What People Are For

Thoughts on AI-based tutoring

In the months before I left my position as Director of Academic Intervention for a network of schools in Los Angeles, one out of every five conversations I had was about AI. I am sure this ratio has only increased for my replacement. 

Last week, OpenAI released a new version of its signature application, Chat-GPT. It’s called GPT-4o. This one is multi-modal; it can “look” and “talk” at the same time. One of their demo videos shows the app giving feedback to a man prepping for an interview. A flirtatious, female-coded entity encourages him, teases him when he puts on a hat. This voice caused a stir and has already been discontinued. But for now, you can see it here. 

More to our point, Sal Khan, known for creating a freely-accessible bank of explanations of academic content called Khan Academy, appears in a similar GPT-4o demo. It shows a patient female-coded entity helping Khan’s son with trigonometry. You can see it here. 

Tutoring has been all the rage since the pandemic. It was one of the few interventions with an evidence base that the field can recommend to combat learning loss. The problem has been how to do it at scale. Where will all the tutors come from, and how will we pay them? 

To many education leaders and entrepreneurs, AI provides an answer to both problems. Many of my friends and colleagues are optimistic that we may soon have a private tutor for all children. For the motivated student, of any age, it will certainly be a useful educational tool. But for the children who need tutoring most? 

It won’t work. 

Alien, by Florence

Calvin has settled into being a class clown. He walks awkwardly, his movements influenced by a neurological irregularity. He packs his folders until they are swollen; sometimes they explode into the hallway. When this happens, he leans into the slapstick of frantically picking up each paper, bumping into walls and his peers. Sometimes, as a finale, he throws himself onto the ground. His classmates laugh. He is on track to repeat the 7th grade. 

Aisha doesn’t want to be in my remedial literacy class. During the phonics warm-up, we do choral practice to link sounds with letters. “A, apple, aaaa. O, octopus, awww.”  I notice she’s just mouthing the sounds, not really saying them. She’s not doing the gestures that go with each one, holding out an imaginary apple, wiggling your fingers like an octopus. 

I know from the diagnostic test that she doesn’t know her short vowels. She’s in the 6th grade but reading at a kindergarten level. So I’m not going to let this be. “Let’s start again, with everyone,” I say. “A, apple, aaa.” She looks away from me. “Aisha, you need to say the sounds.” Massive eye roll, huge sigh. But she does it. 

Fanta is a B+ student. She does her assignments, but she has been arguing with the teachers. She seems annoyed by her classmates. I don’t know her very well. 

I like to give every student a fist-bump as they leave the classroom. One day in the fall, at the end of class, Fanta walks toward me, pulls her arm back, and hits my fist as hard as she can. I wince, hold up my hand and let it dangle as if broken. It might be.

As she files past me: the hint of a smile. 

*** 

These are students I taught at a Harlem charter school in the early 2010s. (I have changed their names). I found myself thinking about them last week as I browsed the responses to GPT-4o demos on X (Twitter) and LinkedIn, looking for some skepticism about the promise of private AI-tutors. 

I would much rather join the enthusiasm. It is a marvel that Chat-GPT and other programs can even begin to answer the questions we ask. Every accurate or even nearly-accurate answer should thrill us. The visuals and films are astounding, too, though derivative, as a rule. And the new, more interactive programs are only going to become more sophisticated. 

But when it comes to the kind of tutoring that can change a student’s trajectory, the technology is not there. The theory of those boosting AI-based tutors is that what students need, above all, is easier access to knowledge. That is certainly a useful thing to have. But in my experience, when tutoring is really needed, it is because there are obstacles in the way of learning that current AI-driven programs still cannot perceive, much less alleviate.

***

Calvin was in the 7th grade, an outsider due to his personality and probable neurodivergence. Rather than be “the awkward guy,” he chose to be “the guy who makes people laugh.” As it happened, he was also a fabulous writer. His handwriting was not easy to decipher, but once you did, an incredible sense of sentence structure, and a unique personal voice shone through.

One day before class, I pulled him aside, asked him if I could use his writing as a model. We had a magnificent piece of technology in our classroom: a document camera. During each class, I placed one of his more choice paragraphs under the lens. There was some tittering as his wild handwriting appeared on the board. I said, “Let me read this to you all. I think it’s the best paragraph I have read in a long time.” I asked students to explain why they thought I chose it. We talked about how handwriting and quality of writing were two separate things. There was no more tittering. 

Calvin’s handwriting improved overnight, and he stopped falling down. That winter, he wrote me a letter. “You helped me not to be scared to write how I feel and to show people how smart I am, so I can be taken more seriously.” 

Aisha was a more long-term project. She fought me for a month, refusing to do the movements, refusing to say the sounds. She got very angry with me sometimes, as I would not relent in my requests that she follow along. Why wouldn’t I leave her alone? Let her develop at her own pace? 

Because she was a 13-year old who could not read. A laissez-faire approach was not compassion; it was neglect. I was in my tenth year of teaching at this point, and tried every strategy I could think of to win her over. Ultimately, what worked was success. I was using a technique called Direct Instruction, which is not lecturing, as many people think, but the careful apportioning of learning into achievable bits. Over and over, I asked the class to answer the questions we had practiced, chorally, at first, and then individually. And little by little, she would answer them. Annoyed, but participating.  

I can’t say everything was smooth sailing after that. But by Halloween, the ratio of correct answers to angry sighs had flipped, and by June, she had grown four grade levels according to our standardized assessment. I had the pleasure of telling her this with her mother present. She gave me one of the more powerful hugs I have experienced, and as both of us teared up, thanked me for “not letting her fail.”

As for Fanta, she wasn’t going to fail. But her arguing, her talking back, was getting so bad that some teachers didn’t want her on the field trips. She was not living up to her potential. 

I didn’t know what to do. It seemed too subtle a problem for intervention. But I kept thinking about the little smile. In talking with her, I began to see that she liked her teachers, and she truly liked to argue. She put her all into the persuasive essay unit. And every time I truly listened to what she was saying while “talking back,” I had to admit she made good points. 

My only action was to accept these facts. I didn’t get upset when she wanted to argue; I decided to try and relish the fight, too. As the leader of the 7th grade team, I determined she would be allowed on the field trips. And at the end of every class, I let her punch my fist with all her might. 

There was no great change that I can remember. But eventually her GPA climbed into the A- range, then the A. Her name dropped off the list of students we discussed in team meetings. I know she had great teachers in high school because the last time I heard from her, it was to let me know she had been accepted to Princeton. 

***

Teaching, and tutoring, remain profoundly human endeavors. To observe someone over time, seek to understand them, build a relationship; to express care as well as concern; sometimes to do nothing but hold their feelings; to understand adolescence, and the social context children live in; to make eye contact, human contact–that is what I want from my children’s teachers, and from any tutor who might work with them.

My concern is not with doom-case scenarios, nor do I think we should ban or ignore AI. I am certain this technology will be useful to educators in ways we cannot yet predict. Right now, it can help with translation to better communicate with parents, and these new personal assistants may be invaluable to students with disabilities related to sight or hearing. What I’m worried about, primarily, is that in our excitement, in our credulity, we will spend money on interventions that won’t work for the students who need them most. 

Because here’s the bottom line: the students I mentioned here, when I met them, would turn off the AI tutor at their earliest opportunity. That is what I know from 20 years in schools. And I feel an urgency to say it, because the idea that human tutors can be replaced by an app, even a very sophisticated one, serves a larger narrative that public schools are failing, relics of the past, and should be abandoned. 

Who knows? One day we may invent a program that really can improve upon human teaching. We would likely have to embody its neurology in a physical form, so that it can reach the complexities of perception, not to mention alignment, that only a body in the world can currently achieve. It would be an incredible achievement, a milestone in our collective history.

If I live to see the invention of such an entity, I will happily recant these words. Just please, keep it away from my children.