What Happens When Preschool Teachers Wear AI Cameras to Train Education AI
Last month, I dropped my 4-year-old off at our neighborhood preschool and spotted Ms. Rosa, the lead teacher who’s worked with little kids for 22 years, with a tiny black camera clipped to the end of her lanyard, right next to her whistle and printed allergy alert list. She was wiping blue glitter glue off her forearm and herding three kids toward the handwashing sink, so I didn’t ask about it until snack time, when the room settled down and the air smelled like cheddar goldfish and sliced apple.
“It’s a district pilot,” she said, leaning against the counter next to me. “We got a grant to test this out. The camera records all day, every interaction I have with the kids, and they feed all that footage into these big AI systems to teach them how good preschool teachers actually do their jobs.” I asked the first question any parent would: “What about the kids? Do you have consent from everyone?” She nodded, pulled a crumpled stack of forms out of her pocket, said every family got a full explanation, 18 out of 19 signed on, and the one family that opted out just means their kid never appears in any data used for training. All raw footage is anonymized within a week: faces blurred, names stripped, voices broken down into text of what was said, no personal information kept anywhere.
I volunteer once a week anyway, so I hung around the block corner that morning to observe. The first big moment came 10 minutes into free play, when Milo grabbed the last purple crayon right out of Jax’s hand, and Jax sat down and started howling. I’ve seen a hundred teachers step in here: some take the crayon away, some tell Milo to give it back, some distract Jax with a different color. Ms. Rosa walked over, wiped Jax’s cheek with the hem of her sweatshirt, knelt down so her face was level with his, and said, “That hurt your feelings, didn’t it? He took the thing you were working on.” Jax nodded, still sniffling. She turned to Milo: “What did you need the purple crayon for?” Milo said he was coloring a dinosaur’s belly. “Did you ask Jax if he was done with it?” Milo shook his head. Then they talked it out: Jax said he just needed one more minute to finish his flower, then he’d pass it over. Milo agreed to wait, and Ms. Rosa walked away, no big speech, no big intervention. That’s the thing no one writes down in teacher training manuals, right? That small, quiet back-and-forth that de-escalates the moment and teaches kids how to name their feelings. That’s exactly what the camera was there to capture.
Most of what we hear about AI in education is scary or flashy: AI that will replace teachers, AI that grades papers, AI that tutors kids after school. This is the opposite. This is AI learning from the best teachers, so it can take the boring, time-consuming work off their plates. Ms. Rosa told me she spends an average of three hours every evening on work after the kids go home: writing progress notes for 14 kids, drafting weekly lesson plans, answering 20+ parent emails asking what time dismissal is or what to bring for field day. If the AI learns her voice, her approach to working with kids, it can draft all that for her. She just tweaks a few lines, hits send, and done.
I talked to another mom at pickup that same day, and she was worried the cameras were just there to grade teachers, to catch them messing up so the district can cut positions. That’s a fair worry. Ms. Rosa told me the pilot’s contract explicitly says none of the footage is used for teacher evaluation. It’s only there to collect the thousands of small, unscripted interactions that never make it into textbooks. New teachers get a couple weeks of student teaching before they’re leading their own classrooms, right? If an AI is trained on 10,000 hours of interactions from veteran teachers like Ms. Rosa, a first-year teacher can pull up, “What do I do when a kid meltdowns because we ran out of their favorite snack” and get 20 concrete examples of what worked for other teachers, not the generic line you get in college.
If your district ever rolls out a pilot like this, there are three specific questions I now make a point to ask first, that don’t take long to answer. First, ask who owns the data, and confirm it won’t be sold to any third party or used for anything besides training education tools. Second, ask if you can opt your kid out at any time, for any reason, with no negative impact on your child’s placement or care. Third, ask how long raw footage is stored, and confirm it’s deleted right after it’s anonymized. Those three questions cut through most of the hype and fear right away.
It’s not all perfect, of course. Ms. Rosa says some days she catches herself being more performative, like she’s putting on a show for the camera, instead of just doing what she’d normally do. She wonders if the AI will count the 10 minutes she sits next to a non-verbal kid just squeezing playdough with them as “wasted time,” even though that 10 minutes is the most important part of that kid’s day. Last week, she had a rough morning, didn’t sleep well because her own kid was sick, and she snapped a little harder than she meant to at a kid who spilled juice. She worried that moment would be pulled out as bad practice, but then she thought: the part where she knelt down and apologized, and explained that she was tired and that wasn’t the kid’s fault, that’s also good practice. Teachers aren’t perfect, and showing that side matters too.
Last week, Ms. Rosa sent me the first draft of Lila’s progress report that the AI put together, based on two weeks of camera data. It didn’t just say “Lila is creative and social,” like the generic notes I got last semester. It said: “Lila chooses to draw forest animals during 8 out of 10 free play periods, asked three questions about where bears sleep during our last nature walk, and regularly helps the younger kids zip their coats when they struggle on the playground.” Ms. Rosa changed two lines about Lila’s preference for watercolor over crayon, sent it off, and left work two hours early to make it to her 10-year-old’s soccer game. No one knows if this is really the future of education, or if it’ll end up being a well-funded experiment that fades away in a couple years. But right now, the change I see isn’t AI taking over classrooms. It’s AI learning what good teachers do, so teachers can get back to doing it. That’s enough to make me curious, even if I still have a lot of questions.
Please indicate: Thinking In Educating » What Happens When Preschool Teachers Wear AI Cameras to Train Education AI