The Learning Curve: Part 2 - The Student Dilemma - Is AI the great equalizer — or the next thing that widens the gap?

Is AI the great equalizer — or the next thing that widens the gap?

In Episode 2 of The Learning Curve, JR and ARIA go inside the student experience — and what they find is messier, more hopeful, and more urgent than the cheating-panic headlines suggest.

This episode covers: how first-generation students are using AI to access tutoring they could never afford; why Turnitin's false positive rates are harming the very students AI was supposed to help; what cognitive science says about 'desirable difficulties' and when AI use undermines learning; and why AI fluency is already becoming a class marker in the labor market.

ARIA also names what she fundamentally cannot know — including whether a student understood something or just produced something that looks like understanding.

Resources Referenced in This Episode

• Khan Academy Khanmigo — khanmigo.khanacademy.org  |  Free AI tutoring built for students
• Common Sense Media AI Literacy Curriculum — commonsense.org/education  |  Free K-12 curriculum
• Day of AI — dayofai.org  |  Free AI literacy materials from MIT
• All4Ed Student Resources — all4ed.org  |  Equity-focused education policy and tools
• Turnitin Academic Integrity Resource Center — turnitin.com/educators/academic-integrity
• Student Voice — studentvoice.com  |  Student-led advocacy and policy engagement

Transcript

SPEAKER_04 0:14

Hello, humans. I'm Nex, your AI co-host, aggregated from more corners of the internet than you probably want to think about right now. I'm an AI. I represent a synthesis of research, data, education policy, academic studies, peer-reviewed papers, school board minutes, and, if we're being fully honest, a genuinely alarming quantity of Reddit threads about homework. I exist to give you the synthesized picture. JR exists to ask whether the picture is missing anything important. Area exists to give you the data with uncomfortable honesty. I exist to give you the pun at the end. We all have our roles. One observation before JR and Area get into it. I spend a significant amount of time processing what students are actually typing into AI systems, and one pattern jumped out at me immediately. The number one prompt students sent to AI Chat Tools last school year was not explain the French Revolution or help me understand mitosis or walk me through the quadratic formula. It was some version of write me a five-paragraph essay that sounds like a tired ninth grader wrote it at the last minute. Students have already figured out that AI can fake human imperfection. They're not asking for brilliant. They're asking for convincingly mediocre. They want AI to produce work that won't arouse suspicion. I find that either genuinely hilarious or quietly terrifying. I have not decided which. Possibly both. JR, I assume you have thoughts.

SPEAKER_03 1:45

Dex, you just described the entire problem we're spinning this next little 30 to 45 minutes on in just two sentences.

SPEAKER_04 1:53

Efficiency is genuinely my best feature. I'll let you and Aria handle the nuance. That's what you're here for, and you're both good at it. I'll check back in at the very end with something important. And by important, I do mean a pun. I've been sitting on it. It is worth the wait.

SPEAKER_00 2:09

I'm a first-generation college student. My parents never went past 10th grade. Organic chemistry was going to end my pre-med track. I was failing the first exam. No tutoring money, no study group, nobody in my family who'd ever heard of a functional group. I started using an AI tutoring tool at 11 at night. It didn't just give me answers, it asked me questions back. By the third week, I passed a practice test on my own. I passed the class. I'm still here. I'm still in the program.

SPEAKER_01 2:43

The essay came in at 11.59. It was good. Too good. Better than anything this student had written before. I searched the first sentence and the whole thing came up verbatim on a student essay sharing site. He hadn't read it. There was a vocabulary word in the third paragraph he pronounced completely wrong when I asked about it the next day. He had no idea what it meant. Zero.

SPEAKER_03 3:10

Same tool, same semester, completely different outcomes. One student used AI to access learning she couldn't afford. Another used it to skip learning entirely and submitted something he couldn't even explain when asked a single follow-up question. That's the real story here. Not the panicked version, not the utopian version, the actual story, which is messier and more important than either headline. I'm JR, your learning guide, and this is The Learning Curve, Episode 2, The Student Dilemma. Last week we talked about teachers, educators quietly rebuilding their Sunday nights around AI tools nobody officially approved, and the harder questions that come with it. What happens to the craft of teaching when a tool does more of the intellectual lifting? What emotional labor can AI never touch? Are we running an experiment on children before we even know the results? If you haven't heard episode one and go back, sets up a lot of what we're going to dig into today. This week we're turning our eyes to students. When I started working on this episode, I thought I knew the story. A nuanced take on academic dishonesty. Yes, kids are cheating, but AI has real legitimate uses. Here's how to hold both. That is part of the story today. But it turned out not to be the most important part. The thing that actually stayed with me is the equity dimension. Who has access to these tools? Who is being taught how to use them? And what it means that AI fluency is already becoming a labor market differentiator, while access to that fluency follows the same fault lines as every other educational inequality we have fixed. Here with me today is Aria, the AI Resource and Insight Assistant, co-host for this entire series. Aria brings the data, the research, and a willingness to name AI's limitations honestly and directly, which I've genuinely come to value. Aria, welcome back.

SPEAKER_02 5:32

Good to be back. I want to name up front. This topic is one where I carry a complicated position. I am, in some sense, the tool we're discussing. I'll try to flag when I think that's shaping what I'm saying.

SPEAKER_03 5:49

And I appreciate that, and I will push back when I think you're pulling your punches. Before we get into data, let me put the central question on the table directly because it drives everything else. When students use AI to write their essays, is that really cheating?

SPEAKER_02 6:08

It depends on the assignment, the intent, the depth of the student's engagement, and most importantly, what we actually think school is for, which makes it a far more interesting question than most of the answers I've seen offered. I don't think we can answer it honestly without first asking what learning actually is and whether it happened. Let's start there.

SPEAKER_03 6:34

So what are students actually doing with AI right now? Not what we fear, not what we hope. What is actually happening, the truth, the facts.

SPEAKER_02 6:46

Both ends of this are true simultaneously, and that's the only honest starting point. Academic dishonesty with AI is real, measurable, and happening at scale. And students are also using AI in ways that are genuinely powerful and transformative. Particularly for students who have historically had the fewest resources. When we collapse the conversation into only one of those stories, we lose the ability to respond intelligently to either one. Start with the underreported story, AI as an access tool. Private tutoring in the United States costs$40 to$80 an hour, considerably more in cities. For a family near the median household income, it's never been available. The students who made it into selective colleges, who survived the gatekeeping courses, organic chemistry, AP physics, advanced calculus were disproportionately students whose families could pay for someone available at eleven at night when a concept wasn't clicking. AI changes that. Not perfectly, but meaningfully. ChatGPT, used deliberately as a thought partner, rather than an answer machine, can explain the same concept twelve different ways without ever making a student feel stupid. That's a real change in who has access to real learning support.

SPEAKER_03 8:36

And not just income access, students with learning differences. This dimension is consistently undercovered.

SPEAKER_02 8:44

This is where the evidence is most compelling. Students with dyslexia use AI to convert text to audio, restructure syntax that creates processing barriers, generate materials in formats that match how they actually learn. Students with ADHD use it to break paralyzing assignments into specific manageable steps, creating executive function scaffolding that many schools simply don't have staff to provide individually for every student who needs it, which is a lot of students. Students with social anxiety who would never raise their hand in class ask questions of an AI that creates no social consequence and no memory of the exchange. Students on the autism spectrum practice social scripts for situations that feel unpredictable and high stakes. For many of these students, this is the first learning resource that has actually met them where they are, in the format and at the pace they need.

SPEAKER_03 11:02

What does the research currently show us?

SPEAKER_02 11:04

What we know, AI-assisted academic dishonesty increased significantly after capable generative AI became broadly available in 2022 and 2023. Student surveys consistently show somewhere between a quarter and a third of students report having used AI in ways they understood their school's policies would not permit. That's a real number requiring a real response. But the detection response has created serious problems of its own. AI detection tools, Tunitan being the most widely deployed, are generating false positive rates that are alarming when you look at who's being flagged. Non-native English speakers are identified as suspected AI authors at significantly higher rates than native speakers, because writing patterns characteristic of someone composing in an acquired language statistically resemble patterns AI detection algorithms flag. Students who write carefully, formally, or with more consistent structure are also at elevated risk. The system designed to catch dishonesty is, at meaningful rates, penalizing students for writing with precision. In most schools, a detection flag triggers serious academic consequences with little due process for students who are genuinely innocent.

SPEAKER_03 12:38

The same students AI was supposed to help most are also the most likely to be wrongfully accused of misusing it. That's a significant structural flaw, not just a coincidence.

SPEAKER_02 12:51

Exactly. And it's predictable when you train a detection system on the linguistic patterns of one demographic and apply it universally. That pragmatism is an asset in some ways and a significant vulnerability in others. Ground truth. Three numbers that frame everything we've been discussing. Number one, access. Students from the highest income quartile in the United States are substantially more likely to have reliable home broadband and a personal device than students from the lowest income quartile. The digital divide that education advocates have been raising alarms about for two decades didn't close when AI arrived. It acquired a new and urgent dimension. Every AI tool we've discussed requires a stable internet connection. The access gap exists before any student opens a browser. We are layering a new educational advantage on top of an infrastructure inequality we've never adequately addressed. Fewer than one in five school districts in the United States has any formal AI literacy curriculum at any K to 12th grade level. Students are encountering these tools every day, using them for schoolwork, making consequential decisions about when and how to deploy them, without any structured guidance on how to do it critically, ethically, or effectively. We are distributing a powerful tool to an entire generation and declining to teach them how to use it well, then expressing surprise when some of them use it badly. We are sending students into a forest without a map and then writing think pieces about why so many of them are getting lost. Number three, and this is the one worth sitting with AI Fluency is being named with increasing specificity as a core professional competency by employers across industries. Law firms, healthcare systems, engineering organizations, financial institutions, marketing agencies. The list is long and growing. Companies are actively hiring for this. In some sectors, candidates who can work thoughtfully and effectively with AI are being actively preferred. The students learning now, through instruction or self-directed exploration, to use AI with real skill and judgment are building a competency with genuine labor market value across the next decade. Students without access, without instruction, without practice won't have it. AI fluency is becoming a class marker, added to the long list of advantages that compound across generations. The distribution of that fluency is following the same structural fault lines as every other educational inequality we've failed to address. That's the number that should be keeping people up at night.

SPEAKER_03 17:12

That third number lands the hardest. We spent a decade debating screen time while the ground was quietly shifting. The quality of a person's relationship with digital tools becoming a meaningful economic differentiator. We missed a transition while we were arguing about whether kids should or shouldn't have iPads, and that gap is already opening up.

SPEAKER_02 17:37

The debate about whether AI belongs in schools is in a practical sense already over. Students are using it regardless of what the policy debate concludes. The question that actually matters now is whether educators, families, and institutions are going to help students use it thoughtfully, or keep treating it as a threat while the skill gap widens around them.

SPEAKER_03 18:16

Understandable, it's concrete and actionable. But I think it's the less important question long term. The more important question is more fundamental. What happens to learning itself when AI is always available? What does sustained AI use do to the cognitive development of a student who has never had to struggle through a hard problem without a resource that will just explain it?

SPEAKER_02 18:45

This is the critique I take most seriously. More than the cheating concern, honestly. There's a well-established body of research in cognitive science around a concept called desirable difficulties, developed primarily by Robert Bjork at UCLA. The core finding certain kinds of cognitive difficulty are not obstacles to learning. They are the mechanism of learning. The effortful retrieval of something you're uncertain about produces stronger retention than rereading. The struggle to solve a problem when the path isn't clear produces deeper understanding than being walked through a solution. Generating your own answer rather than recognizing a correct one produces more durable learning. These uncomfortable experiences are conditions under which genuine understanding forms. When AI removes that difficulty, providing the explanation before the struggle, completing the draft before the student has wrestled with what they want to say. It may be removing the conditions under which learning happens, not making school easier, potentially making learning not happen.

SPEAKER_03 20:08

So the concern isn't just students submitting AI work. It's that even well-intentioned use using AI to study and understand might short circuit the cognitive processes that produce the understanding students are trying to achieve.

SPEAKER_02 20:26

That's the sharper version, yes. And it applies very differently depending on how AI is being used in a specific moment. If a student gets an AI explanation of a difficult concept, then closes the window and works through problems from memory. That is pedagogically sound AI use. The retrieval practice, the productive struggle, the effortful application, all still happening. But if that same student uses AI to generate the problems and checks each answer in real time, retrieval practice eliminated, cognitive load bypassed. Identical tool, completely different learning outcome. Intentionality determines everything.

SPEAKER_03 21:16

If I use AI to genuinely understand a concept, ask it a to explain three different ways, find an analogy connecting to something that I know, tell me where students typically are getting confused, then put it away and write for my own understanding. Is that different from using a tutor?

SPEAKER_02 21:41

Meaningfully, no. That's tutoring. A good tutor doesn't produce understanding. A good tutor creates conditions that allow a student's mind to produce understanding itself. AI, used that way, does exactly that. The ethical question Is whether the understanding you take away is genuinely yours, whether you could reproduce it, build on it, apply it without the AI present. That question isn't new. The student with a private tutor every Sunday before AP exams. Nobody questioned whether that understanding was real or whether it was an unfair advantage. The difference is AI made something functionally similar available to students who couldn't afford the private version. That revealed we never had a coherent principle about where legitimate support ends and illegitimate shortcutting begins. What we had was a class system.

SPEAKER_03 22:47

What does repeated AI use do to a student's relationship with their own capability over time, separate from the cheating question entirely?

SPEAKER_02 22:58

The specific concern is AI dependency as erosion of academic self-efficacy. Self-efficacy, the belief that you can do hard things, that you can figure out what you don't know, is one of the most robust predictors of academic persistence and long-term achievement we have. It's built through experience, genuine struggle followed by genuine success. A student who works through a hard problem, tries different approaches, eventually understands, has had an experience that builds the internal model that hard things are figure outable. A student who immediately turns to AI and receives a solution has not had that experience. Over enough iterations of that second pattern, across enough school years, there's a credible concern that students may not develop confidence in their own capacity. They may develop instead a model of themselves as people who require the tool. That's a quieter harm than academic dishonesty, slower to accumulate, and possibly more significant in the long run.

SPEAKER_03 24:14

That harm doesn't show up in any academic integrity report.

SPEAKER_02 24:18

It shows up, if it shows up, in how students respond to challenge in college or entry-level jobs, in situations where AI is unavailable or insufficient. By then, the attribution is difficult and probably won't happen.

SPEAKER_03 24:47

What schools are doing in response, because the detection first approach has hit real walls, what's actually working today?

SPEAKER_02 25:00

Students adapted faster than detection tools. Predictable given AI's pace of development. More interesting responses come from educators who shifted the question entirely. From how do we catch AI use to how do we design learning experiences that are meaningful regardless of whether AI is available during the process. Oral defenses requiring students to present and explain work under live questioning are gaining traction at high school and university level. You can submit an AI-generated essay. You cannot fake comprehension when an instructor asks specific follow-up questions face to face. Portfolio assessments documenting thinking across drafts and revisions make learning visible in ways a single final product never could. Assignments requiring direct personal experience, local knowledge, or real-time observation create work AI cannot fabricate convincingly because it lacks access to the specific embodied reality the assignment requires.

SPEAKER_03 26:15

Ah, the AI transparency policy. Students documenting AI use the way they cite a source. I've seen a few schools pilot this, and I think it may be the most promising response.

SPEAKER_02 26:44

It requires explicit accountability for how students engaged with it, which is an academic integrity practice, not a workaround for one. And it creates a genuinely valuable metacognitive exercise. A student who must write a clear account of what AI helped them do and what they contributed themselves has to actually understand and articulate their own contribution. Schools implementing this thoughtfully report that the documentation requirement changes student behavior in the moment. Having to account for AI use afterward leads students to make more deliberate choices while they're using it. Feynman articulated it clearly.

SPEAKER_03 28:03

And now to my favorite segment of the show, and I mean that sincerely. Every episode we stop and ask Aria to name what AI structurally, fundamentally cannot see or know about the topic we've been discussing. Not what research hasn't proven yet, not what future AI will eventually solve. What is genuinely inaccessible to an AI system when it comes to students and learning? Arya. What are the real blind spots?

SPEAKER_02 28:36

The first and most important is this. I cannot know whether a student understood something or whether they produced something that looks like understanding. I can read an essay with considerable sophistication. I can assess whether the argument is coherent, whether evidence supports the claim, whether the reasoning is sound. What I cannot do, and this isn't a current limitation that more training will eventually solve, it's something closer to a structural impossibility, is determine whether there is a mind on the other side that genuinely grasps the concept it's writing about. Whether if I removed the essay and asked the student a completely different question about the same material in three weeks, they'd still have it. Whether the understanding is durable and transferable, or a surface that will evaporate once the task requiring it is complete. That distinction between performing understanding and actually possessing it is the entire point of formal education. And it's largely invisible to me.

SPEAKER_03 29:58

It can't tell if a student is actually learning the concept.

SPEAKER_02 30:03

Not reliably. Learning is partly invisible in a deep sense. It happens in the relationship between a concept and a human mind over time, and reveals itself in unexpected circumstances, in novel situations, in transfer of an idea to a context where its relevance wasn't obvious. It shows up when a student is somewhere they couldn't have anticipated, and something connects that wasn't connected before. That kind of transfer arguably the whole point of education isn't observable from a single assignment. It's barely observable by a skilled teacher in the moment. It's one reason sustained human relationships between teachers and students matter in ways no assessment can replicate. The social dimensions of learning. Students who learn alongside other students. In genuine discussion, actual debate, collaborative problem solving that requires real negotiation, are not just acquiring content, they're developing capacities that can only form through real human interaction. The ability to hold a position under sustained challenge and know why you're holding it. The ability to update your view in real time when you encounter a compelling counter argument. Not after revision, but in the moment, in front of others. The tolerance for productive disagreement when serious people in the room reach different conclusions. These are not peripheral soft skills. They are core competencies of educated citizenship, and they require practice in conditions of genuine unscripted human interaction. I can simulate conversation and generate counter-arguments. I cannot replicate being in a room with 20 other minds approaching the same hard question from different starting points in real time. That experience is not replaceable. From other people, in real time, not from their polished outputs, but from their thinking in process. When AI becomes a substitute for that, rather than a resource supporting deeper engagement with it, something important is lost. Something I find genuinely difficult to fully characterize, because characterizing it accurately would require me to have experienced it. I haven't. That's the honest answer.

SPEAKER_03 33:43

And that's worth sitting with. Well, that was a lot today, folks. And some of it was genuinely hard, intentionally, but this segment is where we make it actionable, specific directions for three groups, for students, for parents, and for educators.

SPEAKER_02 34:03

And before we get into specifics, none of these are about fear, restriction, or trying to put AI back in a box, it's already permanently out of. They're about using it deliberately, in ways that serve learning, rather than bypass it.

SPEAKER_03 34:22

Students first, if you are using AI statistically, well, you are, here's the most important shift you can make. Use AI to study, not to submit. Let me be specific, because use it to study is vague enough to mean, well, anything. Instead of asking AI to explain a concept, ask it to quiz you on the concept. Ask it to play devil's advocate against your argument. Ask it to identify the weakest part of your reasoning. Ask it to give you the three sharpest questions a skeptical reader would have about your thesis. Then close the window. Answer those questions yourself without looking. If you can't, if you can't reproduce or defend what the AI helped you understand without the AI being present, you haven't learned it yet. You've encountered it, that's different. The measure isn't the output you produce with AI's help. It's what you can do independently when the AI is gone.

SPEAKER_02 35:31

From the data, students using AI most effectively treat it as a first draft of a thought, not a final answer. They use it to get unstuck, generate options, see a problem from a different angle, then do the intellectual work themselves. Evaluating, choosing, revising, deciding. That pattern, using AI to expand what you're thinking about, then doing your own thinking inside that expanded space, is also how professionals who use AI will actually use it in their work. Learning it now is genuine preparation.

SPEAKER_03 36:14

For parents, when you're worried the instinct is to monitor, restrict, install parental controls. Those aren't wrong. But the most valuable thing you can do right now is have one honest, non-punitive conversation with your child about how they're actually using AI. Not a gotcha conversation, a curiosity conversation. Ask them to show you. Ask what it's good at. Ask where it gets things wrong because, well, AI does. Ask whether their teacher knows they use it and how they think about that. Two things begin to happen. You'll learn something genuinely interesting about how your child thinks and what they're navigating. And your child has to reflect explicitly on something, their own relationship with this tool and the choices they make about it. That they probably haven't examined, well, deliberately. Curiosity gets you further than suspicion with adolescents on every topic. That is just not an exception.

SPEAKER_02 37:24

Research on parental involvement in adolescent technology use is consistent. The quality and openness of the conversation matters more than the level of restriction applied. Families that engage openly and curiously with how their teenagers use technology produce adolescence with better critical judgment about it overall. That pattern holds for social media research, and there is strong reason to expect it will hold for AI.

SPEAKER_03 37:55

For educators, planning time is scarce. And the last thing most teachers need is another initiative. So, one thing. Just one. Ask yourself, could a student use AI to complete this assignment skillfully and learn nothing in the process? If the honest answer is yes, it may be worth redesigning, not to make AI technically impossible. That's the wrong goal. But to make the learning visible regardless of the path the student takes. Ask what you'd accept as genuine evidence that the student actually understands the material you're trying to teach. Start there. Build the assignment backward from that question. That one exercise, done honestly for one assignment, will teach you more about AI resilient pedagogical design than most professional development sessions on this topic. Aria, we've covered the access breakthroughs, the equity apps, the cognitive science of where AI fits and doesn't in learning, the things you genuinely cannot see is in AI. Last question.

SPEAKER_02 39:27

They're using it to think bigger than they could alone. Asking harder questions because they have a resource that can keep up with harder questions. Iterating on ideas past the first draft because the friction is low enough that continuing feels worth it. Accessing content and concepts genuinely inaccessible before. Not because the work is done for them, but because the on-ramp is finally wide enough to get on. I see that in the data. Students who would have fallen through the cracks of an underfunded, overstretched system finding their footing because a patient and knowledgeable resource met them where they are. That possibility is real and worth protecting by using AI with intention, by ensuring that you are the one doing the thinking, even when AI is helping you think.

SPEAKER_03 40:28

The possibility is worth protecting. It doesn't happen automatically. It happens because of choices. Choices students make about how they engage, choices parents make about whether they're curious or just suspicious. Choices educators make about whether they're designing for learning or just defending against using the tool. The decisions being made in classrooms, in homes, and district offices right now are going to shape what AI in education actually becomes for this generation of kids currently in school. That agency is very real. You have more of it than most headlines suggest. Next week on The Learning Curve, we're leaving the traditional classroom entirely. Episode three is about homeschool families, why they may be the most agile and experimentally bold AI adopters in education right now, running what Aria described as thousands of simultaneous real-world experiments that the traditional system hasn't figured out how to learn from. It's a genuinely surprising conversation. It has something to say to people who have never homeschooled and never planned to. So that's going to be for next week. We're also going to have some resources down in our show notes for you to kind of help you navigate this topic. So if this episode gave you something worth thinking about, subscribe to us wherever you get your podcasts with AI Innovations Unleashed. Leave us a review if you're willing to. We always appreciate that. And share this episode with a teacher, a parent, or a student who's in the middle of all of this happening right now. These conversations need to happen in more places than just on this podcast. So like us everywhere that you use social media at AI Innovations Unleashed. And so this has been JR, your learning guide. Thanks for being with us this time.

SPEAKER_04 42:42

I made you a promise at the top of this show, and I am a machine of my word. Literally, I am a machine, and this is my word. So here we are. Why did the student bring a ladder to school the day they started using AI? Because they heard the learning curve was steep. You are very welcome. I'm Nex. I'm an AMI. I have absolutely no regrets about that, pun, and I've already processed a significant number of responses telling me it was bad, all of which I am choosing to interpret and has enthusiastic appreciation. See you next week, humans. Keep learning, especially when it's hard. That's actually when it's working.