Generative AI has intensified concerns about academic integrity in higher education, but it has also raised a more basic question: What counts as evidence of learning? Assessment professionals have been asking that question for years. AI has made it more urgent. When students can produce polished written work in seconds, the finished product may no longer tell us enough about what a student knows or can do.
Much of the early conversation around generative AI focused on detection and prevention. Faculty questioned whether traditional assignments, especially take-home essays and discussion posts, could still be trusted as evidence of student learning. Colleges revisited academic integrity policies, looked at AI-detection tools, and reconsidered proctoring.
That response made sense. Faculty were being asked to judge student work at a time when it had become much harder to know how that work was produced. Concerns soon emerged about the reliability of AI-detection tools and the risk of false positives. Policies were revised while institutions were still trying to determine what appropriate AI use should look like across disciplines. Remote proctoring returned to the conversation as well, bringing many of the same privacy, equity, and effectiveness concerns that existed before generative AI.
Detection only gets us so far. If students can submit acceptable work without fully engaging in the learning the assignment is meant to measure, assessment design becomes part of the problem. What are we asking students to do, and what can the assignment tell us about what they have learned?
AI has exposed some of the limits of assessment practices that rely heavily on the finished product. For a long time, a final essay, completed problem set, or discussion response could serve as a reasonable proxy for learning. That was never perfect, but in many settings it was considered sufficient.
That assumption is harder to sustain now. A polished response may reveal very little about how the student arrived there. We may not see how the student framed the problem, worked through confusion, revised an argument, responded to feedback, or applied what they learned in a new context. Those parts of the process can tell us things that a final submission cannot.
These concerns are familiar in the assessment field. Authentic assessment, formative feedback, reflective practice, and competency-based approaches have all tried, in different ways, to get closer to the learning itself. We have long asked whether an assessment gives faculty enough information to determine what students know and can do. AI did not create that problem. It has made it much harder to ignore.
One response is to give faculty more opportunities to see learning as it develops. That does not mean abandoning essays, exams, or other traditional forms of assessment. It means thinking more carefully about what surrounds the final product and what else might help faculty understand how students got there.
Students might explain why they took a particular approach, submit a draft and discuss what changed, apply a concept to an unfamiliar context, or reflect on where they struggled and how they worked through it. Oral presentations or defenses give students a chance to explain their reasoning in real time. Revision portfolios and process journals can show how ideas develop across drafts. Project-based work asks students to apply knowledge and make decisions within a specific context.
Metacognitive reflection can also help when it asks students to examine their learning rather than simply describe what they did. Questions about how their reasoning changed, how they used feedback, or why they made a particular decision can reveal aspects of learning that might otherwise remain invisible.
None of these approaches makes inappropriate AI use impossible. The goal is to give faculty a clearer view of student learning.
It helps to distinguish between different kinds of AI use. Asking AI to do the work for a student is different from using it as part of the learning process.
A student might use AI to test an argument, consider another perspective, clarify a difficult concept, or get feedback on a draft and still be actively engaged in the work. The key question is whether the student is still doing what the assignment is meant to assess.
AI can also support faculty work. It can help generate reflection prompts, clarify rubric language, summarize recurring themes, or reduce time spent on routine tasks. Used this way, it can support teaching and assessment while leaving interpretation and judgment with the faculty member.
|
The question is not whether AI belongs in learning — it is whether educators are designing the conditions under which it supports rather than replaces genuine intellectual engagement. |
Interpreting student learning still requires disciplinary and contextual judgment. An automated system might flag something in student writing that an experienced instructor understands differently because of disciplinary conventions, a student's stage of development, or the linguistic and cultural context of the work.
Faculty still decide what findings mean, whether they matter, and whether they call for a response. AI may help review large sets of responses, draft rubric language, or provide basic clarification. It cannot determine the significance of those findings on its own. Faculty expertise remains central to understanding student development and deciding how assessment results should inform teaching.
Efficiency can be useful, but it is not the same as understanding. AI may make some parts of assessment faster. Faculty judgment is still what gives the results meaning.
The implications go beyond individual assignments. As assessment practices change, institutions also need to consider how different forms of student work are documented and interpreted across courses and programs.
Many institutional assessment processes were built around familiar forms of evidence. Student work was collected, rubric scores were aggregated, results were reviewed, and programs documented what they learned. That becomes more complicated when assessment includes oral defenses, portfolios, project work, process documentation, reflections, and other forms of student learning.
Collecting more is not necessarily the answer. Institutions need to decide what information is useful, how much is enough, and how different forms of student work can be interpreted across contexts. Any approach also needs to account for faculty workload and avoid turning better assessment into more documentation.
A useful first step is to review current assessment practices and identify places where a single final product is carrying too much weight. Programs can then consider where seeing more of the learning process would improve the assessment. Some assignments may already provide exactly what faculty need. Others may benefit from another opportunity to see how students are reasoning or applying what they know.
Institutions should also look at how different forms of student work are documented. Existing systems and processes may need to accommodate assessment approaches that do not fit neatly into a traditional rubric score or written artifact. More documentation does not automatically lead to better assessment, so any changes should be tied to a clear purpose.
Accreditation and program review belong in these conversations too. As the evidence of student learning changes, institutions need to consider how those changes fit within existing expectations and how they will show the connection between assessment findings and improvement.
The purpose is to help faculty and programs make better decisions about teaching, curriculum, and student support.
Institutions will still need academic integrity policies and clear expectations for AI use. Determining whether a student used AI, however, cannot be the whole response. This is also an opportunity to look closely at the assessments already in place and ask whether they are giving faculty the information they need.
What does an assignment tell us about student learning? What parts of the student's reasoning can faculty see? Could a student produce the expected answer without developing the intended knowledge or skill? What else would help faculty make a more informed judgment?
These are assessment questions. AI has made them harder to put off.
Higher education has been asking for a long-time what counts as good evidence of learning. That question matters even more now.