Orbrya
All posts
Orbrya2026-03-12

The Difference Between Teaching Kids to Use AI and Teaching Kids to Audit AI

Most AI education teaches kids to use AI more effectively. A smaller, more important category teaches kids to evaluate what AI produces. Here's why that distinction is the one that matters for your family.

The Difference Between Teaching Kids to Use AI and Teaching Kids to Audit AI

A parent named Cristina Mugilia, a homeschool mom who also works at the Department of Veterans Affairs, put it plainly: "A machine that just has lots of facts and data is going to spit out whatever you give it. Use your mind. Put your thoughts together. Make an argument."

That's not an anti-AI sentiment. It's a description of the one thing AI can't do for a child no matter how well they learn to use it: develop the capacity to evaluate whether what just came back is actually right.

We've spent the last four weeks building toward this distinction. This post is where it lands fully. The difference between teaching kids to use AI and teaching kids to audit AI isn't a minor pedagogical detail. It's the entire question, and the answer determines what kind of AI relationship your child will carry into adulthood.

What "Teaching AI" Usually Means

When a school, a course, or an app claims to teach AI literacy, the content almost always falls into one of two categories.

The first is AI awareness: understanding what AI is, how it works at a conceptual level, what it's being used for in the world. This is valuable background knowledge. It isn't a skill.

The second is AI usage: learning to write better prompts, use AI tools for specific tasks, integrate AI into workflow and study habits. This is genuinely useful. Most people who use AI effectively have developed some version of these skills, often through trial and error rather than explicit instruction.

Both of these categories teach students something real. Neither of them teaches the skill that determines whether AI use strengthens or weakens a child's capacity to think.

Sam Altman, the CEO of OpenAI, called ChatGPT "a calculator for words" in remarks reported by the Harvard Gazette in May 2024, and the analogy has been widely adopted by educators who see AI tools as productivity amplifiers that free students to focus on higher-order thinking. It's a reasonable analogy in some respects. It breaks down in one crucial way: calculators don't produce confident, grammatically correct, plausible-sounding wrong answers. When a calculator fails, it visibly fails. When an AI language model fails, the output looks identical to the output it produces when it succeeds.

That asymmetry is what makes the usage-only approach insufficient. A student who learns to use AI effectively, and nothing else, has learned to work efficiently with a tool they can't reliably evaluate. The efficiency is real. The inability to catch errors is also real. Both things are true at the same time.

What Auditing Adds

Teaching kids to audit AI means teaching them to ask, for any given output: is this actually right, and how would I know?

That sounds like a small addition to AI usage skills. In practice, it's a different orientation entirely. A student who has learned to use AI treats its output as a resource to work with. A student who has also learned to audit it treats its output as a claim to be evaluated. The tools they use are identical. The cognitive posture they bring to those tools is not.

The research behind why this matters was the subject of our post on the two types of critical thinking. Researcher Chahna Gonsalves at King's College London identified the distinction formally: critical thinking toward the AI is the evaluative layer that usage skills alone don't build. Most educational AI content develops the downstream type, the thinking that happens after the output has been accepted, while leaving the upstream evaluative layer to chance.

Joshua Lockair, a homeschool parent who teaches computer science at the university level, described the consequence with precision: "Students must understand how AI works to evaluate whether it's giving the right answer. Even if it does give the right answer, they're not developing the process or skills to understand the concept. It's like skipping to the end."

Skipping to the end produces students who get assignments done. It doesn't produce students who can stand behind what they submitted, explain the reasoning behind it, or catch it when the AI led them to a wrong conclusion they accepted without noticing.

Why the Frontier Makes Auditing Non-Negotiable

Harvard Business School researcher Ethan Mollick and colleagues ran a field experiment with 758 consultants at Boston Consulting Group, studying what happened when professionals used AI on real work tasks. The headline finding was that AI users significantly outperformed non-users on tasks within AI's capability range: more tasks completed, faster, at higher quality.

The less-cited finding is what matters here. On tasks that fell outside AI's reliable range, users of AI were about 19 percentage points less likely to produce correct solutions than consultants who worked without it. The same tool that boosted performance on one category of task actively hurt performance on another. Mollick described this as the "jagged frontier" — an invisible boundary between what AI handles reliably and what it doesn't, with no visible signal when you're crossing it.

The consultants in that study were adults with professional expertise. They had the domain knowledge to notice, at least some of the time, when an AI output seemed wrong. Children working on assignments in subjects they're still learning don't have that backstop. If AI produces a confident, well-formatted claim about a historical event, a scientific process, or a mathematical relationship, a student who hasn't learned to verify has no reliable mechanism to catch the error.

Teaching usage skills without auditing skills is, in this light, teaching children to move efficiently through a landscape with hidden hazards while leaving them without the tools to see the hazards coming.

The Hingham Case: Both Sides of the Frontier in One Story

In December 2023, a junior at Hingham High School in Massachusetts submitted a National History Day project on Kareem Abdul-Jabbar's civil rights activism for his AP U.S. History class. The student, identified in court documents as RNH, had a 4.3+ GPA, a 1520 SAT score, and a perfect ACT. His teacher ran the project through three AI detection tools and raised concerns about AI-generated content.

What followed illustrates the consequence of the auditing gap from multiple directions at once.

The family's position is that RNH used AI for research and outlining, not for writing the final text. The court's preliminary ruling, issued in November 2024, found the student had included text generated by Grammarly along with citations to books that don't exist — including "Hoop Dreams: A Century of Basketball" by a "Robert Lee" and "Muslim Pioneers" by a "Jane Doe." Neither book is real. Both were presented as sources.

The school had no AI-specific policy in its handbook at the time. RNH received a failing grade, Saturday detention, and was barred from the National Honor Society. His parents sued in federal court. The judge denied the family's motion for a preliminary injunction, writing that the school had the better of the legal argument at that stage, though the case has not concluded.

Jennifer Harris, the student's mother, said: "They told us our son cheated on a paper, which is not what happened."

The framing matters and both sides deserve acknowledgment. But the detail that cuts across any framing debate is the fabricated citations. Whether RNH wrote the text himself or not, citations to books that don't exist were included in the project and submitted. The verification step that would have caught them, looking up the sources before including them, didn't happen. That step is exactly what auditing teaches.

The Hingham case isn't primarily a story about a student who cheated or didn't cheat. It's a story about what happens when a student uses AI without the evaluative layer that would have kept them safe, regardless of their intentions. The consequences were real: a failing grade, legal proceedings, a college record at risk, and a family's account of a years-long ordeal, all traceable to citations that a five-minute verification check would have caught.

What the Wage Data Says, Carefully

PwC's 2025 Global AI Jobs Barometer analyzed close to a billion job postings and found that positions requiring AI skills command a wage premium of 56% over comparable roles without that requirement. That figure has been cited widely as evidence that AI literacy matters for workforce outcomes.

The caveat worth knowing: the PwC data measures the same occupation with and without AI skill requirements in the job posting. It doesn't distinguish between usage skills and auditing skills. It measures AI competence as a category, not the specific type of AI competence that drives the premium.

What the research can support is that AI-competent workers earn significantly more. What it can't tell us, because no study has measured it directly yet, is how much of that premium accrues to workers who can use AI versus workers who can also evaluate it. That distinction is the frontier where the research is still forming.

What the legal cases from our earlier post do suggest is that the professional consequences of usage without auditing are already visible and already costly. The $10,000 fine a California attorney received for submitting AI-fabricated citations didn't come from using AI. It came from using AI without evaluating what it produced.

The Decision Families Are Making Right Now

The families reading this post are making a practical decision, whether they've framed it that way or not. Every day their child uses AI without explicit instruction in evaluation, usage habits form and deepen without the auditing layer attached. Habits formed before the auditing layer arrives are harder to retrofit. Habits formed with the auditing layer built in from the start don't require retrofitting at all.

This isn't an argument for restricting AI access. The research on AI as a learning tool is genuinely mixed, and there are real benefits to children having access to capable tools. The argument is for sequencing: making sure the evaluative habits are established before or alongside the usage habits, rather than hoping they develop on their own downstream.

The tools for doing this don't require a curriculum, a subscription, or technical knowledge. They require the Three Questions Framework applied consistently. They require the metacognitive pause before accepting output. They require the age-appropriate games that make verification feel like a habit rather than a chore.

Mugilia's formulation is the right one: use your mind. AI is available, it's capable, and it's not going away. The question is whether children learn to be the boss of it or to be dependent on it. That answer gets determined by what they're taught to do with the output once it arrives.

That closes out our first four weeks. Starting next week, we move into the practical side of building these habits systematically, including what a structured AI literacy curriculum actually looks like across different ages and how families can start immediately before the formal curriculum launches in August 2026.

If you want early access to the curriculum and to be notified as new posts and resources go live, join the waitlist.

Orbrya builds AI literacy curriculum for K-12 families — homeschool and supplemental — focused on the verification and evaluation skills that separate effective AI use from dependent AI use. Curriculum launches August 2026.

Sources cited in this post: