职人员和律师，并证明他们掌握了一套固定的权威文本。

The industrial-era mass university expanded access dramatically but kept the same architecture: Authoritative knowledge lived inside the institution, delivered through standardized curriculum, fixed time-to-degree, and a credential at the end.

工业时代的大众大学极大地扩大了入学范围，但保持了相同的架构：权威知识存在于机构内部，通过标准化课程、固定的获得学位的时间和最终的证书来提供。

The internet later weakened this monopoly but did not break it.

后来互联网削弱了这种垄断，但并没有打破它。

Raw information became abundant, but interpretation, explanation, and feedback still required faculty.

原始信息变得丰富，但解释、解释和反馈仍然需要教师。

The mass university could be compared, in its essential design, to a modern industrial factory.

从本质上讲，大众大学可以与现代工业工厂相比较。

Students are the raw material, faculty and curriculum are the production process, and the credentialed graduate is the output.

学生是原材料，师资和课程是生产过程，合格的毕业生是产出。

Instruction is standardized so the line can run at scale, time-to-degree fixes the throughput, and the diploma certifies that the product met specification.

指令是标准化的，因此生产线可以大规模运行，时间到程度固定吞吐量，并且证书证明产品符合规格。

This is not a hostile description—the industrial-age model democratized higher education and did enormous good.

这并不是一个充满敌意的描述——工业时代的模式使高等教育民主化，并带来了巨大的好处。

But a factory is built to turn standardized inputs into standardized outputs efficiently, and standardization is precisely the kind of work intelligent automation performs exceptionally well.

但工厂的建立是为了有效地将标准化输入转化为标准化输出，而标准化正是智能自动化表现出色的工作。

What has changed in the past few years is not simply access to information but the cognitive layer itself.

过去几年发生的变化不仅仅是信息获取，还有认知层本身。

AI does not merely provide information.

人工智能不仅仅提供信息。

It explains, customizes, gives feedback, and synthesizes the type of work that justified the lecture hall and office hours.

它解释、定制、提供反馈并综合证明演讲厅和办公时间合理的工作类型。

This is in line with what Benjamin Bloom identified in 1984 as the “ two-sigma problem ”: one-on-one tutoring produced learning outcomes two standard deviations better than classroom instruction but tutoring at scale was economically impossible.

这与本杰明·布鲁姆 (Benjamin Bloom) 在 1984 年提出的“二西格玛问题”一致：一对一辅导的学习成果比课堂教学好两个标准差，但大规模辅导在经济上是不可能的。

For 40 years the problem was unsolved, blocked not by pedagogy but by economics.

四十年来，这个问题一直没有得到解决，阻碍的不是教育学而是经济学。

AI has now removed the economic barrier: Every student can have a patient, responsive tutor on demand.

人工智能现在已经消除了经济障碍：每个学生都可以按需拥有一位耐心、反应灵敏的导师。

Whether these systems reliably reproduce Bloom’s full two-sigma gain remains an open empirical question.

这些系统是否可靠地再现了布卢姆的完整二西格玛增益仍然是一个悬而未决的经验问题。

But it is clear the constraint that made one-on-one explanation a luxury has collapsed.

但很明显，使一对一解释成为一种奢侈的限制已经崩溃。

This does not mean AI is correct, complete, or unbiased.

这并不意味着人工智能是正确的、完整的或公正的。

In fact, it is none of those things consistently.

事实上，这些事情都不是始终如一的。

But universities no longer have a near-monopoly on knowledge production and dissemination.

但大学不再几乎垄断知识的生产和传播。

The student no longer needs to come to campus to receive an explanation of the invisible hand, the second law of thermodynamics, or the Federalist Papers.

学生不再需要来到校园接受关于“看不见的手”、热力学第二定律或《联邦党人文集》的解释。

The explanation comes to each of them, on demand, in whatever form and at whatever time they request.

他们每个人都会根据需要、以任何形式、在他们要求的任何时间得到解释。

The boundary that defined the institution for nine centuries has, in a single technological generation, become quite porous if not disappearing in total.

九个世纪以来定义该机构的边界，在单一的技术一代中，即使没有完全消失，也变得相当多孔。

To see why this matters, it helps to be precise about what the modern university bundles together.

要了解为什么这很重要，有必要准确了解现代大学的捆绑内容。

Three products are sold under a single tuition price.

三种产品以单一学费出售。

The first is content delivery, including lectures, readings, and problem sets, that transmit established knowledge from those who hold it to those who do not.

第一个是内容交付，包括讲座、阅读材料和问题集，将已有的知识从掌握知识的人传递给不掌握知识的人。

This is a student’s daily experience, and it has been the primary justification for the university’s cost and time since the rise of the industrial mass university in the late 19th century.

这是学生的日常经历，自 19 世纪末工业大众大学兴起以来，这一直是大学成本和时间的主要理由。

The second is credentialing—the degree as a signal to the labor market that its holder has completed a recognized program at a recognized institution.

第二个是认证——学位向劳动力市场发出信号，表明其持有者已在认可机构完成了认可课程。

The credential is what employers actually purchase when they require a bachelor’s or master’s degree, and it is what makes the university economically essential for most students even when the content could be acquired elsewhere.

该证书是雇主在需要学士或硕士学位时实际购买的东西，也是使大学对大多数学生在经济上至关重要的原因，即使这些内容可以在其他地方获得。

The third is networking and formation.

第三是网络和形成。

Networking is the slower, harder-to-measure work of meeting people, establishing connections, joining an intellectual community, and entering jobs, communities, and professional circles that follow from them.

人际交往是一种较慢、较难衡量的工作，包括与人会面、建立联系、加入知识社区，以及进入随之而来的工作、社区和专业圈子。

Formation is the cultivation of judgment, ethical sensibility, and the capacity to think seriously under uncertainty.

形成是判断力、道德敏感性以及在不确定性下认真思考能力的培养。

Both happen through sustained relationships with faculty and fellow students, through encounters with hard cases, through participation in traditions of inquiry, and through the experience of being taken seriously as a thinking person by people who have lived a life of inquiry themselves.

两者都是通过与教师和同学的持续关系、通过遇到疑难案件、通过参与探究传统以及通过自己被过着探究生活的人们认真对待为一个有思想的人的经历而发生的。

For roughly a century this bundle held together.

大约一个世纪以来，这个捆绑在一起。

Content delivery was a visible product.

内容交付是一个可见的产品。

Credentialing followed automatically from completing the content.

完成内容后自动进行认证。

Networking and formation came along almost as a byproduct of the time students spent on campus with faculty and peers.

网络和形成几乎是学生在校园与教师和同龄人一起度过的时间的副产品。

AI breaks this bundle in multiple ways.

人工智能以多种方式打破了这一束缚。

Content delivery is the factory’s production line, and it is the most substitutable component.

内容交付是工厂的生产线，是最可替代的环节。

A capable AI tutor, integrated with course materials, can increasingly deliver explanation and feedback at a quality that exceeds the average large-lecture experience, meaning the part of the university that most resembles a factory is the part most exposed to automation.

一位有能力的人工智能导师与课程材料相结合，可以越来越多地以超过平均大型讲座体验的质量提供解释和反馈，这意味着大学中最像工厂的部分是最容易受到自动化影响的部分。

Credentialing remains sticky but is under pressure.

资格认证仍然具有粘性，但面临压力。

Employers across technology, finance, and government have announced moves toward skills-based hiring, though research suggests practice still lags the proclamations.

技术、金融和政府领域的雇主已经宣布采取基于技能的招聘举措，尽管研究表明实践仍然落后于公告。

Networking and formation are the hardest for AI to replicate, because they are fundamentally relational and embodied in personal networks.

网络和形成是人工智能最难复制的，因为它们从根本上来说是关系性的，并且体现在个人网络中。

They require human judgment encountering other human judgment over time, in conditions of mutual stakes.

随着时间的推移，它们需要人类的判断在相互利害的情况下与其他人类的判断相遇。

It is difficult to predict what the university will become, but the direction of change is already visible, and it is running along several distinct tracks at once.

很难预测大学会变成什么样子，但变化的方向已经可见，而且它同时沿着几条不同的轨道运行。

The first track is unbundling and speed.

第一条轨道是分拆和速度。

Western Governors University has built a competency-based model in which students advance by demonstrating mastery rather than by seat time.

西部州长大学建立了一个基于能力的模型，学生通过展示掌握程度而不是通过上课时间来进步。

Southern New Hampshire University has also developed competency-based, direct-assessment models that prioritize demonstrated competence over time spent in courses.

南新罕布什尔大学还开发了基于能力的直接评估模型，该模型优先考虑展示的能力而不是花费在课程上的时间。

Google’s career certificates sit just outside the university system but show the same pressure: short, job-oriented credentials designed to connect learners with employers without requiring a full degree.

谷歌的职业证书位于大学系统之外，但也显示出同样的压力：简短的、以工作为导向的证书，旨在将学习者与雇主联系起来，而不需要完整的学位。

Arizona State’s partnership with Starbucks offers eligible employees 100% upfront tuition coverage for an online bachelor’s degree through the university.

亚利桑那州立大学与星巴克的合作伙伴关系为符合条件的员工提供 100% 的大学在线学士学位预付费用。

Abroad, South Korean universities have developed company-linked contract departments with firms such as Samsung and SK Hynix, tying curriculum, scholarships, and employment pathways more tightly to corporate workforce needs.

在国外，韩国大学与三星和 SK 海力士等公司建立了与公司相关的合同部门，将课程、奖学金和就业途径与企业劳动力需求更紧密地联系在一起。

This track trades away the residential experience and most of what came with it, but it protects access for working adults, career-changers, and students who could never afford four years of foregone income.

这条道路牺牲了居住体验和随之而来的大部分东西，但它保护了工作成年人、转行者和那些永远无法承担四年损失收入的学生的机会。

A fast and legible credential is thus not a lesser choice but the only workable one.

因此，快速且清晰的证书不是一个次要的选择，而是唯一可行的选择。

The second track is AI-embedded instruction.

第二条轨道是人工智能嵌入式指令。

Arizona State again offers a leading example: Its institution-wide partnership with OpenAI is framed around using ChatGPT Enterprise to enhance teaching, learning, and discovery across the university.

亚利桑那州立大学再次提供了一个领先的例子：其与 OpenAI 的全机构合作伙伴关系是围绕使用 ChatGPT Enterprise 来增强整个大学的教学、学习和发现而构建的。

Georgia Tech ‘s online master’s in computer science—low-cost, high-volume, and fully online—showed a decade ago that elite instruction could scale; AI can now be layered onto that earlier model of scaled digital instruction.

佐治亚理工学院的计算机科学在线硕士学位——低成本、大批量、完全在线——在十年前就表明精英教学是可以规模化的；人工智能现在可以分层到早期的规模化数字教学模型上。

Here, the flipped classroom becomes the natural architecture; content is absorbed at home through an AI tutor, and class time is reserved for discussion, critique, and application.

在这里，翻转的课堂成为自然的建筑；内容通过人工智能导师在家吸收，课堂时间用于讨论、批评和应用。

This track forgoes the traditional lecture and much of the routine explanation faculty once provided but protects the classroom itself, repurposed for the work that still requires a room full of people.

该课程放弃了传统的讲座和教师曾经提供的大部分常规解释，但保护了教室本身，并重新用于仍然需要满屋子的工作。

The third track is a quieter shift in assessment.

第三条轨道是评估中更安​​静的转变。

If AI can produce a polished paper, a working proof, or clean code, the finished artifact stops being what a university can meaningfully grade on its own.

如果人工智能可以生成一份精美的论文、一份工作证明或干净的代码，那么完成的作品就不再是大学可以自行进行有意义评分的东西。

Some programs have responded by lowering the stakes attached to any single piece of work.

一些项目的应对措施是降低任何单项工作的风险。

In one emerging model, students choose assignments from a pool larger than any one student is expected to complete.

在一种新兴模型中，学生从比任何一名学生预期完成的作业池都要大的作业中选择作业。

Most of these earn credit and detailed faculty feedback but no evaluative grade; only a few are graded outright.

其中大多数获得学分和详细的教师反馈，但没有评估等级；只有少数是直接评分的。

Students decide how much evaluative weight to take on, and faculty spend their time responding to student work rather than scoring it.

学生决定承担多少评估权重，教师则花时间回应学生的作业而不是评分。

This is early and uneven, but the direction is significant: less weight on any single output an AI could have produced, more attention on the ongoing exchange between student and instructor.

这是早期且不均衡的，但方向很重要：减少对人工智能可能产生的任何单一输出的重视，更多地关注学生和教师之间正在进行的交流。

The fourth track runs in the opposite direction from the first: a smaller set of institutions doubling down on intensive human formation.

第四条轨道与第一条轨道相反：较小的一组机构加倍致力于密集的人类形成。

St. John’s College conducts its curriculum largely through small seminars, tutorials, and laboratories centered on primary texts rather than conventional lectures.

圣约翰学院的课程主要通过小型研讨会、教程和实验室来进行，这些研讨会、教程和实验室以初级文本为中心，而不是传统的讲座。

The service academies treat the formation of judgment and character as their explicit institutional purpose.

军校将判断力和品格的形成视为其明确的制度目的。

Medical residencies and law school clinics have always worked this way.

住院医师实习和法学院诊所一直都是这样运作的。

What this track trades away is scale and cost-efficiency; what it protects is the one product in the bundle that AI cannot manufacture.

这条赛道牺牲的是规模和成本效率。它保护的是人工智能无法制造的捆绑产品中的一种产品。

These tracks are not mutually exclusive, and most institutions will likely combine elements of several.

这些轨道并不相互排斥，大多数机构可能会结合多种元素。

But they are not equivalent either.

但它们也不等同。

The first two compete on cost, speed, and delivery—the terrain where AI’s advantage is largest and still growing.

前两者在成本、速度和交付方面展开竞争——人工智能在这方面的优势最大，而且仍在不断增长。

That is a wager that the credential can outrun the substitution.

这是一个赌注，即证书能够超越替代品。

It may hold for some institutions but asks a great deal of luck for others.

对于某些机构来说可能是这样，但对于其他机构来说则需要很大的运气。

This is why the fourth track deserves more attention than its current scale suggests.

这就是为什么第四条轨道比目前的规模更值得关注。

The small seminars, capstones, and clinical placements do more than transmit knowledge—they build sustained relationships through which both networking and formation happen.

小型研讨会、顶点课程和临床实习的作用不仅仅是传播知识，它们还建立了持续的关系，通过这种关系，网络和形成得以发生。

A student who spends years in the same seminar or a rotation in the same clinic forms working relationships with faculty and peers that outlast the degree itself.

一个学生在同一个研讨会上呆了多年，或者在同一个诊所轮流工作，与教师和同行建立了比学位本身更持久的工作关系。

AI can supply the relevant knowledge, but it cannot decide what to do with it in a particular case, under uncertainty, with real stakes.

人工智能可以提供相关知识，但它无法决定在特定情况下、在不确定性和真正的风险下如何处理这些知识。

It cannot be held responsible for the choice.

它不能对选择负责。

It cannot reliably tell a serious argument from a merely fluent one.

它无法可靠地区分严肃的论证和仅仅流利的论证。

And it cannot make a person a participant in a living tradition of practice—a physician, an engineer, a public servant, or a scholar—rather than a consumer of its outputs.

它不能使一个人成为活生生的实践传统的参与者——医生、工程师、公务员或学者——而不是其成果的消费者。

Judgment under uncertainty, responsibility for outcomes, discernment, and membership in a community of inquiry are what formation produces, and networking follows naturally from the same relationships, once formation has given students something worth connecting over.

不确定性下的判断、对结果的责任、洞察力和探究社区的成员资格都是形成所产生的，一旦形成给了学生一些值得联系的东西，网络自然就会从相同的关系中产生。

This is less a leap forward than a return.

这与其说是一次飞跃，不如说是一次回归。

The university did not begin as a factory but rather a guild.

这所大学最初并不是一家工厂，而是一个行会。

Arguably the first universities—Bologna in the late 11th century and Paris soon after—were a universitas, or a sworn corporation of masters and students organized to pursue a craft.

可以说，第一批大学——11世纪末的博洛尼亚和不久之后的巴黎——是一所大学，或者是一个由大师和学生组成的宣誓团体，为追求一门手艺而组织起来。

Their characteristic exercise was not the lecture alone but the disputatio: the formal disputation in which a student took a position and defended it under challenge.

他们的特色练习不仅仅是讲座，而是辩论：学生在正式辩论中表明立场并在挑战下为其辩护。

To learn was to be initiated into a community of argument, conducted face to face, under masters, judged by the standards of the craft.

学习就是要开始进入一个争论的社区，在大师的指导下，以工艺标准来评判，面对面地进行。

The industrial-era university preserved apprenticeship at the research level but expanded undergraduate education through standardized curricula and mass credentialing—a real democratic achievement that nonetheless traded the guild’s formative core for the factory’s throughput.

这所工业时代的大学保留了研究层面的学徒制，但通过标准化课程和大规模认证扩大了本科教育——这是一项真正的民主成就，但却以行会的形成核心换取了工厂的吞吐量。

AI is now dissolving that factory layer and thus exposing the older logic beneath: education as apprenticeship into a community of practice.

人工智能现在正在溶解工厂层，从而暴露了其背后的旧逻辑：教育作为学徒进入实践社区。

None of this requires abandoning AI.

这些都不需要放弃人工智能。

AI can take over much of the explanation and feedback that currently consumes faculty time, freeing faculty for the work that only humans can do.

人工智能可以接管目前占用教师时间的大部分解释和反馈，从而使教师能够从事只有人类才能完成的工作。

A faculty member whose AI handles initial drafts of problem-set feedback has more time for the seminar discussion, the one-on-one meeting, and the careful reading of a difficult capstone.

人工智能处理问题集反馈初稿的教员有更多时间进行研讨会讨论、一对一会议以及仔细阅读困难的顶点。

AI used this way enables formation by removing the routine work that has always crowded it out.

采用这种方式的人工智能可以通过消除一直排挤它的例行工作来实现形成。

There is no avoiding the question of cost.

成本问题是无法回避的。

Formation-centered education is more expensive per student, not less.

以形成为中心的教育对每个学生来说成本更高，而不是更低。

It requires smaller classes, more faculty time per student, deeper community infrastructure, and stronger ties to external practitioners.

它需要更小的班级、每个学生更多的教学时间、更深层次的社区基础设施以及与外部从业者更紧密的联系。

The change will be painful, and the pain will be uneven: Selective universities with strong endowments will find this path far easier than tuition-dependent regional ones, which is itself a problem that deserves serious attention from policymakers and accreditors.

这种变化将是痛苦的，而且痛苦将是不平衡的：拥有雄厚捐赠的选择性大学会发现这条道路比依赖学费的地区大学容易得多，这本身就是一个值得政策制定者和认证机构认真关注的问题。

The first instinct of many institutions will be to bolt AI onto the existing assembly line.

许多机构的第一反应是将人工智能应用到现有的装配线上。

Courses will become more efficient, advising more automated, grading more scalable, and content delivery more personalized.

课程将变得更加高效，建议更加自动化，评分更加可扩展，内容交付更加个性化。

Some of this will be useful, and much of it may be necessary, but it will not answer the deeper question AI has placed before higher education.

其中一些是有用的，其中大部分可能是必要的，但它不会回答人工智能向高等教育提出的更深层次的问题。

The industrial university was built for a world in which explanation was scarce, expertise was localized, and access to knowledge required institutional mediation, but that world is disappearing.

工业大学是为一个解释稀缺、专业知识本地化、获取知识需要机构调解的世界而建立的，但这个世界正在消失。

The question, then, is not how universities can preserve the old model with better tools.

那么，问题不在于大学如何用更好的工具来保留旧模式。

It is what remains worth preserving when the old model no longer has the same economic logic.

当旧模式不再具有相同的经济逻辑时，它仍然值得保留。

The answer is not content delivery.

答案不是内容交付。

Nor is it credentialing alone, though credentials will remain important for some time.

尽管凭证在一段时间内仍然很重要，但它也不仅仅是凭证。

The answer may be formation: the slow work of bringing students into communities where judgment is practiced, arguments are tested, ethical responsibility is learned, and intellectual seriousness is modeled by other human beings.

答案可能是形成：将学生带入社区，在那里练习判断、检验论点、学习道德责任，并以其他人为榜样，培养智力上的严肃性，这是一项缓慢的工作。

AI will do what factories do best: standardize, scale, and deliver.

人工智能将做工厂最擅长的事情：标准化、规模化和交付。

The university should not try to out-factory the machine.

大学不应该试图将机器外包出去。

Its reason for being is older and harder.

它存在的理由更加古老和艰难。

It is to form people capable of knowing what to do with knowledge once knowledge is everywhere.

一旦知识无处不在，它就会培养人们能够知道如何利用知识。

If knowledge has moved outside the walls, the university’s task is not to pull it back in.

如果知识已经移出墙外，大学的任务不是将其拉回来。

It is to form the people who know how to use it well.

是为了培养懂得如何用好它的人。