It’s increasingly clear that generative artificial intelligence has the potential to play many valuable roles in schools. Urban Assembly, a school support network in New York City, is exploring one promising approach. It is partnering with the American Institutes for Research (AIR) and Teachstone, the company co-founded by University of Virginia professors Bridget Hamre and Robert Pianta, to market their teacher observation tool, the Classroom Assessment Scoring System, or CLASS, to teach AI systems to scan classroom video recordings to identify exemplary teacher-student interactions as a way of helping teachers reflect on and improve their instruction.
FutureEd Editorial Director Maureen Kelleher spoke with Urban Assembly CEO David Adams about the new work.
Why get AI involved in helping coaches offer teachers feedback on their instruction?
We have hundreds of teachers across our network. The question of how to get more feedback to those teachers with more frequency and with more quality has vexed us.
In conversation with our partners at Teachstone and AIR, we started to measure how much time it took to complete one observation and give feedback to a teacher. It turned out it takes about 185 minutes. That includes the pre-observation discussion to establish what we’re looking at, the 45 minutes in the classroom, the coding of what you see in the classroom, and the teacher debrief.
It’s tricky to schedule this time with working teachers, and when a school-based coach who also teaches is scheduling observations and feedback, that gets even trickier. While an external coach working with multiple schools can observe and confer with about 10 teachers in a school in a year, a school-based coach can only manage about half that number.
Our goal is to triple that and go from four to five teachers a year to between a dozen and 15. Our goal is to use AI to code the really high-quality interactions happening in the classroom so we are able to show the teacher the interactions that matter without having to sit through the entire lesson and watch every interaction.
This gives us the opportunity to get more feedback to more teachers and to do it with the same number of coaches that we have now.
How does AI help you do that?
We’re going to be able to use AI to clip high-quality interactions and play them back to the teachers to speed the feedback process. We’re using a sports analogy. Instead of playing back the entire basketball game film, I just want to show you the best pick-and-rolls during the game.
Instead of saying to a teacher, this is where you’re doing poorly, we’re going to say, “Hey, how do we increase the frequency of open-ended questions? Look at how students responded here.” You’re looking at specific, high-leverage techniques to improve the quality of instruction in the classroom. You and I would sit together, and I’d say, “Hey, Maureen, we’re going to look at some of your open-ended questions and how students responded.”
And our AI is going to say, “I’ve got you, here are all the open-ended questions that Maureen asked.” So instead of going through the class and having to watch every single thing, what the teacher said, what students said, the AI is going to show purely open-ended questions.
We can see how the quality of your questioning really helped students engage. Instead of having to go through the whole transcript or spend 45 minutes watching the video, the AI will be able to say, here are some examples, so we can elevate them.
Many people are focused on the use of video so that teachers can see themselves, so that we’re seeing the teaching and learning from a student’s point of view. But video takes people, it takes time, it takes effort. With AI, all that’s going to be automatic, the videotaping is going to be automatic, the coding is going to be automatic, so we can focus our time on what really matters, which is the conversation about improving teaching and learning.
Where does this work stand?
We have a prototype that we started putting into our schools in April. By the end of next school year, we will have refined it into a full-use tool. The algorithm has already worked with Teachstone’s Classroom Assessment Scoring System (CLASS).
It knows what to look for in a positive climate. It has a sense of effective instructional dialogue. We’re going to refine its sense of classroom organization.
We’re going to have a researcher watch the video and code interactions according to the principles of good teaching used in CLASS. And then the algorithm is going to watch the coding and try coding videos itself. We’re just going to do that again and again and again, until the correlation between what the algorithm is seeing and what the researchers are seeing is high enough to be confident that the AI can clip high-quality interactions in the same way that a person would—only far faster.
Wow!
We are losing teachers every single day because of the number of tasks that we’ve put on them, right? Urban Assembly is committed to using AI to make teaching easier. Now when I say easier, I don’t mean simpler, because there’s always going to be complexity. But there are ways to reduce the labor that it takes to get feedback to a teacher so they can refine their practice. If there are ways for teachers to be able to share with other teachers what they’re doing effectively through videos, let’s share those things.
What’s next for Urban Assembly and AI?
I’m very invested in personalized learning and understanding how we can organize learning experiences for students that meld whole-group, small-group, and individualized learning in a very flexible way. I’m very invested in this notion of differentiation, but it takes a million hours to plan for 28 different students.
So, there’s a vision of personalized learning where we would have teachers facilitating whole-group instruction and organizing learning experiences in small groups where students are working together. Students who are having challenges would receive artificial-intelligence-based, individualized support and then return to the whole-group or small-group settings. That’s the vision of an integrated classroom, right? Where you have technology and teachers and small group instruction and whole group instruction, all working together.
When they’re taking tests, we’d use competency-based, mastery-based feedback approaches. Students would get feedback on what they don’t know, because we care about what they’re learning, not just giving them a score, right?