MIT Report on
AI & Education

An Interactive Briefing for Moore School Faculty

MIT's recent report on generative AI examines how the technology is reshaping teaching, learning, assessment, and the student experience. This interactive briefing highlights the ideas and findings most relevant to faculty.

The central challenge

AI raises big questions for higher education

MIT's committee was charged with assessing AI use, identifying innovations in teaching and assessment, and proposing an AI-use policy.

But it concluded that AI raises a much deeper set of questions about “the structure, meaning, and value” of an MIT education—questions with obvious implications for higher education more broadly.

What should students learn?
How should they learn it?
How do we know they have learned it?
What parts of education should remain fundamentally human?

Nine recommendations

What Should Educators Be Doing

The report calls for reconsidering what students learn, how learning happens, how we assess it, and what institutions need to do to support the transition. Explore nine of the report’s most consequential recommendations.

§3.1.1 · Revisit course goals

MIT argues that instructors should reconsider the learning goals of every course in light of AI. Some longstanding goals may remain essential and need to be protected. Others may matter less when AI can perform the task readily. And AI may make entirely new learning goals possible.

The report recommends a backward-design approach: first determine what students should know, be able to do, and learn to value; then design instruction, assessment, and AI use around those objectives.

Begin with what students need to be able to do, not what AI can do.
§3.1.2 · Ensure durable learning

AI weakens the assumption that strong completed work necessarily demonstrates individual understanding or mastery. MIT recommends richer ways to determine what students actually know and can do, including oral exams, portfolios, presentations, and out-of-class assignments paired with in-class conversations.

The report also cautions against simply moving more weight onto controlled, high-stakes exams. The goal is not merely to create assessments AI cannot complete, but assessments that provide credible evidence of the learning we care about.

§§3.1.3 and 3.1.5

MIT recommends expanding learning-by-doing: projects, collaborative work, research, internships, and other experiences that place students in authentic contexts.

AI can make these experiences considerably more ambitious by allowing students to undertake projects that previously would have been unrealistic within a course.

The opportunity is not simply to use AI to do the same work faster. It is to raise the level of what students can meaningfully undertake.
§3.1.4 and §§3.2.1–3.2.2

MIT treats AI as a social-learning challenge as well as a cognitive one. It recommends intentionally building meaningful interaction into courses through collaborative problem solving, group projects, discussion, peer feedback, mentoring, and instructor or TA interaction.

Human interaction is not an inefficient substitute for information delivery. It is part of what education is meant to develop.
§§2.4 and 2.7

MIT worries that AI can allow students to bypass the difficulty through which genuine learning occurs. It describes the risk of “cognitive surrender”: turning to AI at the first sign of struggle without developing the underlying knowledge, confidence, persistence, or judgment.

Use AI to augment learning, not automate it away.
§3.1.8

MIT rejects both ambiguity and a single universal rule for AI use. Students should know whether AI is permitted, limited, required, or prohibited, and why that choice supports the learning objective.

The most important part of an AI rule may be the explanation of why that rule supports learning.
§3.1.9

MIT recommends against relying on AI-detection software, citing false positives, difficulty detecting nuanced AI assistance, an escalating arms race, and damage to trust.

It points instead toward drafts, intermediate milestones, version histories, regular project meetings, and conversations about student work.

The goal should be better evidence of learning and clearer expectations—not simply better detection of misconduct.
§3.2.4

Effective use includes working productively with AI, evaluating outputs, verifying claims, and knowing when AI is the wrong tool.

Responsible use includes understanding augmentation versus automation, taking responsibility for outputs, and disclosing AI’s contribution appropriately.

Ethical use includes questions of bias, intellectual property, training data, authorship, environmental impact, and broader consequences.

§3.1.10 and §3.3

MIT calls for sustained experimentation and institutional support: curricular pilots, AI Leads, AI Fellows and implementation teams, pilot funding, faculty communities of practice, ongoing training, data collection, and continuous evaluation.

Meaningful adaptation requires time, expertise, resources, experimentation, and opportunities for faculty to learn from one another.

Student perspective

What Students Say

Selected findings from MIT student surveys show a striking mix of widespread use, perceived value, career expectations, and concern about dependence.

Career readiness gap
70%

of undergraduates said AI proficiency would be important in their careers

25%

felt MIT was preparing students to use AI professionally

Concern
90%

of undergraduates were somewhat or very concerned about overreliance on LLMs

46%

used LLMs daily

30%

used them several times per week

75%

said faculty expectations around AI use were clear

Students are using AI extensively, expect it to matter professionally, see its benefits, but worry about becoming dependent on it.

Choose your path

Explore the Report by What Matters to You

Explore MIT’s recommendations and reasoning by topic. Each section distills the parts of the report most relevant to that issue, with references back to the original sections.

Make learning goals “AI-aware” §3.1.1 · Revisit course goals

MIT argues that every course should reconsider its learning goals in light of AI. That does not mean abandoning longstanding objectives simply because AI can perform the task. Some capabilities may remain fundamental and require new ways of protecting and assessing them; others may become less important; and AI may make entirely new learning goals possible.

The report recommends backward design: identify what students should know and be able to do, then design instruction, assessment, and AI use accordingly.

Begin with what students need to be able to do, not what AI can do.
Decide what students need to master for themselves §§2.4–2.5, 3.1.1–3.1.2

AI can produce a successful output without the student developing the knowledge or capability that the assignment was designed to build. MIT therefore urges instructors to think carefully about which skills and knowledge students must personally possess.

The answer will vary by discipline and stage of learning. Some fundamentals may need to be practiced independently before AI becomes useful; elsewhere, using AI may itself be part of competent professional performance.

What must students themselves understand or be able to do, even when AI can do the task?
Teach students to use AI effectively §3.2.4

MIT defines effective AI use as much more than knowing how to write prompts. Students need to learn how to specify a problem, evaluate an output, verify claims, recognize when models may hallucinate or fail, and decide when AI is—or is not—the right tool.

That makes judgment central to AI proficiency. The goal is not simply to produce better AI outputs, but to understand the capabilities and limitations of the system well enough to remain responsible for the quality of the work.

Teach responsible and ethical AI use §3.2.4

MIT distinguishes among effective, responsible, and ethical AI use.

Responsible use includes understanding when AI is augmenting human capability versus replacing learning, taking responsibility for AI-assisted work, and disclosing AI’s contribution appropriately. Ethical use raises broader questions about bias, training data, intellectual property, authorship, environmental and resource costs, and the potential homogenization of language and ideas.

MIT recommends addressing these issues throughout the curriculum rather than treating them as a separate compliance exercise.

Develop AI capability within disciplines §3.2.4

MIT does not envision AI literacy as something that can be handled once in orientation or a single generic course. It recommends introducing foundational AI literacy and then connecting it to disciplinary practices throughout students’ education.

The report suggests possibilities such as AI-intensive courses within majors, AI components in capstones, and discipline-specific instruction on appropriate AI use in writing, research, and professional practice.

The implication is that sophisticated AI use will look different across fields—and students need opportunities to learn those differences.

Raise the ceiling on what students can accomplish §§3.1.1 and 3.1.3

MIT emphasizes AI’s potential to enable learning experiences that previously would have been too complex or time-consuming.

Its software-engineering example is especially concrete: AI can allow students to build much more sophisticated systems within a semester, creating room to focus on design choices, experimentation, validation, and how systems perform in realistic settings. The report makes similar observations about architecture and other disciplines.

AI can allow educators to ask students to undertake more ambitious work.
Assess capability, not just completed work §§3.1 and 3.1.2

Generative AI can now produce credible responses to many traditional written assignments, including essays, problems, proofs, and code. MIT argues that this weakens the assumption that the quality of a submitted artifact necessarily reveals what an individual student knows or can do.

Assessment therefore needs to provide stronger evidence of the capability the course is intended to develop.

The challenge is not simply determining whether AI was used. It is determining whether the student achieved the intended learning outcome.
Don’t make high-stakes exams the whole answer §3.1.2

MIT understands why instructors are increasing the weight of controlled, in-person assessments. They provide clearer evidence of individual performance when outside work can be heavily AI-assisted.

But the report warns of a tradeoff. If too much assessment moves into short, high-stakes settings, students may have less incentive to invest in the sustained problem sets, projects, and extended work through which deeper understanding develops.

MIT is not arguing against exams. It is warning against allowing the need for AI-resistant assessment to narrow what counts as meaningful learning.
Use richer evidence of understanding §§3.1.2 and 3.1.6

MIT encourages experimentation with forms of assessment that provide different evidence of mastery, including oral exams, semester portfolios, presentations, and out-of-class assignments paired with in-class conversations.

One particularly useful model is to let students undertake substantial work outside class and then require them to explain, defend, extend, or apply it in person.

Rely on multiple forms of evidence rather than assuming that a polished final product by itself demonstrates understanding.
Make the learning process more visible §3.1.9

MIT points toward approaches that reveal how student work develops rather than focusing exclusively on the final submission.

These can include intermediate deadlines, drafts, regular project meetings, version histories, and opportunities for feedback during the work. Such evidence can help instructors understand how a project evolved and whether students can explain their decisions and revisions.

This approach serves two purposes: it can provide better evidence of individual engagement and create opportunities to improve learning before the final submission rather than merely evaluate it afterward.

Reconsider grades and incentives §3.1.6 · Reconsider grades and incentives

MIT does not recommend eliminating grades or rationing top grades. It does ask whether conventional grades should remain the dominant way of signaling mastery as education becomes more project-based and experiential.

The report points to competency- and mastery-based assessment, portfolios, oral feedback, and substantive project evaluation as approaches worth exploring.

It also raises an incentive problem: when students are intensely focused on maximizing grades, AI provides a powerful way to optimize the submitted product—even when doing so bypasses some of the learning the assignment was intended to produce.

Be cautious with detection and surveillance §3.1.9 · AI detectors and online exam platforms

MIT recommends against relying on AI detectors. It cites difficulty detecting nuanced forms of AI assistance, the possibility of false positives, potential disparate effects, and an arms race in which students use increasingly sophisticated tools to evade detection.

The report also worries about the effect of constant policing on instructor-student trust. Where secure individual assessment is necessary, MIT generally favors in-person proctored exams over current lockdown-browser approaches, while encouraging process evidence and clear expectations elsewhere.

The objective is credible evidence of learning—not simply increasingly sophisticated detection.
No one size fits all §2.6 · No one size fits all

MIT explicitly rejects a universal rule for AI use.

A poetry seminar, proof-based mathematics course, design studio, and software-engineering project may appropriately have very different relationships to AI. The same is true across levels of expertise: a novice developing foundational knowledge is in a different position from an advanced student using AI within a field they already understand deeply.

The relevant question is what role AI should play for this learning objective, in this context, for these students.
Let the learning objective determine the rule §§2.5 and 3.1.8

MIT recommends determining AI policy through educational purpose.

AI might appropriately be prohibited for one activity, available only as a support tool for another, and required for a third. What matters is whether its use supports or undermines the learning goal.

The report places particular emphasis on explaining the rationale to students. A rule such as “no AI” communicates a restriction; explaining that students need to develop a particular capability independently communicates what learning the restriction is intended to protect.

Prefer augmentation to automation §2.7 · Augmentation not automation

MIT’s guiding principle is to use AI in ways that expand student capability without replacing the thinking through which learning occurs.

It describes the goal as “pro-learner” AI: technology that helps students tackle new tasks, explore more deeply, receive useful support, and develop expertise while keeping the human intellectually engaged.

The distinction is whether AI helped the student do better thinking or largely did the thinking the student needed to learn to do.
Preserve intentionally AI-free learning where it serves a purpose §§2.6, 3.1.8 and 3.1.10

MIT is not arguing that AI should be incorporated into every learning experience.

There are contexts in which independent work is necessary precisely because the student needs to develop a capability without technological assistance. The report also suggests taking seriously students who prefer to minimize AI use where the subject permits it and even raises the possibility of “AI Light” and “AI Heavy” curricular pathways.

AI-free learning should serve a pedagogical purpose, not simply reflect resistance to the technology.
Require accountability for AI-assisted work §§3.1.8, 3.2.4 and 3.2.6

For MIT, AI remains a tool under human direction.

Students remain responsible for checking factual accuracy, identifying errors or bias, complying with course and professional rules, and appropriately disclosing AI’s contribution. The report also recommends AI-use statements in theses and notes that research communities may establish different standards.

MIT additionally suggests asking students to reflect on when AI helped or harmed their learning or the quality of their work.

Disclosure can become an opportunity to develop metacognition about human-AI collaboration.
Productive struggle §§2.4 and 2.7

MIT argues that difficulty is not merely an obstacle to efficient education. Working through confusion, failed approaches, revision, uncertainty, and effort is often how students develop durable understanding, confidence, persistence, and judgment.

AI makes it much easier to bypass that process by providing a plausible answer immediately. The report describes a danger of “cognitive surrender” when students turn to AI at the first hint of struggle.

The educational challenge is to distinguish friction that adds little value from productive struggle that is itself part of learning.
Learning with other people §§3.1.4 and 3.2.1

MIT reports signs that AI may be changing longstanding patterns of social learning, including office-hour participation and informal study groups.

The report therefore recommends regular structured human interaction: collaborative problem solving, group projects, peer feedback, discussion, and contact with instructors and TAs.

These interactions matter not only because classmates and instructors can explain content. They also develop communication, collaboration, confidence, the ability to receive criticism, and the habits required to participate in a professional and intellectual community.

Human interaction is itself an educational outcome.
Mentorship and apprenticeship §3.1.5 · Preserve and expand out-of-class research and career experiences

MIT makes a particularly strong argument around undergraduate research.

AI agents may sometimes perform work more quickly or cheaply than novice student researchers. But MIT argues that research experiences exist to educate, not merely to supply faculty with labor.

Through apprenticeship, students learn how questions are formed, mistakes are interpreted, disagreement is handled, judgment develops, credit is shared, and knowledge is produced collectively.

Replacing novice participation with AI might improve short-run efficiency while eliminating the experience through which novices become capable professionals.
Human feedback and teaching relationships §3.2.3 · Instructor disclosure around AI use

MIT extends its discussion of AI norms to instructors.

Students told the committee that they notice when instructors use AI for slides, feedback, grading, or other teaching activities—particularly when students themselves face restrictions. MIT recommends transparency about substantial instructor use of AI.

The report is especially cautious when students invest significant effort in work but receive only machine-generated feedback in return.

What kinds of human attention and investment should students reasonably expect from their teachers?
Community and the educational “social contract” §§2.8 and 3.2.1–3.2.2

MIT worries that AI can reinforce a transactional view of education in which assignments are outputs, teachers are evaluators, peers are optional, and credentials are things to optimize as efficiently as possible.

The report instead describes education as participation in a community through which students develop norms, identity, relationships, judgment, responsibility, and a sense of agency.

It calls this alignment among students, instructors, and the institution a kind of “social contract”: a shared understanding of why learning matters and what members of an intellectual community owe one another.

Human judgment and agency §§2.7–2.8

Ultimately, MIT’s goal is not to protect students from AI. It is to prepare them to use increasingly capable systems without becoming passive users of them.

The report repeatedly emphasizes human capacities such as critical thinking, initiative, creativity, communication, collaboration, judgment, and responsibility for consequences.

AI fluency therefore involves more than knowing how to obtain useful outputs. Students must learn when to question the system, when to reject its recommendation, when to work independently, and how to remain the responsible decision-maker in an AI-assisted process.

Give faculty real support for course redesign §3.3.3 · AI Fellows and an AI Implementation Team

MIT describes meaningful adaptation of courses and assessments as a substantial undertaking.

It recommends implementation support combining expertise in AI, technology, and learning science, with people who can work directly with individual instructors as well as develop broader tools and practices.

If institutions expect faculty to make substantial changes to teaching and assessment, they must provide meaningful expertise, time, and implementation support.
Fund experimentation §§3.1.10 and 3.3.4 · Responsible experimentation and an AI Pilot Fund

MIT argues that institutions do not yet know enough to prescribe one settled model for AI-aware education.

It recommends deliberately experimenting with AI-enabled, AI-aware, and even intentionally AI-free learning experiences—and evaluating what happens.

To make this feasible, MIT proposes pilot funding for resources such as AI credits, TAs, undergraduate research assistance, and summer support. It also calls for faster pathways for curricular experimentation.

Make it practical for faculty to try substantive ideas rather than merely make small adjustments around the edges.
Build communities of practice §3.3.5 · Ongoing training and instructor support

Faculty and instructors told MIT that they wanted opportunities to learn from colleagues about what was working, what was failing, and how others were adapting.

The report recommends regular peer-learning opportunities such as lunch-and-learns, hands-on training, workshops, and communities of practice within and across disciplines.

It also explicitly cautions against simply “checking the box” with generic third-party AI training.

MIT treats the university itself as a learning community in which instructors experiment, compare experiences, and improve together.
Create ongoing leadership and governance §§3.3.1–3.3.2

MIT argues that a one-time committee and report cannot resolve a challenge evolving as rapidly as AI.

It recommends an ongoing AI-and-education committee along with school-, college-, or department-level AI Leads to support curricular planning, policy coordination, and adaptation.

AI adaptation needs to become an ongoing organizational capability rather than a temporary initiative.
Measure what is happening—and keep revising §3.3.6 · Develop metrics

MIT’s first guiding principle is humility: today’s solutions will inevitably need revision.

The report therefore recommends tracking AI use, student engagement, satisfaction, and post-graduation feedback, among other possible measures.

The larger point is methodological. Institutions should not treat their first generation of AI policies and teaching practices as permanent answers. They should gather evidence, identify unintended effects, compare experiences, and revise.

AI policy becomes a cycle of experimentation, measurement, learning, and improvement.
Ensure equitable, secure access to capable AI §§3.3.7–3.3.9

MIT argues that access to AI is becoming an educational-equity issue. Students who can personally afford the most capable commercial systems may have advantages over classmates who cannot, particularly when AI use is permitted or expected.

The report therefore recommends institutionally supported access to capable tools while also addressing privacy, sensitive data, model choice, logging, and auditing.

If AI becomes part of education, unequal access to AI can become unequal access to educational opportunity.

The framework underneath

8 Quick Principles

Questions for us

Implications for the Moore School

For the Moore School, several questions raised by MIT's report seem especially worth considering.

  1. 01What should a Moore graduate still be able to do without AI?
  2. 02What should a Moore graduate be able to do with AI that wasn't previously possible?
  3. 03Which of our current assessments no longer tell us what we think they tell us?
  4. 04What kinds of productive struggle should we deliberately preserve?
  5. 05Which human capabilities become more important as AI becomes more capable, and how do we cultivate them?
  6. 06How should we prepare students for the AI-enabled workplace they are entering?

Audio briefing

Listen to the Briefing

26-minute audio discussion

An AI-generated discussion of MIT’s report, created with Google NotebookLM from the original report.

AI-generated audio briefing created with Google NotebookLM.

Continue Exploring