Vermont Teachers Confront Classroom AI’s Learning Tradeoffs
Vermont teachers entered the new school year with a conflict now spreading through Google News: students can use artificial intelligence to learn faster or avoid learning altogether. The distinction sounds simple. Applying it across different ages, subjects, assignments, and software is not.
A September 1 classroom AI report from WCAX captured that disagreement through educators working in and around Vermont. Some teach students to inspect AI-generated work for errors and bias. Others want little or no classroom use, especially before high school.
This is not a familiar argument between technology enthusiasts and traditional teachers. Both sides recognize that students will encounter AI outside school. Their disagreement concerns when exposure becomes education, when assistance becomes substitution, and who must decide.
Vermont has already published statewide guidance, but it deliberately stops short of settling every classroom decision. Districts, school leaders, and teachers still must translate broad principles into practical rules. That leaves them managing the tension between educational judgment and consistent policy.
The immediate question is not whether AI belongs somewhere in education. It is whether schools can define uses that preserve the thinking an assignment was designed to develop.
Vermont’s Classroom Debate Has Moved Beyond Bans
The new development is a shift from asking whether schools should block AI to deciding which uses support a specific learning goal.
WCAX reported that Claudine Bedell, an education professor at St. Michael’s College, asks students to critique material produced by an AI tool. Students quickly encounter errors and possible bias, she said. The exercise makes evaluation, rather than generation, the point of the assignment.
That distinction matters because generative AI creates text, images, audio, or code from patterns in training data. It can produce fluent answers without establishing that those answers are accurate. A classroom activity can expose that limitation when students possess enough subject knowledge to challenge the output.
Bedell’s approach treats AI literacy as critical literacy. Students must identify unsupported claims, missing perspectives, and language that appears convincing without being reliable. The model becomes an object of study rather than an invisible author.
Christina Asquith, a co-founder of the Burlington-based nonprofit Hack Club, described a more restrictive position. Hack Club teaches teenagers computer science, and participating teens use AI to write code. However, Asquith told WCAX that students sometimes submit projects they do not understand.
Hack Club rejects those projects. According to Asquith, three-quarters of those students resubmit within one or two weeks with a better understanding of their code. The organization permits the tool while refusing to accept output as evidence of learning.
That example captures the central problem. A working program shows that software produced an operational result. It does not show that a student can explain the program, debug it, or recreate its logic.
Asquith said Hack Club participants were producing code about 100 times faster than in 2023. That figure is the organization’s account, not an independent measure of learning. Faster production can coexist with weaker comprehension.
She therefore draws a much narrower classroom boundary. Asquith told WCAX that AI has no role in kindergarten through eighth grade and should have a very limited high school role.
The disagreement is real, but it has common ground. Bedell and Asquith both reject passive acceptance of machine output. Each wants students to understand the work in front of them.
One approach develops that understanding through structured criticism. The other limits access until students have stronger foundations. Google News surfaces the disagreement as a timely local story, but Vermont schools must turn it into durable operating rules.
Google News Highlights a Statewide Guidance Gap
Vermont has a statewide framework, yet the consequential decisions still sit with districts, school leaders, and individual teachers.
The Vermont Agency of Education issued a 50-page AI education guide on January 23, 2026. It addresses administrative work, personalized learning, AI literacy, responsible use, tool review, and developmentally appropriate access.
The document provides direction without declaring one universal classroom policy. It tells educators to consult approved software lists, district technology teams, school leadership, and applicable privacy requirements.
That flexibility respects local educational judgment. It can also produce different answers for students taking similar classes in neighboring districts.
The state guide separates expectations by grade range. Elementary educators should focus primarily on using AI to streamline their own work. Middle school is described as a transition point where students need structured guidance. High school students can take on more demanding AI-supported tasks after developing AI literacy.
Those distinctions make developmental readiness central. A tool suitable for a high school coding exercise may be unsuitable for an elementary writing lesson. Even within one grade, readiness varies by student, subject, and task.
The framework also distinguishes access from supervision. A district might approve an education-focused chatbot, but approval does not define how a teacher should use it. The instructional question remains whether the tool strengthens the targeted skill.
Consider a writing assignment. AI can suggest an outline, provide feedback, or produce a complete response. Those actions occur through the same interface, yet they have different educational consequences.
A teacher might permit brainstorming but prohibit generated prose. Another might ask students to compare an AI draft with their own. A third might prohibit the tool because the assignment measures unaided reasoning.
These differences become confusing when schools communicate only whether AI is “allowed.” Students need task-level expectations: what they can do, what they must disclose, and which parts of the work must remain their own.
National reporting shows that this ambiguity extends beyond Vermont. An academic honesty analysis from the Associated Press described students encountering different AI rules across classrooms. Grammar correction, outlining, rewriting, and idea generation frequently sit in uncertain territory.
A useful policy must therefore describe actions rather than brands. Blocking one chatbot does not answer whether students can use AI features embedded in search engines, writing software, document editors, or learning platforms.
The same principle applies to teacher use. A school can allow lesson-planning assistance without permitting automated grading or feedback. The policy must identify which responsibilities require direct professional judgment.
Vermont’s guidance gives schools a vocabulary for those decisions. It does not eliminate the hard work of making them.
The Real Divide Is Learning Versus Substitution
The primary conflict is not teachers versus technology. It is AI-assisted learning versus AI-assisted substitution.
Substitution occurs when a tool performs the cognitive work an assignment intends to measure. If students must construct an argument, generating the entire argument substitutes for that process. If they must learn syntax, accepting generated code can conceal whether they understand the language.
Assistance supports the process without displacing its purpose. A student might ask a chatbot to create practice questions, explain a mistake, or challenge a conclusion. The student still must evaluate the response and demonstrate independent understanding.
The boundary changes with the assignment. A calculator can support advanced mathematics while undermining an exercise designed to build arithmetic fluency. Vermont’s 2024 state AI policy report used that comparison when discussing generative AI.
The analogy is useful but incomplete. A calculator performs defined operations with predictable outputs. A language model generates probabilistic responses that can contain factual errors, bias, or invented details.
That difference increases the importance of verification. Students need enough knowledge to recognize when a polished answer is wrong. Without that foundation, “checking the AI” can become little more than accepting its confidence.
Bedell’s classroom exercise addresses this weakness directly. Students first inspect the generated material and then identify its flaws. The learning value comes from the critique, not from obtaining text faster.
Hack Club uses a different mechanism. It tests understanding after students build with AI. Rejecting unexplained projects forces students back into the code, where they must develop an account of what the system produced.
Both cases change assessment. Teachers cannot treat a finished document or functioning program as sufficient evidence. They need to examine the process behind it.
That can include drafts, source notes, oral explanations, version histories, in-class work, or short reflections describing how a tool was used. These methods are more informative than an AI-detection score.
Detection software attempts to estimate whether a model generated a passage. Such estimates do not reveal what a student learned, and false positives can damage trust. Process evidence gives teachers a stronger basis for evaluating authorship and understanding.
Clear disclosure also matters. Students should know when they must identify an AI-assisted step and how detailed that disclosure should be. Teachers who use AI-generated materials should follow comparable expectations.
That reciprocity is emerging as an important issue in Vermont. A Vermont educator survey published by Seven Days reported that some students felt less motivated when teachers assigned AI-generated material. Students questioned why their own work faced tighter rules than their teachers’ work.
The concern is not merely symbolic. An inaccurate worksheet can spread errors across an entire class. A generic lesson can also fail to reflect the students, curriculum, or discussion that a teacher knows firsthand.
Teacher oversight must therefore mean more than pressing a button and reading the output once. Educators remain responsible for accuracy, accessibility, instructional fit, and the treatment of student information.
A transparent classroom can make that responsibility visible. Teachers can explain when AI helped prepare a resource, which parts they changed, and why the final material serves the lesson. Students can document their own permitted uses through the same principle.
This model does not give students and teachers identical responsibilities. Teachers design instruction and evaluate learning. It does, however, establish a shared standard: neither side should present unexamined machine output as original, verified work.
Privacy and Age Limits Complicate Classroom AI
Even a sound learning activity can become unacceptable when an AI tool collects sensitive data or exceeds a student’s developmental readiness.
Generative AI systems often retain prompts, generate usage records, or connect interactions with user accounts. A student may reveal personal details, academic struggles, disability information, or unpublished work during an ordinary conversation.
That makes tool approval a data-governance decision. Schools must understand what information a vendor collects, how long it retains that information, and whether it uses inputs to improve its models.
The Vermont guidance directs educators toward district-approved tools and technology teams. It references federal privacy requirements, including the Family Educational Rights and Privacy Act and the Children’s Online Privacy Protection Act.
Teachers cannot resolve every contractual or technical question independently. Districts need a review process that covers data handling, security, accessibility, account controls, and age restrictions before classroom use begins.
Product design introduces another complication. A general chatbot may encourage open-ended conversation, while an education-specific application can restrict the model to a defined subject or activity. A narrower system reduces some risks but does not remove them.
The state guide notes examples such as literacy screening and curriculum-focused tutoring. It also warns that school leaders must examine the actual application rather than assuming an educational label guarantees safety.
Supervision remains limited. Vermont’s guide discusses gaps in administrators’ ability to review student conversations inside some widely used platforms. A school cannot promise meaningful oversight if its staff cannot see how students interact with a tool.
Age makes those limitations more consequential. Younger students have less experience evaluating persuasion, factual uncertainty, and simulated emotional language. They may attribute understanding or authority to a system that generates conversational responses without human awareness.
This helps explain Asquith’s call for no classroom AI in kindergarten through eighth grade. Her position emphasizes protection of foundational learning and development before students encounter tools that can complete work for them.
Vermont’s state guidance takes a more graduated approach. It focuses elementary use on educator tasks, introduces structured AI literacy during middle school, and allows broader high school use after students demonstrate readiness.
Neither position has been settled by long-term classroom evidence. Generative AI tools became widely available only in late 2022, leaving schools with limited longitudinal research on learning, dependence, and development.
That uncertainty should narrow claims on both sides. Schools cannot assume that early exposure automatically creates literacy. They also cannot assume that avoiding classroom tools prevents students from using them elsewhere.
Equity adds another pressure. Students with paid accounts, personal devices, or stronger support at home can gain access regardless of school rules. A simple ban can therefore push AI use outside supervised environments without making it disappear.
Conversely, mandatory use can burden families concerned about privacy, bias, screen exposure, or accessibility. It can also disadvantage students who learn better through human instruction and direct practice.
A defensible policy needs alternatives. If an assignment permits or requires AI, students should have a way to meet the same learning objective without surrendering personal data to an unapproved service.
Schools should also avoid framing every accommodation as permission for unrestricted generation. Accessibility support can be tightly defined around a student’s documented need while preserving the assignment’s purpose.
The difficult truth is that “safe AI” is not a permanent product category. Vendors change features, policies, models, and account controls. District approval must be revisited when those conditions change.
Teachers Are Now the Pressure Point
Vermont’s flexible approach places professional judgment at the center, but it also asks teachers to manage responsibilities that no individual classroom can solve alone.
Teachers must decide whether a use supports learning, explain that decision to students, apply academic-integrity rules, review output for errors, and protect student information. They must do so while AI features spread through familiar software.
A complete prohibition appears easier to administer. Yet students can still access tools through personal accounts, search interfaces, writing assistants, and phones. Enforcement then becomes an expanding surveillance task.
Unrestricted access avoids that enforcement burden but creates another problem. Students can outsource the very reasoning schools are meant to develop.
The workable middle requires more design, not less. Assignments need explicit AI boundaries, meaningful process evidence, and assessment methods that reveal understanding. That increases preparation demands before it reduces any workload.
Professional development therefore matters. Teachers need time to test tools in subjects they understand deeply. That allows them to see typical errors, weak citations, biased assumptions, and patterns of generic language.
Training must also cover when not to use AI. A session focused only on prompting risks treating adoption as the predetermined goal. Educators need authority to reject a tool when it does not serve their students.
Local reporting shows Vermont districts at different stages. Champlain Valley School District has developed principles focused on equitable learning, educator support, and ethical use. Harwood educators have worked toward a shared philosophy intended to reduce inconsistent rules.
Those efforts show why a statewide document cannot be the final layer. Districts need assignment guidance, approved-tool lists, disclosure practices, family communication, and procedures for reviewing disputed work.
School leaders also need to define accountability. A teacher should know who approves a new tool and who responds if student data is exposed. Students should know how to challenge an AI-related accusation or request clarification before submitting work.
Families need plain explanations of what a tool does. Product names and broad claims about personalized learning do not answer whether a student will converse with a model, upload work, or receive automated feedback.
The debate has consequences for vendors, too. Education companies face pressure to provide administrative visibility, age-appropriate controls, accessibility, and clear data practices. A chatbot designed for adults cannot become classroom-ready through branding alone.
Google, Microsoft, OpenAI, and education-focused vendors are all competing for a place in school workflows. Their products differ, but the institutional question is the same: can a district control use without transferring educational authority to the software?
Schools should judge those systems according to local learning goals. A feature that saves planning time is not automatically suitable for students. A tutoring interface is not automatically reliable enough for independent use.
Teachers remain responsible for the relationship that software cannot supply. They notice frustration, prior knowledge, social context, and changes in a student’s work over time. AI can process inputs, but it does not assume that professional duty.
That is why the current pressure cannot be resolved through procurement alone. Vermont’s experiment depends on whether institutions give teachers enough policy, training, and time to exercise the judgment they are being asked to provide.
What Vermont Schools Should Watch Next
The next phase will be measured through district policy, classroom evidence, and student trust rather than the number of AI tools adopted.
The first signal is whether districts convert broad guidance into assignment-level rules. Students should see clear categories for prohibited, limited, disclosed, and required use. If those categories appear across subjects, Vermont will be moving toward consistency without imposing one rule on every lesson.
The second signal is whether assessment changes. Schools that continue grading only finished products will struggle to distinguish learning from substitution. More oral explanations, documented drafts, and classroom demonstrations would show that policy is influencing instructional practice.
The third signal is whether teachers and students report greater clarity. A policy can look complete while leaving daily questions unanswered. Districts should ask whether students understand permitted uses and whether teachers know how to respond consistently.
These signals matter more than adoption figures. A high usage rate does not establish educational benefit. A low rate does not prove that students are learning without AI outside school.
Vermont lawmakers also remain part of the picture. State Sen. Seth Bongartz told Seven Days that he wanted a model policy while expressing reluctance about a blanket prohibition. Future legislative action will reveal whether the state keeps guidance voluntary or establishes firmer requirements.
Vendor changes deserve continuous review. A district may approve a tool under one privacy policy and encounter new features later. Conversation sharing, automated agents, memory functions, or social features can change the risk calculation.
Schools should document review dates and triggers. Approval should expire or return for evaluation when a vendor materially changes student-facing features, data practices, or administrative controls.
Educators should also collect concrete classroom evidence. Did students retain concepts after AI-supported practice? Could they explain generated code? Did feedback improve revision, or did it replace conversation with a teacher?
Those questions are harder than counting logins, but they address the purpose of education. They also protect schools from mistaking activity for learning.
Students belong in the evaluation process. They can describe when rules feel inconsistent, when AI makes an assignment less meaningful, and when structured use helps them practice. Their feedback should not decide every policy, but excluding it would hide an essential part of classroom reality.
Vermont’s public debate is valuable because it resists a simple adoption story. The educators highlighted through Google News are not deciding whether AI is impressive. They are deciding when it is educational.
That standard leaves room for different classroom answers. A coding club may permit AI-generated code and demand a technical defense. A writing class may prohibit generated prose while permitting source comparison. An elementary classroom may keep student-facing AI entirely outside the lesson.
Consistency should concern the decision process, not identical outcomes. Every allowed use should have a learning purpose, an age rationale, a privacy review, and a method for checking understanding.
For students managing research, notes, and source material, a private student workspace can support organization without replacing the reasoning an assignment requires. Whatever tool they use, the student should remain able to explain the final work.
The most useful question for teachers is therefore specific: what thinking must students practice here, and which parts must they perform themselves?
If Vermont schools can answer that before choosing a tool, their policies will remain useful as products change. If they cannot, the do’s and don’ts will keep shifting between classrooms while students learn to navigate the gaps.
