We Built the Detector Before We Built a Way to Tell the Truth

This post is part of a year-long thought leadership series, produced in partnership with Solutions Review’s Insight Jam Mesh Lab, that explores the future of work and learning. Each session brings together educators, workforce development leaders, and industry experts to build a framework for human capability in the intelligence age. Session 6 focused on what AI detection actually measures, and what it is doing to trust.

We Built the Detector Before We Built a Way to Tell the Truth

I opened our sixth session by admitting I have been hiding.

Not the thinking. The help. Both of my organizations expect me to use AI every day, and I do. Then I go quiet about how, because somewhere in me sits the fear that if you knew, you would trust the thinking less.

I said it out loud because I wanted the room to be honest, and I did not think I could ask for that without going first. I didn’t expect that sixty minutes later I would be sitting there questioning whether I wanted to keep hiding at all.

A Score Arrived, He Graded Himself

Early on, one of our panelists mentioned he had run a recent post of his own through a detector. Seventy-two percent. His next words were that he must be the C-minus student.

Nobody had accused him of anything. No policy was violated. A number appeared, and within seconds he had translated it into a grade, assigned it to himself, and accepted the verdict.

Another panelist stopped him. You just put that label on yourself.

I have been thinking about that exchange ever since, because my honest reaction was recognition. I would have done the same thing. Thirty years in education taught me to read a percentage as a statement about my worth, and that reflex doesn’t switch off just because I happen to know better.

What made it worse was what the number described. Work that came out of a project file he had built over eighteen months. His research. His questions, sharpened across hundreds of exchanges. The score read the surface, reported machine, and he believed it about himself before anyone else had the chance to.

A Detector Reports Who Touched It, Not Who Can Do It

I have spent six sessions arguing that we measure the wrong things. That was the first time I watched it happen to a person in real time.

A detection score estimates whether a machine touched the artifact. That is the whole of it. It cannot tell you whether the student understands the material, whether the candidate can do the job, or whether the employee exercises judgment under pressure. It reports provenance. We are reading it as capability.

Researchers call this construct validity. You want to measure something difficult; you measure something convenient instead, and then you treat the convenient thing as if it were the difficult one.

We have done this before. I have done this before. I wanted to know whether a student could think, so I counted whether they could produce five paragraphs on demand. We wanted to know whether a candidate could lead, so we checked a GPA. Every substitution felt reasonable at the time, because the proxy tracked the real thing, roughly, on average, most of the time.

That correlation is gone. And the instrument we grabbed to replace it has a property that should stop us cold. It is most confident about the person who did nothing and least confident about the person who did the thinking. Detection is reliable on raw machine output. It gets unsteady on work someone drafted with a tool and genuinely rewrote.

The artifact was never the capability. It was the evidence we accepted because gathering better evidence cost more than we were willing to pay.

We Did This to Students First and Called It Integrity

One of the professors on the panel said it flatly. Academics who ran student papers through detectors for years are now furious that the same scanner is pointed at their LinkedIn.

I did not have a defense. I ran my own dissertation through plagiarism software and never once questioned it. It never occurred to me to ask what it was actually checking, or what I would have done if it had been wrong about me.

That is the part I keep returning to. We normalized surveillance on people with no power to refuse it. Minors first, with tools less accurate than today’s and consequences that follow a kid into a permanent record. We called it academic integrity, which made it sound like a virtue instead of a policy choice. Three weeks ago the scanner turned toward professors and executives, and the reaction has been very different.

I want to be careful here, because I know who is holding these tools. They were handed a detector, no policy, no training, and more work than any person could verify by conversation at the scale they were asked to teach. Most reached for the only instrument anyone gave them. That deserves grace.

What it does not deserve is repetition. Having no other move yesterday is not permission to keep making the same one now.

And look at where the money is going. Institutions buy detection. Students get sold the workaround, through ads engineered to find them. One detector recently got sharp enough to flag humanized text, which guarantees somebody is already building the next humanizer to sell against it. Money moving in both directions. Nothing learned. Every dollar of it not spent on smaller sections and the time to sit with a student and ask how they got there.

Here is what bothers me most. The safest possible output is now one that sounds like nobody in particular. We have built infrastructure that quietly instructs people to sand off whatever is most identifiably theirs in order to prove they are human. And the skill it develops, producing work that satisfies an inspection, is the one thing machines already do better than any person alive.

Nobody Tells the Truth Where the Truth Is Punished

So the answer is disclosure. Say how you worked.

Except we have not made that safe. One person in the session was turned away from a publication for disclosing AI use, while colleagues published in the same outlet after using it and saying nothing.

Read that sequence again. The system did not punish the use. It punished the honesty. Everyone watching learned the same lesson, and it was not a lesson about integrity.

Disclosure only works if it is safe. If admitting how you worked invites suspicion or costs you the byline, people stop admitting it. Not the dishonest people. Everyone. Concealment becomes the rational choice, and once it is rational, it is universal.

I have spent this entire series arguing that conditions come first. We built the detector before we built any environment where telling the truth was the smarter play.

And I have to be honest about my own position in this. Disclosure is cheap for me. I have standing, a platform, and nobody grading my work. A student facing an academic integrity panel has none of that. Neither does a candidate whose application gets scanned before a human ever reads it. When I say be transparent, I am asking people to take a risk I am not taking.

What came out of the session that I can actually use is smaller and better than a principle. A real disclosure answers three things. What the AI did. What the person did. Who verified it.

I developed the argument. I selected the evidence. I used AI to challenge my reasoning and tighten the structure.

That discloses a process, not a percentage. It tells you where the judgment lived, which is the only thing you wanted to know when you asked whether a person wrote it. And it is the strongest claim available now. Anyone can produce a clean artifact. Almost nobody can walk you through why they made this choice, what they rejected, and how the answer changes if you alter one condition.

That conversation cannot be automated. It is also the thing we stopped making time for, because sitting with one person does not scale, and we built systems that only counted what scaled.

The Hiding Was the Tell

I went into that session convinced disclosure would cost me credibility.

I left thinking it might be the credibility.

What I have been protecting is not my thinking. My thinking is fine. It is mine, and I can defend every choice in it. What I have been protecting is the appearance of effortlessness. The impression that this arrived fully formed, that nobody helped, that I am not the kind of person who needs a mirror to think against.

I do not know yet how to make disclosure safe for the people who have the most to lose by it. I am not sure anyone in that room did either.

But you are already using it. So is nearly everyone around you. So what is it costing you to keep pretending otherwise, and who is learning to hide because they watched you first?

Dr. Michelle Ament leads the Human Intelligence Movement, a grassroots nonprofit dedicated to ensuring humans have the skills to thrive in an AI world. She is also Chief Academic Officer of ProSolve, The Human Skills Company. As a K-12 educator and district administrator, she bridges education and workforce, helping organizations make human capabilities visible, measurable, and actionable.

Connect with Michelle on LinkedIn or at humanintelligencemovement.org.