AI in Higher Education - September 18, 2026

September 18, 2026 · Vol. 33

A weekly brief for the Management Department, Isenberg School of Management, UMass Amherst. By Matthew D. Langenkamp / 雷邁德, prepared in collaboration with Thea 🪻✨.

← Back to archive · ← Back to Dictionary

AI Permission and Evidence of Learning

AI permission specifies whether and how a student may use artificial intelligence for an assignment. Evidence of learning is work or performance that demonstrates the knowledge or skill being assessed. Permission defines the conditions of an assignment; it does not establish that learning has occurred.

This week’s reports concern three related matters: rules for authorized AI use, evidence of educational benefit, and the effects of assessment design. Florida approved rules for its state colleges. Researchers discussed the limits of current evidence on AI-supported learning. A humanities professor questioned whether shorter, controlled assessments preserve the skills developed through extended writing.

For management education, these distinctions apply to case analyses, valuations and recommendations. A finished report shows the quality of a product. An explanation or a new application can provide additional evidence of the student’s understanding. Neither necessarily measures every course objective.

Executive Summary

Florida: The Board of Education approved rules on September 16 requiring the 28 Florida College System institutions to adopt AI policies. These policies must prohibit student AI use on graded assignments or assessments unless the instructor explicitly permits it. The report concerns state colleges, not the separate public-university system.
Research: September 14 reporting in Inside Higher Ed describes limited and context-dependent evidence of AI’s educational benefits. Use, satisfaction and learning are distinct measures.
Assessment: In a September 17 opinion essay, Michael Genovese argues that replacing extended writing with short, controlled exercises may reduce opportunities for sustained analysis and revision. His essay is not a causal study.
Institutional practice: Grand Canyon University’s current guidance emphasizes faculty evaluation and demonstrated understanding rather than reliance on AI-detection software to determine integrity violations. The guidance is undated and is included as background.
Governance: The University of Colorado describes continuing review and campus implementation of its AI policy, effective August 24. The September report concerns implementation, not a new adoption.

1. Permission: Florida’s College Rules

WFSU reported on September 17 that the Florida Board of Education had approved rules requiring the state’s 28 public colleges to adopt AI policies. Those policies must prohibit students from using AI on graded assignments or assessments unless the instructor explicitly permits it.

The report also describes K–12 provisions, including parental consent and a school-district deadline. Those provisions should not be attributed to colleges. The college rules are also distinct from the regulatory process for Florida’s public universities. This brief relies on WFSU’s account; the legal text and college implementation dates were not independently reviewed.

The reported requirement establishes a condition of authorized use. It does not prescribe a method for assessing understanding. An assignment may permit AI and still require students to explain their calculations or defend their conclusions.

Why it matters for Isenberg: Assignment instructions can distinguish among brainstorming, language assistance, spreadsheet troubleshooting, source discovery and drafting. Naming the permitted activity is more precise than referring to AI use in general. The evidence required to demonstrate learning should be stated separately.

2. Educational Benefit: What the Evidence Supports

Educational benefit is an improvement in a specified learning outcome attributable to an instructional practice. Adoption rates and user satisfaction may inform an evaluation, but do not establish that improvement.

Inside Higher Ed’s September 14 report examines the evidence behind institutional investment in AI. Interviewed researchers describe promising and concerning findings, limits on generalization, and difficulties in evaluating tools that change rapidly.

The article’s title, “The ‘Nonexistent’ Research on AI’s Benefits for Education,” does not mean that no research exists. The report discusses existing studies and disagreement about what they establish. It is reporting based on interviews, not a systematic review conducted for this brief.

One distinction concerns the product and its instructional use. A chatbot may provide hints, supply complete answers, question an explanation or correct an error. These uses may have different effects even within the same product. A useful evaluation therefore specifies the activity as well as the tool.

Smaller classroom evaluations can inform local decisions. Their conclusions remain subject to the design of the study, the students involved and the outcome measured. A result in one course does not establish effectiveness across disciplines.

Why it matters for Isenberg: A pilot might examine whether students who use AI to critique a valuation can later identify a flawed assumption in a new valuation without assistance. Completion time and satisfaction are supplementary measures. Differences in prior preparation, instruction and task difficulty also require consideration.

3. Assessment: Sustained Writing and Independent Performance

Assessment validity concerns whether evidence supports the intended interpretation of a student’s performance. An assessment can restrict AI access yet fail to measure an important course objective.

In “The Cures for AI Are Killing the Humanities,” published September 17, University of Kentucky professor Michael Genovese argues that replacing extended essays with shorter handwritten work or oral examinations may reduce opportunities for sustained argument and revision.

His opening account of a graduate-admissions meeting in 2030 is hypothetical. It illustrates his concern; it is not evidence of an observed outcome. The essay offers an argument about assessment design rather than a measured estimate of its effects.

Extended writing allows time to gather evidence, reconsider an interpretation and revise an argument. A brief examination can reveal immediate understanding but may not assess those processes. The methods serve different purposes.

Why it matters for Isenberg: A strategy report and a short defense can provide complementary evidence. The report permits sustained analysis; the defense tests explanation and response to questions. The latter should not be treated as a complete substitute for the former when research and revision are course objectives.

4. Verification: Understanding and Academic Integrity

Verification of understanding is the examination of a student’s ability to explain or apply the knowledge represented in submitted work. It is distinct from determining whether the student violated an assignment’s rules.

Grand Canyon University’s current AI guidance states that the institution does not rely on AI-detection software to determine academic-integrity violations. It emphasizes faculty evaluation, dialogue, applied projects, presentations and process documentation. Students are expected to disclose AI assistance and retain responsibility for their understanding.

This guidance describes institutional practice. It does not establish, through causal evidence, that the practice improves learning.

Difficulty explaining a submission may warrant further instruction or inquiry. It does not alone establish unauthorized AI use. Conversely, a fluent explanation does not establish that all submitted work was independently produced.

Why it matters for Isenberg: A routine explanation, assessed under stated criteria, can provide evidence of understanding without presuming misconduct. Advance notice, consistent grading and appropriate accommodations help distinguish disciplinary knowledge from confidence in speaking.

5. Governance: Colorado’s Continuing Review

The University of Colorado’s September 17 update describes implementation of APS 6012, its responsible-use policy, effective August 24.

CU states that the policy provides common expectations for academic integrity, data protection, transparency and review of AI-generated information. Campuses may develop more specific guidance. The report describes continuing feedback and faculty participation. A systemwide AI group is expected to be established; that step is not reported as complete.

The account illustrates a distinction between a common institutional framework and its application in particular academic settings. It does not report an evaluation of educational outcomes.

Teaching Example: A Report and a Reasoning Check

The following is an original teaching suggestion, not an institutional requirement.

Strategy memo and reasoning check. Follow the assignment’s AI-use permissions and disclose any assistance. Identify two sources you checked and one alternative you rejected. After submission, explain a principal assumption and respond to a changed scenario. The memo assesses sustained analysis; the follow-up assesses understanding and application. Grading criteria are provided in advance.

The follow-up may be written or oral, depending on the objective and required accommodations. A concise account of assistance may be sufficient; a complete private chat history need not be requested when it serves no additional assessment purpose.

Workload also matters. Five minutes for each of 36 students requires three hours, excluding transitions and feedback. A short in-class written exercise may be more practical when spoken explanation is not itself an objective.

The design should specify the capability to be assessed and the evidence needed to assess it. Restricting AI access is a condition, not a measure of learning.