The New Campus Divide: Who Can Use AI and Who Can Judge It

Generative AI is forcing universities to confront a problem deeper than plagiarism: if students can produce polished essays, code and explanations with machine assistance, what still counts as credible evidence that they have learned? A new study titled "Examining Generative AI Disruption: A Repeated Cross-Sectional Study of Faculty and Staff Sensemaking in Higher Education," published in Algorithms by Trini Balart, Gibin Raju and Kristi J. Shryock, captures how that question evolved during the first two years after ChatGPT entered academic life.

Based on open-ended responses from 104 faculty and staff in 2023 and 313 in 2024 at a large U.S. public research university, the study finds a noticeable shift in emphasis. Early responses centered on uncertainty, academic-integrity risks and defensive efforts to protect existing assessment systems; a year later, respondents more often discussed redesigning assignments, teaching students to evaluate AI output and treating AI literacy as an emerging academic and professional competence.

The findings suggest that higher education is moving from asking how to keep AI out of student work toward asking how to preserve learning, judgment and accountability when AI is already inside it.

AI Is Forcing Universities to Redefine What Counts as Learning

Assessment remained the most prominent issue in both survey years. It appeared in 34.6% of the analyzable responses in 2023 and 31.6% in 2024, but the substance of those discussions changed. In the earlier survey, instructors frequently worried that take-home writing, coding exercises and even open-book assessments could no longer reliably demonstrate whether students understood the material themselves.

The immediate response was often defensive: more proctored examinations, oral explanations, presentations and supervised tasks. By 2024, however, faculty were more frequently considering assignments that incorporated GenAI while requiring students to critique answers, verify information, explain their reasoning and reflect on how AI had been used. The goal was shifting from preventing AI assistance to making students' judgment visible despite it.

This is potentially one of the most consequential changes GenAI could trigger in higher education. Universities have traditionally relied heavily on the finished product, an essay, report, code submission or solution, as evidence of competence. When machines can contribute substantially to that product, assessment may increasingly need to capture the process: drafts, oral defenses, source verification, reasoning, revisions and explanations of why particular choices were made.

This does not mean every course should become "AI-integrated." The study stresses that exposure differs by discipline and task. Writing and programming were often seen as immediately vulnerable, while some hands-on, mathematical, design or clinical work was viewed differently, reinforcing the need to align assessment reform with actual learning outcomes rather than impose a universal AI policy across every field.

The Integrity Debate Is Moving Beyond Cheating

Academic integrity remained a persistent concern, but by 2024, it had expanded beyond the familiar question of whether a student had cheated. Faculty increasingly raised issues of authorship, attribution, disclosure, acceptable collaboration and responsibility for claims produced with AI assistance. Integrity-related concerns accounted for 10.6% of coded responses in 2023 and 13.7% in 2024.

The expansion reflects the difficulty of treating all AI use as equivalent. A student might use GenAI to brainstorm, translate, revise prose, summarize material, generate code or produce a complete answer. Each use affects the intended learning task differently, meaning a blanket instruction such as "do not use AI" may be less useful than clearly stating which forms of assistance are acceptable and what must remain the student's own intellectual work.

The study also highlights the limits of solving this problem through detection. Faculty reported greater workload and uncertainty while trying to determine whether student submissions were authentic, while the authors caution that AI-detection systems bring reliability and fairness concerns of their own. The more durable response is likely to combine explicit expectations, redesigned assessment, student education and fair procedures for handling suspected misconduct.

More importantly, AI literacy is beginning to look less like proficiency with prompts and more like a form of intellectual responsibility. Students need enough disciplinary knowledge to recognize inaccurate or misleading output, identify bias, verify consequential claims and know when a system should not be trusted. In that sense, GenAI may actually increase the value of judgment even as it reduces the effort required to generate a first answer.

Classrooms Are Adapting Faster Than University Governance

Faculty experimentation appears to be advancing faster than coordinated institutional policy. In 2023, respondents often described being left to decide individually how AI should affect assessment, authorship and enforcement. By 2024, requests had become more specific: faculty wanted clearer acceptable-use rules, stronger support for assessment redesign, better professional development and access to capable AI tools.

The policy challenge is that neither extreme offers an easy solution. Highly centralized rules may become obsolete as AI capabilities change and may ignore major disciplinary differences. But fully decentralized governance can leave instructors and students facing conflicting expectations from one course to another while individual faculty are forced to resolve ethical, technical and legal questions without sufficient support.

The study points toward layered, adaptive governance. Universities can establish baseline principles around transparency, privacy, responsibility, equity and due process, while departments and instructors translate them into discipline- and task-specific expectations. Policies would need to be revisited continuously rather than treated as permanent rules written for a technology that is changing faster than institutional review cycles.

Higher education is effectively becoming an early test of how institutions govern rapidly evolving AI systems more broadly: balancing common safeguards with local experimentation, protecting human accountability without freezing innovation, and updating rules as technologies and user behaviour change.

The Next AI Divide Will Be About Access, Judgment and Work

Equity concerns became more explicit between the two surveys. Respondents increasingly distinguished between free and paid AI systems and worried that students with greater financial resources could gain access to more capable models, better computing infrastructure and more time or training to learn how to use them effectively. Access-related concerns rose from 1.9% of coded responses in 2023 to 3.2% in 2024.

However, equal access to a tool does not create equal capacity to benefit from it. Students and faculty also differ in training, technical assistance, digital infrastructure and the ability to evaluate outputs critically. Universities that provide licenses without building ethical reasoning, verification skills and disciplinary AI literacy may simply increase usage without improving learning.

The concern becomes even more strategic when education meets the labor market. Workforce-related themes appeared in 12.5% of 2023 responses and 16.6% in 2024. Rather than predicting universal job replacement, later responses more often envisaged hybrid professional roles in which humans supervise, interpret and verify AI-supported work, but faculty also worried that entry-level tasks traditionally used to build professional experience could disappear first.

This creates a hidden risk for universities and employers. If AI automates routine tasks before novices have mastered them, productivity could improve while expertise formation weakens. The study therefore argues for durable capabilities, problem decomposition, evidence evaluation, verification, communication and responsibility, rather than narrow training on specific platforms that may quickly become obsolete.

The research comes from one institution with substantial engineering representation, relies on voluntary responses and compares independent samples rather than following the same people over time. It captures perceptions and expectations, not direct evidence that learning, employment or institutional outcomes changed because of GenAI.

To sum up, generative AI appears to be pushing higher education away from a narrow policing problem and toward a deeper redesign of learning, responsibility and professional preparation. The key challenge is no longer whether universities can stop students from using AI, but whether they can create environments in which students remain capable of thinking independently when intelligent assistance is always available.