Well, at least the photo accompanying a Mary Randolph Scientific American story about Alpha School captures how I feel upon coming across one more puff piece about Alpha School.
I feel you, kid.
Alpha School, if you've somehow missed the noise, is a private school chain created by Texas businesswoman Mackenzie Price, her husband, and a real rich guy. It leans on the idea that a child's basic core education can be fully conveyed by a computer program in about two hours a day: Alpha School is directly connected to 2 Hour Learning, because like many private school grifts, the company is one of many companies, wheels within wheels, all handing money to each other, and through each other ultimately to the owners.
Alpha School is expensive-- as pricey as an ivy league prep school, but the buzz has been phenomenal and ongoing. If Price could do education one quarter as well as she does PR, the school might actually have something to offer.
The Scientific American piece is notable for some of the worst arguments I've seen made for Alpha School, by the guy currently employed as the "senior learning scientist" for Alpha-- Carl Hendrick, who really ought to know better.
Carl Hendrick is an internationally recognized science of learning and instructional design expert. I'm about to favorably review a book he co-wrote. I have on occasion forwarded some of his substack offerings. People I respect, respect him.
And yet, the Carl Hendrick in this piece is really out in the weeds.
First, he compares AI in education to Waymo on the road. See, the autonomous car has to learn to accommodate wobbly cyclists and kids running into the street and non-straight roads-- all the things that Waymo, with its limited training for very specific urban areas, still has trouble doing. Sometimes big trouble. But Hendrick predicts that in "five or ten years" AI will be trained on "learning and all of kids' misconceptions and problems."
Thus waving away one of the problems with AI instruction-- the AI can't analyze where the student has gone wrong, what the glitch in their thinking or process might be. Is Hendrick suggesting that AI can programmed with a menu of every possible student misconception and then programmed with a response to it. Because that seems... unlikely. What he's proposing is that the program need not understand why the student is doing what they're doing, just like Waymo doesn't need to know why the bicyclist is wobbling. It's a strictly behavioral approach-- we need the kid to spit out the correct response.
Randolph slips past some issues. "When used by a teacher, adaptive learning models like Alpha School’s have shown promise in improving student performance and understanding." That sentence is doing so much heavy lifting. Used by a teacher? Models like Alpha School's? Promise? And she cites some objections from some academics. Talking to Kelly Miller, the Harvard professor who has worked in this area. Miller likes adaptive learning, calls it "the future of education." But:
But Miller uses AI and recorded lectures to cover basic material so that her in-person time with students can be spent on more complex work. If there’s no teacher-student relationship—like at Alpha School, where adults supervising students generally aren’t trained educators—“what’s the point?” she says.
Then it's all downhill from there.
Hendrick admits that Alpha School's model isn't perfect, "[b]ut he argues that most kids aren’t getting a good teacher or instruction grounded in learning science in a classroom anyway." So... their human teacher is probably crappy, so why not give them a crappy AI program instead? Hendrick argues that people would be shocked at the varying degrees of quality within schools. In other words, AI "educators" may be bad, but humans are worse-- or at least very inconsistent.
That’s Hendrick’s biggest argument for Alpha School’s model as a solution: “You can’t scale good teaching,” he says. Software, Hendrick argues, can be updated across classrooms at once as new curriculum and instructional research emerges and can give feedback far faster than a teacher grading assignments one by one.
Emphasis mine. Added while picking my jaw off the floor. You can't scale good teaching-- so let's scale mediocre-at-best teaching? And when it comes to feedback for students, which is more important-- being fast or being good? "I got feedback on my essay instantly. It's kind of dumb and useless and even wrong, but at least I got it back quickly."
Randolph nods briefly to the 404 Media report that found all manner of problems with Alpha School. It's not the first negative coverage (here's a Wired piece about unhappy Alpha parents in 2025).
Randolph also allows one quote that questions the selection bias involved (families that fork over that kind of money for private school tend to be highly motivated).
She also notes that the model will be tested in some public schools, which may show if the 2 Hour AI model works with a "broader" assortment of students, but we already have some data about that. But as this piece from The Conversation points out, much of the "data" we supposedly have is bunk. And as Dan Meyer showed, 2 Hour Learning has already been tested on general population students-- and it failed hard.
The irony here is that we are not talking about a revolutionary new technology-- the idea of a teaching machine that can guide students to cough up proper responses is over a century old. Read Audrey Watters Teaching Machines. Alpha School is parked firmly within a forest of red flags, and guys who are learning experts ought to be able to see them all waving.
No comments:
Post a Comment