You Need to Understand That Teachers Process Petabytes Every Day

What would a genAI chatbot do with the class I observed last week?

Okay but what if a generative AI agent knew lots and lots about students, the class, and all of its different stores of knowledge? Someone imagined that future in my comments last week:

… don’t you think we’ll get closer to [some ideal application of AI in learning] once the tool has a memory and can, over time, develop its own (technical) understanding of the child’s accumulated knowledge? [..] I could imagine an AI tool, like above, but this time it holds the knowledge of all of the students in a specific classroom, giving it the ability to engage the students in the class as a group, drawing on each students’ knowledge.

---

I visited a math classroom in Richmond, CA, last week. I walked into the classroom and took in the noise. Healthy buzz. A little apprehension given the extra adults in the room maybe.

I noticed the arrangement of the class. Students at individual desks shaped in a U. That’d affect the kinds of conversations we could have. The flow of energy.

I noticed individual students and their relationships. Some high energy, others more subdued. I tried to intuit why they were subdued or high energy. Could we recruit their energy—whatever it was—into math learning? Some of the students would need to be persuaded it was worth their while to generate energy for math. Others had surplus energy but needed to transfer it to math. Both would require different invitations.

The teacher started with a question about sales tax and tipping, asking students what they knew about those features of modern commerce. Extremely useful sensory information flying at us here. Some students will eagerly participate in a conversation about context but less eagerly in a conversation about math. If you can draw those students into a conversation about math, their participation may convince other students that math conversations are possible for them as well.

During classwork, I noticed a kid who seemed stuck on this screen about a broken cash register. I asked him how I could help. He kept looking at the screen. I asked him what he understood about the question. Still looking at the screen. I noticed where he was looking on the screen. Not at the 7% sales tax rate but at the numbers themselves. I figured he was trying to figure out where those numbers even came from. Were they all just arbitrary?

I asked him how he thought the price and tax made the total. We made progress there. I asked him where he thought the tax number came from. Was it random? What if someone tried to scam you and just put whatever number they wanted there? How would you know if it was the right tax?

He pointed to the 7% on his screen. I worked through the sandwich calculation with him. Okay, here is his elbow partner. She’s plugged into our conversation. What is their relationship like? Can I ask them to work independently for a moment and then check their answer for the donut calculation with each other? Let’s try that.

---

I took a nap in the break room at the end of my first full day of teaching. I slept so hard I swear I hit REM sleep and had a dream. My eyes and ears and every other sense had never ingested and processed so much information over such a brief stretch of time.

I’m not sure how to convert all of that information into whatever a petabyte is but I am telling you that a single classroom is loaded with them. The petabytes. Trying to ingest and process the vibes of a class alone, like—

1. How are we feeling today?

2. Where are we at in the semester?

3. Who is ready for how much more thinking right now?

—is not a task for mortals. And much less a task for generative AI as it exists now at the end of 2023. 

No, I cannot imagine generative AI adequately supporting that student or that class. I cannot imagine imagining it. 

This could be, of course, a failure of my imagination. But it’s striking that the majority of my work with that student was non-verbal. He didn’t ask for my help but every one of my senses told me he needed it. As I asked him questions, he answered not by typing thoughts into a chat interface or even forming them out loud. Rather, he answered me, at first, by looking anxiously around a screen. 

To help him, I used visual and auditory and cognitive systems that have been evolving under natural selection for millions of years. My mammalian ancestors hid from predators in the savannah so I could notice a student struggle to calculate tax on a sandwich. I would love nothing more than to see new tools help more students love learning math but it is not a critique of these tools to say, “You are neat, but maybe find a lane that suits you better than classroom teaching.”

It is a critique to say that many of you should hang out in classrooms and watch the work of teaching more often. You should ask yourself more often, “How on earth did that teacher know to make that decision in that moment with that student?” After you answer that question across the hundreds of different moments that every teacher has with their students every day, and maybe even after you try that work once or twice yourself, I promise that you will have everything you need to answer the question, “How will I train a machine to do this work?”