A Veteran Teacher Explains How To Use Ai In The Classroom The Right Way

When ChatGPT launched successful 2022, Jen Roberts had been school mediate aliases precocious schoolhouse students for much than 26 years and was moving connected fumes. The pandemic had pushed galore educators into burnout, but wherever others saw artificial intelligence arsenic a threat—a exertion that facilitated student cheating—Roberts saw a instrumentality to thief her survive.

An English coach astatine Point Loma High School successful San Diego, Roberts is simply a pioneer of acquisition technology. She has taught pinch one-to-one laptops since 2008, years earlier astir schools adopted them. When generative AI emerged, she was speedy to trial whether it could make feedback faster and grading fairer.

Scientific American said to Roberts astir really she guards against nan misuse of AI and why she believes nan exertion tin thief teachers conflict their ain biases.

On supporting subject journalism

If you're enjoying this article, see supporting our award-winning publicity by subscribing. By purchasing a subscription you are helping to guarantee nan early of impactful stories astir nan discoveries and ideas shaping our world today.

[An edited transcript of nan question and reply follows.]

Many teachers spot AI chiefly arsenic a cheating instrumentality for students. You saw it differently. How did you commencement utilizing it?

I’ve recovered it’s very effective for feedback. When a student takes an [Advanced Placement (AP) English Language and Composition] test, [the free-response conception of] their trial is scored by 2 people. And if those scores disagree, there’s a 3rd score. I thought: What if AI is nan 2nd scorer? If I people it and person nan AI people it, I spot if we agree. If we disagree, nan AI and I person a small chat astir which 1 of america is right.

In my time-constrained world, my remark mightiness beryllium little aliases terse. What nan AI comes up pinch is usually spot-on. I for illustration to opportunity that AI doesn’t really prevention maine time; it conscionable lets maine do much pinch nan clip I have. When I’m utilizing AI-suggested scores and feedback, my students get their penning backmost successful days alternatively of weeks. That intends they’re revising much and revising better.

I besides usage education-specific tools. MagicSchool provides a student-facing action wherever I tin adhd my rubric and duty explanation and past springiness students a link. I’ve seen students put their activity into that four, five, six times successful a azygous period. It’s contiguous feedback that I can’t supply to 36 students simultaneously.

How do you defender against students utilizing AI to constitute for them?

Nothing is simply a magic bullet. It’s a operation of devices and strategies that psychologically person my students I will cognize if they usage AI inappropriately. I require them to do each penning successful a Google doc wherever I tin spot type history. I usage Chrome extensions to analyse nan writing—I tin watch a video playback of their penning process. I besides usage nan old-school method: you’re going to bring your penning to your penning group. Students are cavalier astir turning AI penning successful to me. But if they person to bring it to a penning group, publication it aloud to peers and explicate what they wrote and why, they won’t do that.

I show them ways to usage AI responsibly. You can’t usage it to constitute for you, but you tin usage it for feedback, condemnation frames, outlines and organizing thoughts. If I show ethical usage cases, they’re little apt to usage it unethically. I do an activity wherever I springiness them 3 paragraphs and inquire which 1 is AI. They each instantly know. I say, “You could tell, truthful I tin tell.”

A batch of nan hype astir AI successful acquisition focuses connected customized instruction plans. Is that nan reality?

AI instruction plans are mostly crap. I don’t usage AI often for instruction planning. I usage it specifically to build materials. There’s a Chrome hold called Brisk that lets maine return thing students are reading, creation learning objectives for it and create an interactive tutor to show students really overmuch they understand.

Also, I tin return a page that’s a wall of text, springiness it to [Anthropic’s AI assistant] Claude and say, “Help maine rewrite this. Improve nan clarity, usage color-coding, emojis.” Now I person a page that’s beautiful and easy to understand, pinch colored boxes astir important parts. When students understand what they’re expected to do because directions are clear, that’s really helpful.

In what ways does AI thief pinch nan cognitive load of teaching?

Lots of ways. I often request to travel up pinch a penning prompt. Am I tin of that? Absolutely. Am I tin astatine 4:15 P.M. connected a Thursday day erstwhile I’m really tired? Maybe not. I’ll show nan AI what we’ve been studying and inquire for suggestions. It’ll spit retired 5 aliases six options, and we’ll prime nan 1 that works.

Another example: I was doing an activity pinch a agelong reference that I wanted to break into smaller sections. I didn’t want to walk 45 minutes rereading it to create sensible breaks. I gave Claude nan PDF, and it took only 5 minutes for [the AI] to thief maine reorganize nan material. I besides asked for 40 vocabulary words students mightiness struggle with, organized successful nan bid they appeared successful nan article. That is support I would ne'er person had clip to supply manually.

What warnings would you springiness teachers who are starting to usage AI?

Do not require aliases propose students usage ChatGPT aliases Claude. Those devices are not [compliant with] COPPA [the Children’s Online Privacy Protection Act] and FERPA [the Family Educational Rights and Privacy Act]—federal laws covering children’s privateness and acquisition privateness rights. It’s amended to person students usage devices wrong MagicSchool aliases Brisk that are compliant and that let teachers to show conversations.

Second, do not supply personally identifying accusation astir students to AI. Instead of giving nan full IEP (Individualized Education Program), return nan 1 extremity you’re supporting and say, “How could I support a student pinch this goal?” You get nan aforesaid thief without providing student information.

Can you talk much astir AI-assisted grading?

According to a University of Michigan study, a statistically important chunk of students astatine nan extremity of nan alphabet got little grades and worse feedback, astir apt because teachers get tired. I deliberation of AI arsenic my equilibrium check. When I get to nan student whose past sanction starts pinch Z, and [they had] irritated maine today, americium I giving them a adjacent grade? Often nan AI says, “No, you should beryllium giving them a higher grade.” I look astatine nan activity again and americium like, “It’s right.” If AI tin mitigate that [issue], that’s bully for my students. I spot it arsenic a fairness issue, making judge students get accordant scoring.

Every clip I show teachers astir [the University of Michigan] study, heads nod. We displacement really we people complete a azygous grading session—firm astatine first, loosened up by nan 10th essay, tired and grouchy astatine nan end. We’re human. For each nan concerns astir AI bias, I person much concerns astir quality bias.

A type of this article appeared successful nan March 2026 rumor of Scientific American arsenic “Jen Roberts.”