Episode Transcript
This is a computer-generated transcript. While our team has reviewed it, there may be errors.
Morgan Sung: What are your thoughts on, on generative AI?
Jeremy Na: Generative AI is a fancy autocomplete, in my opinion. Trusting it is no different than trusting a magic 8-ball.
Morgan Sung: Jeremy Na is a Bay Area High School English teacher. He’s a great teacher, and I know this because he’s my older cousin, and he spent a lot of his teenage years at my parents’ dining table as my math tutor. Let’s just say I wasn’t the most cooperative student. One of the first hurdles of his teaching career was probably getting me to understand basic algebra. The second, getting me to actually sit down and do my math homework. And now, like many teachers, he faces another challenge: AI in the classroom.
Jeremy Na: When my students tell me a lot about how they trust AI answers and stuff like that, I reference the SpongeBob episode where, you know, everyone in Bikini Bottom is trusting a magic conch shell to give them answers and decide their life.
SpongeBob SquarePants: Oh magic conch shell, what do we need to do to get out of the kelp forest?
SpongeBob SquarePants: Nothing.
SpongeBob SquarePants: The shell has spoken!
Jeremy Na: So I think it’s a big scam. Or rather, I know it’s big scam because, you know, it just sucks up money and is burning the environment for nothing. That’s my personal view of generative AI.
Morgan Sung: So yeah, my cousin Jeremy, not a fan of AI. In recent years, the use of AI tools has been a major point of contention between students and teachers. Scroll through any social media platform and you’ll see students give tips for getting away with submitting AI essays.
TikTok: I use ChatGPT, but this is the way to use ChatGPT and not get flagged for any plagiarism or anything like that.
Morgan Sung: Teachers fretting about their students’ dependence on AI.
TikTok: This is a nightmare that is destroying learning.
Morgan Sung: And then there are the students who don’t use AI but get dragged in anyway.
TikTok: I was falsely accused of using AI on my final paper last term, it was flagged as 60% AI.
Morgan Sung: In his nine years of teaching, my cousin Jeremy has seen his fair share of cheating attempts. His students have used ChatGBT since it launched about three years ago. But he quickly realized that his students were using it for more than just an academic shortcut.
Jeremy Na: They were asking me stuff like, Mr. Na, uh, how can I use AI to help me with this assignment? Or Mr. Na, you know, I was using AI the other day and it really helped me with X, Y, Z problem. At first, I thought these students were just kinda trolling cause like I thought to myself, there’s no way anyone would trust AI. This is the machines that like tell people to staple cheese to pizza to make it stick better.
Morgan Sung: Google AI actually recommended adding glue to pizza, but his point still stands.
Jeremy Na: But no, I realized soon enough that my students were like serious about trusting AI. And that’s when I realized, oh, my students, you know, they’ve fallen for the propaganda.
Morgan Sung: There’s still no consensus on the role of AI in education. Some teachers embrace it as a feature of their lesson plans. Others ban it from their classrooms entirely and blame AI for their students’ atrophied critical thinking skills. And at the center of this debate is a question that’s haunted educators throughout human history. What do we do about cheating? Now that everyone can carry little AI cheating machines in their pockets, AKA their phones, that debate has kicked into overdrive.
This is Close All Tabs. I’m Morgan Sung, tech journalist, and your chronically online friend, here to open as many browser tabs as it takes to help you understand how the digital world affects our real lives. Let’s get into it.
Today’s episode is a little different, because instead of following one thread, this internet rabbit hole will take us down a few different paths. We’re going to look at three different approaches to curbing AI cheating in school.
Back when the internet was brand new, students suddenly had access to a virtually endless trove of information to copy and paste from. Academics needed a way to tell what was original and what was copied. So plagiarism detection was born. An entire industry dedicated to identifying and flagging stolen work. But with generative AI blowing up, checking for plagiarism isn’t enough.
First up, a new tab. What’s up with AI detection?
Morgan Sung: To get into this, I called up Max Spero, co-founder and self-described chief slop janitor of Pangram Labs. Pangram is an AI detection tool that flags texts generated by all the major models like ChatGBT and Claude and Grok, and also detects AI-generated edits from tools like Grammarly.
Max Spero: It could be an essay, it could be a review, it could a social media post. And then so we help people tell you like, yeah, what’s AI and what’s human.
Morgan Sung: Cleaning up the slop.
Max Spero: Cleaning up the slop as they say.
Morgan Sung: Slop is the colloquial term for that low-quality, AI-generated content flooding the internet. Pangram didn’t start out as an educational tool. Max said that the company’s first customers were actually websites trying to detect and delete fake reviews. Pangram offered a free trial version of the tool on its website.
Max Spero: But then we were looking at the signups and we kept on seeing signups from like EDU addresses, people who are putting in what’s clearly student papers and people are doing this like at huge volumes. They’re just pasting in like dozens of papers a day. And then, so I think we were realizing like, hey, there’s a much bigger market here on this consumer side as well, where we can help teachers out because this is clearly a big need.
Morgan Sung: Pangram’s detection works by showing its model two writing samples. The first, written by a human, and the second, similar but AI-generated. They do this over and over and again. The model learns to recognize the way humans write, and distinguish it from AI trying to replicate it. That’s how certain sentence structures or words like “delve” and “rich cultural tapestry”, or even, my beloved em-dash, became associated with AI writing. Humans use all of those features in writing, but AI models overuse them. They’re predictable. There’s a misconception that students can bypass detection tools by paraphrasing AI-generated writing. That may have been the case a few years ago, but detection tools are getting better.
Max Spero: We found that you need to rewrite at least 30% to 40% of some text before it’ll come back as human written. So you really have to like rewrite a large portion of the text before you’re able to erase these signs of AI writing. At that point, you might as well just write your essay yourself. But in the end, like all of these tools, at Pangram and other AI detectors, we can still train on these, the outputs of these tools, so we can detect, not only does this look AI generated, but it also looks like it was run through a humanizer.
Morgan Sung: That’s a program that rewrites AI-generated text to sound more, well, human. And of course, humanizers use AI to do that.
Max Spero: And that’s even more clearly a sign of acting in bad faith. Like this is a clear indication that no, you didn’t just like misunderstand that you weren’t allowed to use ChatGPT, but like you actually were actively cheating here.
Morgan Sung: It’s not impossible to bypass AI detection, but Max said that many AI detectors, including Pangram, can still flag humanized text as AI. He added that since these students are often trying to take the path of least resistance, the threat of detection could be enough to deter cheating.
Max Spero: Adding a little bit of friction here goes a long way to helping put down these guardrails and say like, okay, fine, you know, if it’s not really easy for me to just use AI to generate my assignment, I’ll just do it myself. It’s not that bad.
Morgan Sung: Some teachers have tried to AI-proof their assignments by adding a random prompt in white text, like “mention bananas in every paragraph.” The human eye can’t see the white text on a white digital background. If a student copies and pastes it into ChatGPT, they’ll get a generated essay that has nothing to do with the actual assignment, but a lot to do with bananas. It’s a dead giveaway that they used AI to complete the paper.
There’s also some confusion over what detection scores mean. A score of 60% doesn’t mean that 60% of the essay is AI-generated, it’s a confidence score. It means that the detection tool is 60% certain that the text it analyzed is AI generated, which is pretty far from certain. That’s why students who wrote completely original papers have been accused of cheating.
These AI-proofing hacks and the knee-jerk reaction to accuse students of using AI without hearing them out says a lot about the relationship between students and teachers right now.
Max Spero: I think there’s a big problem in trust right now in education, especially because, um, the nature of it is so adversarial. I think like we really need to take a step back and realize that we have a shared goal. The goal is to get the student to learn. And I think a lot of this starts with like how we understand assignments. Like, teachers need to be very clear to students like, hey, these are the guardrails.
Like you can use AI, for example, to do brainstorming or an outline, but don’t use AI to fully produce your assignment. Um, and similarly, once the student has like a very clear understanding of like, these are the guidelines, this is what I can and can’t do, um, then it’s easier for them to work together.
Morgan Sung: What’s your stance on teachers using Pangram to grade? Are AI detection tools like the end all be all? Can teachers rely on it completely? How do you best see Pangram being used in the grading process?
Max Spero: It’s like really only the start. It is best used as a smoke detector to tell you, hey, something’s wrong. I should look into what’s going on here. I don’t think it’s appropriate to say like, hey, this AI detector flagged your work as AI, so I’m gonna give you a zero and then just like move on. I think that’s, that’s lazy and that doesn’t really turn the opportunity into a teaching moment.
Morgan Sung: Detection tools are just one part of a holistic grading process. Max added that some teachers integrate AI detection into their classroom platforms so that students can check themselves before they turn in their assignments. But other teachers are taking a wildly different approach. Instead of leaning into detection, they’re going back to the olden days. We’ll dive into the return of handwriting after this break.
Morgan Sung: We’re back. Let’s take a look at the next strategy teachers have employed to hold the line against AI. If you were in college before the pandemic started, you might remember this vintage classic. Okay, new tab. Back to blue books.
Morgan Sung: My colleague, Marlena Jackson-Retondo, is joining us for this part of our deep dive. Marlena has been reporting on AI and K through 12 education for MindShift, another KQED podcast about the future of learning.
Marlena Jackson-Retondo: So I guess now we can say that blue books are an old-fashioned, in quotes, tool that, um, a lot of professors and high school teachers use to test students. And they’re used for a lot midterms and finals, and are an alternative to now what we know as, you know, digital testing tools.
Morgan Sung: I’ve used blue books. You’ve used blue books. I feel like it wasn’t that long ago that we were using blue books.
Marlena Jackson-Retondo: We’re not that old.
Morgan Sung: Blue books are exactly what they sound like. They’re little booklets with a baby blue cover and lined paper. They’ve been around since the 1920s and they’ve been a staple of written exams in high school and college. I still remember the absolute horror I felt my junior year of college when my pen exploded just as I finished my constitutional law final and I had to turn in a blue book covered in purple ink.
But blue books were phased out when the pandemic shut down schools in 2020. Everything involved in learning, from lessons to homework to exams, took place online. And then, late last year, Marlena saw a viral post from Jason Coupe, a professor in Atlanta.
Marlena Jackson-Retondo: And he was talking about using blue books, bringing them back to the classroom for his first midterm of that school year.
Morgan Sung: Can you talk about his reasoning behind moving to blue books? How did other teachers react to that?
Marlena Jackson-Retondo: He and a couple other professors in his department got together and discussed ways to mitigate cheating, use of AI, and also to reengage their students. When we spoke last year, there was a lot of discussion about really getting students to think on their feet, think critically, respond to questions in ways that they might in the real world. And he wasn’t seeing a lot of that in digital exams.
So this wasn’t just about cheating. It wasn’t just about AI. There was also that sense of reconnecting the students to the coursework. And it was a learning curve for a lot of students, um, as I’ve heard from multiple professors, but it went well, and a lot of folks will continue to, to use these blue books.
Morgan Sung: By 2024, you know, this is a generation of students that have spent at least four years, you know, learning digitally in some capacity. How did they respond to having to go back to handwriting?
Marlena Jackson-Retondo: Yeah, so a lot of Coupe’s students didn’t know what a blue book was. They had never taken a handwritten exam. They didn’t what to do with the blue book, where to write their name. So he had to teach them. And I remember him saying that it reminded him of his time in teaching in elementary school, um, having to really break down certain processes for students. But, you know, they learned, they took the exam by hand, and he and other professors noticed a difference right away.
Aside from deterring cheating, how else does prioritizing handwritten notes impact learning?
Marlena Jackson-Retondo: The research on handwriting is actually really interesting. I spoke with Sophia Vinci-Booher out of Vanderbilt University. She talked to me about handwriting in a really interesting way in that it creates these neural connections to what you’re learning. And this is what she called the visual motor learning system. So it’s combining these two systems, the motor system of handwriting and the visuals of learning something that might be written on the board or a PowerPoint, and it’s combining them and it’s reinforcing what you’re learning when you’re writing by hand.
It’s been shown that the mode of taking notes, when it correlates with the mode of having to recall that information, like taking an exam. When those modes are synced, so let’s say there’s a student who takes notes by hand, and then they have to go and take an exam on a blue book, the recall is better that way.
So essentially, students are “learning more”, that’s in big quotes, because obviously there are other factors involved in learning. But there’s better learning happening when the mode of note taking and recalling those notes is the same. So, you know, I think there’s a lot more research to look at and to be done, but there are certain benefits to writing by hand.
Morgan Sung: And blue books are hot again. Earlier this year, The Wall Street Journal reported that blue book sales were up more than 30% at Texas A&M University and almost 50% at the University of Florida. At UC Berkeley, blue book sales shot up 80% over the past two years.
Of course, this return to our roots has its limits. Blue books are great for exams, but they’re not a realistic option for longer assignments like research papers. And neither approach we’ve covered so far, AI detection and blue books, addresses the core reasons students cheat.
Let’s look at one final strategy in a new tab. The testmaxxing problem.
Let’s go back to my cousin, Jeremy, who’s a high school English teacher in the Bay Area. We both grew up in New York City and went to very academically rigorous public high schools.
Jeremy Na: The standards of the time, this was in, you know, the early 2000s, was very much, let’s call it testmaxxing.
Morgan Sung: In gaming, there’s this practice called minmaxxing — maximizing your stats with a minimal amount of effort. On the internet, adding maxing to the end of any word is kind of like a joke about optimizing. So by test maxing, Jeremy is referring to the way that students are encouraged to shape their whole approach to school around succeeding on tests instead of actually learning. He believes that’s the reason students cheat.
Jeremy Na: So, when I became a teacher, I set out to make sure that like no one else has that same experience as I did that, you know, students don’t have this miserable high school experience where they’re treated like cattle, basically. Like you gotta you gotta get those numbers up. Your entire worth is decided by these numbers. Our society has placed a lot of importance on the end result of education rather than like education itself. What I mean by that is it’s more important what grade you get on the test than what you learned through the process before taking the test.
Morgan Sung: To counter the pressure of testmaxxing, Jeremy does things differently in his classroom. He never assigns homework. All the work, including reading and writing assignments, is done in the classroom. He does assign long-term projects, like essays, but he has his own way of AI-proofing them.
Jeremy Na: This methodology, this pedagogy that I’ve developed is developed from shifting priorities away from the end result over into focusing on the process, right? So the fact that it’s inconvenient to use AI was just kind of like a happy little bonus.
Morgan Sung: He really does make it inconvenient to cheat in his classes.
Jeremy Na: Now, I don’t see a lot of AI usage. It’s always like one or two out of my 150 students. And the reason I don’t see a lof of AI usage is because when we’re doing it, a long-term assignment like an essay, I make sure to break down that assignment into as many granular pieces as possible.
So for example, with my freshmen, ninth graders, when we were doing an essay I’ll often walk them through, not only like how to construct each paragraph, but how to construct like the parts of each paragraph. Like sometimes we’ll go sentence by sentence. In that scenario it would be kind of absurd to use AI, right? Because like Mr. Na is telling you, okay, write one sentence about your opinion on this part of the book, right. Why, why would you go ask Grok or ChatGPT opinion about the book when you can just write your own opinion in one sentence.
Morgan Sung: What is your classroom AI policy, even though, you know, you personally despise AI? Like do you use a detection tool for your students?
Jeremy Na: I don’t use detection tools at all. I would not trust AI to tell me what the weather is. Why would I trust it to read student reports and, you know, analyze them? That’s absurd.
Morgan Sung: Despite his evident disdain for generative AI, my cousin doesn’t have the same zero tolerance policy that a lot of other teachers have.
Jeremy Na: I don’t explicitly call out their paper for being AI. But what I will do is, you know, I’ll sit with them and be like, “You know, this, this paper’s got a lot of problems. This assignment that you wrote has lots of problems, like it’s, it’s overflowing with problems. I got to sit you down here and we got to talk for like 15 minutes. Sentence by sentence and deconstruct this essay to fix it.”
I feel like in that scenario, students do learn like what’s wrong with using AI. Conceivably, they could learn to cheat better this way, but in my view, what they’re actually learning is that the assignment I’m asking them to do is not something outside of their capabilities. So yes, they are learning what the flaws of AI are, but they’re also learning that they are more capable than they might think.
Telling a teenager not to do something never works. Yeah. If you want a teenager to do something, right? If you want that horse to drink that water, you gotta make the water appealing. You gotta explain why it’s in their best interests not to so, right? They gotta come to that conclusion on their own. The downside is that this process takes a lot of time. So yes, I don’t get through many bucks throughout the school year, but I feel like the quality of the work and the learning that students receive is a lot better for that sacrifice. And I’m willing, I’m, I am willing to make that sacrifice
Morgan Sung: I feel like on Reddit, I’m always seeing teachers and professors talk about the ways that they AI-proof their assignments, like adding white text to trip up the prompts or splitting up instructions into multiple documents so it’s really inconvenient to give it to ChatGPT. Have you and your colleagues tried any of these? What do you think about these, like, AI-proofing tricks?
Jeremy Na: I think getting into a arms race with AI is unhelpful. So coming up with different ways to detect AI, and then AI comes up with ways of getting around that, you know, this arms race that you’re describing, I don’t see a point to engaging with it. If you want students to stop using AI, you got to address the core issue, right, rather than just constantly dealing with these symptoms like some absurd game of whack-a-mole.
Morgan Sung: As you might be able to tell, Jeremy feels pretty strongly about this. He believes that putting so much value on the final product, whether a final paper or a final test score, deprioritizes actual learning, overwhelms students, and incentivizes cheating. Still, he has to teach within the system that values end results. His students still have to take California’s standardized tests. But he said that prepping them for the annual state assessments doesn’t take that much time. For the rest of the school year, he can focus on breaking down assignments and, through that, developing their critical thinking skills. It’s not that grades don’t matter, but by focusing on actual learning, students feel less pressure to cheat.
Jeremy Na: Everyone is telling them to participate in this rat race, that they have to. That they have no choice but to participate in this rat race to climb to the top, you know. If they are learning, right? I tell students not to deride all of their self-worth from the numbers they get at school.
Morgan Sung: It seems like the larger question we’re trying to answer throughout the episode is how do you maintain that sense of trust between students and teachers when generative AI tools makes cheating so easy?
Jeremy Na: I don’t think, say that like trust between student and teacher has eroded. I think it’s more that teachers are becoming more suspicious overall because the tools that they’ve developed in the past no longer work. Students using AI is not a sign of, I don’t know, like moral degeneracy, and the youth of today are worse than they have been in the past, no. But you know, students are cheating nowadays, just like students would have cheated back when I was in high school or when my parents were in high score when my grandparents were in high school. It’s not a sign of moral decay. It’s a sign of, I don’t know, of the downfall of Western civilization or nothing like that. It’s just a new chapter in the book that’s been written forever. Cheating.
Morgan Sung: It’s not just teachers trying to curb their students’ use of AI. Students are also upset with teachers for using ChatGPT to write lesson plans. In a New York Times report this year, students complained about their professors cheating them out of their tuition money, because in a way, it was ChatGpT teaching them. These issues have existed in education long before generative AI tools did. The existence and widespread use of AI just magnified them. Including this adversarial relationship between students and teachers. But in this scenario, teachers can set an example. If they don’t use AI, then maybe students won’t feel like they need to, either. Okay, now let’s close all these tabs.
Morgan Sung: Close All Tabs is a production of KQED Studios, and is reported and hosted by me, Morgan Sung. This episode was produced by Maya Cueva and edited by Chris Hambrick.
Our producer is Maya Cueva. Chris Egusa is our Senior Editor. Additional editing by Jen Chien, who is KQED’s Director of Podcasts. Our audio engineer is Brendan Willard.
Original music, including our theme song and credits, by Chris Egusa. Additional music by APM. Audience engagement support from Maha Sanad. Katie Sprenger is our Podcast Operations Manager and Ethan Toven-Lindsay is our Editor in Chief.
Support for this program comes from Birong Hu and supporters of the KQED Studios Fund. Some members of the KQED podcast team are represented by The Screen Actors Guild, American Federation of Television and Radio Artists. San Francisco Northern California Local.
Keyboard sounds were recorded on my purple and pink Dustsilver K-84 wired mechanical keyboard with Gateron Red switches.
Ok, and I know it’s podcast cliche, but if you like these deep dives, and want us to keep making more, it would really help us out if you could rate and review us on Spotify, Apple podcasts, or wherever you listen to the show. Don’t forget to drop a comment and tell your friends, too. Or even your enemies! Or… frenemies? Your support is so important, especially in these unprecedented times. And if you really like Close All Tabs and want to support public media, go to donate dot KQED.org/podcasts!
Also! We want to hear from you! Email us CloseAllTabs@kqed.org. Follow us on instagram @CloseAllTabsPod. Or TikTok @CloseAllTabs. And join our Discord — we’re in the Close All Tabs channel at discord.gg/KQED.
Thanks for listening!
