Rethinking Assessment in the Age of AI

When the Finished Assignment Tells You Less Than It Used To

For generations, higher education has relied heavily on the finished product.

The essay. The problem set. The take-home exam. The research paper.

Students complete the work, instructors evaluate it, and the resulting grade is expected to represent some level of understanding or mastery. Generative AI is making that relationship much more complicated.

In a recent Inside Higher Ed article, Emma Whitford examines recommendations from MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training. The committee argues that AI is disrupting foundational parts of the MIT educational experience and forcing instructors and institutions to reconsider how they teach, assess, and ultimately determine whether students have learned.

One reason is straightforward: generative AI can now produce plausible responses to many of the assignments higher education has traditionally used as evidence of learning.

That leaves colleges with a much bigger question than whether a student used AI.

If AI can produce the finished assignment, what should we actually be assessing?

MIT Is Asking Higher Education to Look Beyond the Grade

Assessment and grading are central to the MIT committee’s recommendations.

As Inside Higher Ed reports, rather than simply making conventional assessments harder to complete with AI, the committee asks MIT to reconsider the role grades play in education and explore alternatives, including competency-based approaches and other ways of demonstrating mastery.

The original MIT report goes even further.

It recommends approaches including oral exams, portfolios, out-of-class work paired with in-class conversations, project-based learning, and opportunities for instructors to observe progression rather than evaluating only an end product.

The common thread is important.

Assessment becomes less about one artifact and more about evidence of learning.

That distinction could become increasingly important as AI gets better.

A polished essay can demonstrate something. But instructors may also need to know what happened before the essay existed.

What did the student read? What did they notice? Which evidence did they choose? What confused them? How did they arrive at their interpretation? Could they explain why they reached that conclusion? What happened when someone challenged it?

Those moments tell us something a finished document alone may not.

AI Detection Doesn’t Solve the Assessment Problem

One response to generative AI has been to try to determine whether an assignment was written by a student or a machine.

But MIT cautions against relying on that strategy.

The committee raises concerns about the reliability of AI detectors and the consequences of false positives. It also argues that increased policing can create distrust between students and instructors. Inside Higher Ed describes the resulting tension as faculty feeling pressure to police AI use while students worry about being falsely accused.

This concern extends beyond MIT. Universities have begun moving away from AI detectors, with institutions and teaching centers instead encouraging faculty to focus on process, reasoning, authentic engagement, and assessment redesign.

The distinction matters.

AI detection asks: Did AI produce this?
Learning assessment asks: What does this student understand?

Those are not the same question.

And as generative AI becomes more sophisticated, the second may prove much more useful than the first.

Make the Learning Process Part of the Evidence

If the final product is becoming less informative on its own, instructors can create more opportunities to see what happens before students submit it.

Reading is one place to start.

Consider a student preparing to write an essay about a difficult academic text.

Before writing anything, that student has to make dozens of intellectual decisions.

They identify important passages. They distinguish arguments from evidence. They encounter unfamiliar ideas. They make connections. They question assumptions. They decide which information matters and eventually begin constructing an interpretation.

Traditionally, much of that work is invisible. The instructor sees the essay that comes out at the end.

Social annotation can make more of what happens before that point visible.

With Hypothesis, students can highlight passages, ask questions, explain ideas, identify evidence, and discuss interpretations directly alongside course materials.

The annotation itself does not prove mastery.

But it can provide another window into how a student is engaging with the intellectual work behind an assignment.

What Visible Learning Can Look Like

Imagine two students submit similarly polished essays.

If an instructor sees only the final papers, those artifacts become the primary evidence available for evaluation.

Now imagine that the course also includes collaborative annotation.

Before writing, one student identified three passages they considered central to the author’s argument. They asked a question about contradictory evidence. They responded to a classmate with a different interpretation. Later, they returned to their original position and refined it.

Suddenly, the instructor can see more than an answer.

They can see elements of a learning process.

This does not mean every annotation should become another item to grade. Nor does it mean social annotation should replace essays, exams, presentations, or other forms of assessment.

Instead, annotation can become part of a broader learning design in which students have multiple opportunities to demonstrate how they are reading, reasoning, questioning, communicating, and developing ideas.

Assessment Can Begin Before Students Start Writing

The AI conversation often focuses on what happens when students sit down to produce an assignment. But some of the most important learning happens earlier. Students need to understand the material before they can meaningfully write about it.

That makes reading an important part of the assessment conversation.

If a student can ask AI to summarize a text, formulate an argument, and generate a response, then instructors may want to design more intentional opportunities for students to interact directly with source material before moving to the finished product.

Social annotation asks students to do something with the reading itself.

Identify the sentence that changes the argument.
Explain why a piece of evidence matters.
Find something you disagree with.
Ask a question about a confusing section.
Respond to another interpretation using evidence from the text.

Those activities make reading an observable part of learning rather than simply something students are expected to have completed before the “real” assignment begins.

Social Learning Gives Instructors Another Kind of Evidence

MIT’s recommendations are not only about grades.

The committee also places substantial emphasis on social learning.

Inside Higher Ed reports that MIT’s committee heard concerns about declining peer study groups and office-hours participation and ultimately called for greater emphasis on social learning and human interaction.

That matters for assessment because understanding is often revealed through interaction.

Students demonstrate knowledge when they explain an idea to someone else.
They reveal uncertainty when they ask a question.
They exercise judgment when they disagree.
They develop arguments when they have to defend an interpretation.

And they may revise their thinking after encountering evidence or perspectives they had not previously considered.

Social annotation creates a space for those interactions directly around academic content.

Instead of discussion occurring only after reading is finished, students can engage with one another while they are still constructing meaning.

Independent Thinking Should Come Before Peer Influence

Social learning is valuable, but students also need opportunities to form their own ideas.

That balance has become particularly relevant in an environment where students can immediately ask AI what a text means before attempting to interpret it themselves.

Paced Social Annotation offers one approach.

Students first read and annotate independently without seeing their classmates’ contributions. At an instructor-defined checkpoint, annotations are revealed and the collaborative phase begins.

That sequence creates two different kinds of evidence.

First: What did this student notice and think independently?

Then: What happened when they encountered other perspectives?

The goal is not surveillance. It is an intentional learning design.

Students get space to develop an interpretation before peer influence, followed by an opportunity to test and expand that thinking through conversation.

Alternative Assessment Doesn’t Have to Mean More High-Stakes Assessment

There is another risk in responding to AI.

If instructors lose confidence in take-home assignments, the instinct may be to move everything toward supervised exams.

MIT specifically warns against relying too heavily on that solution.

The committee argues that an excessive shift toward time-limited, in-person assessment could reduce opportunities for the sustained projects and difficult intellectual work students need to build mastery.

That leaves institutions searching for a balance.

How do we preserve thoughtful, extended work while also gaining confidence that students are actually developing the skills the work is supposed to measure?

One answer may be to layer evidence.

An essay can still matter.

But so can the reading that preceded it.

The annotations.
The discussion.
The draft.
The oral explanation.
The peer feedback.
The student’s reflection on how their argument changed.

Instead of relying on a single artifact to represent learning, instructors can design a series of moments where students demonstrate different parts of the process.

AI Could Push Higher Education Toward Better Assessment

There is an interesting possibility buried inside MIT’s recommendations.

AI may expose weaknesses that existed long before ChatGPT.

A student could always skim a reading, memorize enough material for an exam, imitate the structure of a strong essay, or optimize for a grade without developing the deeper understanding an instructor intended.

Generative AI makes those gaps harder to ignore.

MIT’s committee describes the disruption as an opportunity to reconsider familiar approaches to teaching and testing rather than simply defending them.

That could move higher education toward a more useful question.

Instead of: How do we make the old assignment impossible for AI to complete?
Ask: What evidence would actually convince us that a student has learned?

For some outcomes, the answer may be an oral examination.

For others, a portfolio, presentation, project, experiment, or performance may make more sense.

And when the learning objective involves critical reading, evidence evaluation, interpretation, communication, or collaborative knowledge building, the student’s interaction with the source material itself may deserve a place in that picture.

Hypothesis Isn’t an AI Detector. That’s the Point.

Hypothesis does not attempt to determine whether AI wrote a student’s work.

Instead, it gives instructors another place to design for the behaviors they want students to practice.

Reading closely.
Asking questions.
Selecting evidence.
Developing interpretations.
Explaining reasoning.
Responding to other people.
Reconsidering an idea.
Those activities were valuable before generative AI.

They may become even more valuable now.

The Hypothesis LMS integrations allow social annotation to become part of existing course workflows in platforms including Canvas, Blackboard, D2L Brightspace, and Moodle.

The objective is not to add another layer of policing.

It is to give students more opportunities to demonstrate learning while learning is actually happening.

Maybe We Should Assess More Than the Answer

AI has made answers abundant. That does not make learning less important. It makes the path toward understanding more important to pay attention to.

MIT’s recommendations, as reported by Inside Higher Ed, suggest higher education may need to reconsider some of its most familiar assumptions about grades, assessment, and what counts as credible evidence of mastery.

That does not mean abandoning essays or grades tomorrow. It means recognizing that a final product may no longer tell the whole story.

Students still need to read difficult material, wrestle with ideas, evaluate evidence, communicate with other people, revise their thinking, and develop judgment of their own.

The challenge for course design in the age of AI is to make more of that learning matter.

And, whenever possible, make it visible.

Related Resources

• Inside Higher Ed: MIT AI Report Calls for Alternative Grading, More Social Learning
• MIT Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training
• Hypothesis for Higher Education
• Paced Social Annotation
• Generative AI and Social Annotation Case Study
• Hypothesis LMS Integrations