Can AI Help Colleges Identify Course Barriers?

Every university improves courses after students encounter challenges. Faculty review student feedback, analyze performance and revise learning experiences after barriers become apparent.

But what if institutions could identify some of those barriers before the first student ever opens a course?

As higher education explores the role of artificial intelligence (AI), much of the conversation has focused on improving efficiency, generating content and supporting student success. Another emerging opportunity is whether AI can help institutions identify course-design barriers before students encounter them.

Every course contains barriers its authors never intended and often never knew existed. A lesson may use language that is clear to a subject matter expert but confusing to someone encountering the topic for the first time. It may contain an unrecognized cultural assumption or introduce concepts in an order that leaves learners without the foundational knowledge they need.

Traditional course reviews can identify many of these issues. Yet no instructional team can fully represent the range of language backgrounds, cultural experiences, knowledge levels and life circumstances students bring to a learning experience.

AI may help institutions examine course material from additional perspectives earlier in the design process. The goal is not to replace educators or student feedback. It is to determine whether AI can help instructional teams ask better questions before unintended barriers affect real learners.

Rethinking the Idea of a Synthetic Learner Agent

The concept began with a simple question: How do we know a course will work for real students before the first learner experiences it?

At WGU, that question led our team to explore whether a single AI model could account for the many factors that shape a learner’s experience. As the research progressed, the team shifted toward specialized agents focused on individual dimensions of the learning experience.

One agent may test how a student agent profile experiences course language. Others examine cultural context, accessibility, emotional friction, or the logical structure of learning. Each agent performs one task. Together, they create a more complete picture of how different students may experience the same learning material.

Throughout the development process, our team learned that a general model pretending to be four specialists is not four specialists.

An AI System Designed to Ask Better Questions

Specialized agents also require a way to determine which perspectives are relevant to a particular piece of content.

A lesson containing complex language may need to be evaluated by language-focused agents. Content discussing family relationships may warrant review for cultural context or emotional response. A technical lesson may benefit from agents designed to evaluate prerequisite knowledge and concept sequencing.

An orchestration layer can determine which agents should evaluate a specific piece of content. Because each agent performs a narrowly defined function, its findings can be transparent and traceable. Instructional teams can evaluate where potential issues are identified, why they matter and which recommendations may warrant further review.

Consider the sentence:

“If you’re struggling with this material, reach out to your instructor or a trusted family member for help.”

Most reviewers would likely see a supportive message.

 Specialized learner agents may identify several opportunities for improvement.

A language-focused agent may recognize that “reach out” is an idiom some learners find difficult to interpret. Another may identify the assumption that every student has a trusted family member. A third may observe that “struggling” is subjective and offers little guidance about when or how students should seek support.

None of these observations necessarily mean the sentence is inappropriate or must be changed. Instead, they give instructional teams additional questions to consider. An educator may decide the original language is appropriate, revise it or provide more specific guidance about available support.

The value is not in treating every observation as a required correction. It is in giving educators a structured way to examine assumptions that may otherwise go unnoticed.

Testing Knowledge, Not Just Words

The approach can also evaluate something standard content-review tools may not: the structure of learning itself.

A course can be clearly written and still create barriers for students if concepts are introduced in the wrong sequence.

For example, a lesson may ask students to interpret nutritional information before explaining vitamins and minerals. Every sentence may be accurate, but the learning sequence may still be incomplete.

That distinction matters because content can be accurate at the sentence level while the overall learning path remains difficult to follow. By identifying where concepts depend on knowledge that has not yet been introduced, agents may help instructional teams review the connections between ideas, not just the ideas themselves.

Structure-focused agents can analyze relationships among concepts, identify missing dependencies and evaluate whether ideas appear in a logical progression.

 While this work is still being validated, the broader question is whether AI can help extend quality assurance beyond content review and into learning design.

Learning design still depends on context, disciplinary expertise and an understanding of what students are expected to know at each stage of a course. Faculty and instructional designers remain responsible for determining whether an identified issue is meaningful and whether a change would improve the learning experience.

Establishing Safeguards

The value of synthetic learner agents will depend on how carefully institutions design, govern and validate them.

People must remain responsible for every meaningful decision. AI supports educators rather than replacing them. Agents may identify potential issues and recommend improvements, but instructional designers, subject-matter experts and faculty determine which recommendations should be implemented.

Using synthetic learner profiles rather than real student information may offer privacy advantages by reducing the need to expose sensitive student data during testing. Models should also undergo ongoing bias evaluations, expert reviews and human validation. Their findings should be understandable, open to challenge and evaluated against evidence from actual learners.

Most importantly, synthetic findings must be compared with actual student outcomes. A model’s observation remains a hypothesis until institutions determine whether it reflects real learner experiences. That validation will help establish where the approach can support instructional design, where refinement is needed and where the technology may not provide useful insight.

For colleges considering similar work, safeguards should not be treated as a final review conducted after a system has been built. Decisions about privacy, bias, transparency and human responsibility must shape the system from the beginning.

The Larger Opportunity for Course Quality 

At WGU, where approximately 77% of students belong to one or more underserved or vulnerable populations, identifying barriers before a course launches has the potential to improve learning experiences at meaningful scale. The next phase of the work is validating synthetic learner agent findings against actual student outcomes to ensure the models accurately reflect real learner experiences and where additional refinement is needed. Throughout that process, synthetic feedback should complement, not replace, feedback from real students, accessibility reviews, learning science or faculty expertise.

The measure of success, at WGU and for any institution pursuing this work, should not be how many issues an AI system flags. It should be whether its findings help educators make better decisions and contribute to stronger learning experiences.

The larger opportunity is not simply another application of AI. It is a chance to change when institutions learn about course quality. Feedback from real students will always be essential, but colleges may not have to wait for students to encounter every barrier before they can act. AI will not define educational quality or replace the people responsible for it. If validated and governed carefully, however, it could help institutions give more learners a clearer path from the moment they log in.

Originally published in University Business.