600 People Watched Him Read Words That Weren't His

New on The Ryan Vet Show: Dr. Jared Cooney Horvath on AI in schools → LISTEN or WATCH

Quick housekeeping: last week’s essay went out twice, back-to-back! Totally user error on my part_._ I’m sorry for cluttering your inbox!

On Saturday, I was attending an event where a speaker got up in front of the room presenting slides that were absolutely AI-generated. Honestly, as a former graphic designer, I would prefer AI-modified designs as opposed to ugly slides, but the content was completely machine-fabricated too. Every single tell was there, from the formatting to the AI ticks to the verbiage.

And then, in front of about 600 people, he started reading “his notes” from his iPad. He was reading AI prose. The short, fragmented sentences. It wasn’t his, and you could tell it wasn’t his, because he was stumbling over the words. It sounded a lot like me as I’m trying to make out the broken Cockney accent of the Cabby from The Magician’s Nephew, as I read to my kids at night.

Instead of passing a note, in a completely Millennial move, I texted the guy sitting next to me who happened to be one of the organization’s leaders. My message said, “What is the organization’s AI-use policy?”

“Working on it,” was his response.

The use of generative artificial intelligence in education and in corporations has been a topic of great debate and one where the sides and the arguments change minute-by-minute. But if we are still working on it, we are likely too late.

The ongoing AI-use policy debate

We’ve seen seismic shifts in both directions in education recently. From grade school through graduate-level programs, education institutions are rushing to define what AI use means for them.

Purdue now requires AI competency to graduate. Ohio State wants every undergraduate “bilingual” in it (Purdue University, 2025; The Chronicle of Higher Education, 2026). Singapore’s national university made it compulsory for freshmen and handed ChatGPT to the whole campus. Meanwhile, 29 New York City council members asked for a 2-year ban and a school board in Kansas swung from a total ban to teacher-supervised access just 2 days before school started (KCTV, 2026).

All of that happened in the last 10 weeks.

However, almost none of this dialogue is an argument about honesty and integrity. Every one of those is a fight about whether students may use it, or about who gets their data. New York’s council never argued the technology was bad for learning. They said the city had not protected student information in its contracts with AI companies (K-12 Dive, 2026). Much of the conversation has to do with whether to ban or embrace the technology and how to protect data. Very little emphasis is placed on building the character of the learner.

Columbia Law comes closest, putting the responsibility for everything a student turns in back on them (Columbia Law School, 2026). Oklahoma makes districts tell families each year which AI tools they use.

We’ve seen books getting canceled due to alleged AI use, national magazines and media platforms banning AI entirely, and other organizations embracing generative artificial intelligence to help increase capacity and output.

The Gym and AI

New York City’s guidance says that effective AI integration preserves the intellectual work of learning (New York City Public Schools, 2026). I don’t agree. I fundamentally don’t believe that. I think AI use augments the work of learning, but I don’t think it preserves it. That is a big difference.

I grew up in Chicago going to occasional Cubs games. In the late 1990s I, along with the rest of the nation, had the exciting pleasure of watching Sammy Sosa (Cubs) and Mark McGwire (Cardinals) race for home run records. I had (or still have in my parents’ attic) their baseball cards. Then in the early 2000s came the whole conversation about performance-enhancing drugs and corked bats. Did those modifications make Sosa and McGwire better players? Yes. They absolutely were better players. They hit harder and further. Were their methods okay and acceptable? Most people would say no. And many fans were left disappointed.

They were not using a machine to swing for them. It was their own hand-eye coordination, and they still had to make contact with the ball. The steroids and the corked bats augmented the work. They did not preserve it, because what they put into their bodies changed those bodies.

Dr. Jared Cooney Horvath, an educational neuroscientist who was on my podcast yesterday, has a cleaner way of describing it. Interestingly, Horvath is one of more than 260 organizations and experts who signed an April letter calling for a five-year pause on all student-facing generative AI in PreK-12 schools, so he is not a neutral party in this debate (Fairplay, 2026). He said an expert handing a task to a tool is offloading, because their expertise lets them check what comes back. A novice doing the identical thing is outsourcing, because nothing returns that they can evaluate. Same tool. Two completely different acts.

He used the story of his sister-in-law. She is a gym rat. Her husband is not. They rented e-bikes to tour their city and, as always happens, the batteries ran out. In her case, it didn’t matter, because she was fit. She pedaled the bike right back home. When her husband’s bike ran out of juice, he had to call an Uber. He didn’t have the skill, he was just using a tool to get something done, and when the tool collapsed he had no recourse. Shameless plug to support a fellow author, Jared’s new book, The Digital Delusion, came out today, and it is worth your time.

We are arguing about how much is enough and how much is too much. Is AI a performance-enhancing drug, or is it an e-bike, or is it something else entirely? For the gym rat, an e-bike is offloading, and it is not really a problem. For the guy who never trained, the same bike is outsourcing, and he is stranded the moment the battery dies. It’s the same machine, yet two entirely different outcomes based on how people are relying on it.

Caught in the act

Let’s bring this back to education and the workplace for a moment…I teach a couple of courses a year as an adjunct instructor. My job is to empower students to learn and prepare them for the workforce. Part of learning is the character that gets built along the way, and AI has walked straight into the middle of character development.

The first time I ran into AI outsourcing in my own classroom, nobody had drawn a line yet either.

Six weeks into 2023, in the early days of ChatGPT being available to the public, I was grading a student’s paper. Immediately, I saw some huge AI tells. It was mostly formatting— the line breaks, big headings that aren’t natural in writing, bullet points, and the telltale em dash. The prose itself was perfect, but it was really the formatting and how it looked. I didn’t even read the essay. I glanced at it and could tell.

So I called him into virtual office hours. “I’d like to talk about your project,” I wrote to him in an email.

When he hopped into the Zoom room, I asked him to tell me about his project. He gave a very superficial overview. I wasn’t satisfied. So I asked, ”Did you use AI to help in the writing of your project?”

The student’s image on my screen was locked in with a blank stare; I thought he had frozen for a moment. “What is AI?” he asked, breaking the silence.

“ChatGPT, or something similar.”

He immediately claimed he had never heard of it and denied using it.

A professor’s job is to teach, so I used this as a teaching opportunity. I shared my screen and pulled up ChatGPT. I copied and pasted the assignment prompt into the field, picked the exact company he’d picked for the subject of his project. Real time, he’s watching his essay get rewritten almost verbatim. Same sentence structure. Same arguments. Same breakdown between the points. He still denied using AI.

I reported him to Academic Affairs for an honor code violation.

To be honest, it didn’t concern me at all that he used generative AI. It was an honesty and integrity issue, and it was also something deeper than the lie. Why did he feel he had to hide AI use? It’s like the kid who still has chocolate above their lips but denies eating the fresh-baked chocolate chip cookie off the counter. We hide things when we know they are not right. It’s human nature.

So I stopped arguing and measured it

Over time, I have experimented with the impact of AI in my classroom. This summer I layered hidden instructions into two assignment prompts. They were invisible on the page, so if you read the assignment and completed it the way you were supposed to, you would have turned in exactly the outcome I was looking to grade. But if you selected that prompt, copied it, and pasted it into a chatbot, the model received my hidden instruction alongside the visible one and obediently followed both. One was subtle enough that it changed nothing material, and it was marked optional. One student actually caught it and asked whether it was a real requirement, which is exactly the behavior I want. The other one would have redirected the entire project.

Six of the eleven students who submitted handed back projects responding to an assignment they could never have seen with their naked eye. The only way they would have responded that way was by copying and pasting that prompt.

Graded on substance, they’d done the work. They hit the learning objectives and had a stand-up paper. Several had clearly reworked the output before turning it in. But what they didn’t know was that these silent marks were present in their work because of the prompt injection I had carefully disguised. Not one of them disclosed their AI use.

My students can pass my course 100% using AI. In fact, they can pass most courses from any institution on any subject using LLMs. The concern is that a day after the grades posted, I’d bet several of them couldn’t tell me what their final project was even about. What is even more concerning is that they don’t realize they have hindered their own learning. They’re spending tens of thousands of dollars a year to attend this university, and they are getting nothing of value in return if they’re outsourcing the earning of a degree to artificial intelligence.

The line nobody can draw

What makes AI a unique beast is that its use is almost impossible to police.

The AI detectors don’t work well. I have run writing through them and get back one hundred percent human on things I know were machine-written, and one hundred percent AI on things I know a person wrote. Then there are all the combinations in between that I know just aren’t accurate because I created the documents. I haven’t run a rigorous enough study to give you an exact percentage rate. But anecdotally, it feels wrong more often than right.

What if we could know beyond a shadow of a doubt that AI was used in the creation of something? Would that solve our problems and cure distrust? The newest attempt is from Anthropic. They announced this month that Claude will mark its output with something they are calling a Claude mark. This is a statistical watermark woven into the text as it’s generated. But it applies at launch only to models released on or after August 2, and every model in service today predates that. No detection tool has shipped. And the company says plainly that a mark shows the model processed the text, not that it wrote it.

That is where this problem becomes a double-edged sword. I’m notorious for disagreeing verb tenses. I slide between you and I and we and one. You can even see that in this essay so far. If I write something entirely myself and hand it to a machine to clean up the grammar, is that acceptable? Most would say yes, because that is what would happen if I had a human editor reviewing my work. But now, Claude will be inserting a hidden watermark saying AI touched that work. So where’s the line?

We can draw a line at 0.08 for BAC in driving, at 16 to get a license, or 18 to buy a lottery ticket. Those are measurable and definitive, and you clear them, or you don’t. When it comes to AI, it has influenced so many aspects of everyday life that there is no clear equivalent here.

Where do we draw the line

I am not against using tools to make life easier. Like everyone else, I joined the sourdough fad during the pandemic lockdown. I made bread by hand for a while, and then I started using a stand mixer, and the speed at which I could do things went way up. Does that take some of the friction out? Absolutely. But I honestly found an increase in quality, because the stand mixer didn’t get tired of mixing. I probably got better at bread making too.

So is the stand mixer a shortcut or just a useful tool? It seems like a benign example, but somewhere we have to draw a line. It’s like an X-ray in a dental office or at an orthopedic office. AI can identify the break or fracture or cavity, but it can’t tell you what to do with it. It does not have that human judgment, which is so important.

Right now, as I’m working on my next book, I have a notebook. A physical, college-ruled notebook that I’m writing in. I’m minding the margin, that faint magenta line on the left and the right, that my 4th grade teacher always penalized me for crossing. I’m writing up to that line, and I’m just double spacing so that I can go back and correct my writing in between. It’s been cathartic, honestly, because even in Google Docs, where I do most of my writing, the tools are no longer telling me I confused my verb tenses; it’s trying to rewrite my work. For example, I’ve had to stop using Grammarly because it’s just rewriting my entire sentences. It’s replacing the humanness (no matter how flawed) in me.

Closing thoughts…

What will AI use allow a student to do tomorrow that they couldn’t do today?

I don’t mean write papers faster, or research faster. I mean what practical skill, the kind that stays with them the rest of their life, is it actually going to change? The answer is probably nothing. Logic gets learned on a playground, or over a chessboard, or in a math problem worked out by hand. Conflict management gets learned with the tough roommate or lazy group project team member.

I sent a note to the students who hadn’t disclosed. I told them they’d met the requirements of the course but hadn’t disclosed the use of machine-generated content, and that it’d be helpful to have their own ideas and not those of a machine.

Not a single student responded to that.

Which brings me back to Saturday’s “keynote.” That organization is still working on its policy, and while it works, it let a man get up in front of a room full of its core stakeholders and look like a machine’s puppet. He clearly did not trust his own wisdom, his own insight, or his own words. That could be a symptom of any number of issues: was he being lazy? Was too much output expected from his job that he had to turn to AI? Did he lack personal confidence? Did he really think the machine said it better?

I do not have a clean answer. But I think we are arguing about the wrong thing when we argue about how much is too much. The question underneath it is what is human and what is not. What is the quiet trade? What are we trading with all of these things?