This is an iHeart podcast. grade class somewhere in Michigan.
It was filmed in January of 1990, so the video is a bit grainy.
The teacher's name is Deborah Loewenberg Ball.
She's a professor at Michigan State University, who as part of her research teaches a one -hour math class at a local elementary school.
On the day in question Ms. Ball begins by asking her students about the previous day's lesson, which was about even and odd numbers.
I'd like to hear from as many people as possible what comments you had or reactions you had to being in that meeting yesterday.
A little boy with black hair raises his hand.
His name is Sean. Um, I don't have anything about the meeting yesterday but I was just thinking about 6.
Sean was thinking about the number 6.
I was thinking that it can be an odd number too.
Because there could be two, two -four -six, three -twos.
Now, it could be six and two -threes.
It could be an odd and an easy number.
And four. Three things to make it?
There could be two things to make it.
Sean doesn't understand what odd and even means.
He thinks that just because you can break down six in an odd number of parts, and an even number of parts, that six must exist in some magical middle category.
And when you listen to the Sean videotape, you keep waiting for the teacher to say, Oh no, Sean, you misunderstand.
But Debra Ball doesn't do that.
She never tells him he's wrong.
Instead, she simply asks him to explain his thinking.
And the two things you put together to make it were odd, right?
Three and three are each odd?
Anything bottom, bottom, top, bottom two were even.
Baldwin asks the class to give their views.
Other students jump up and explain their theories on the blackboard. For the next fifteen minutes, she deftly guides the class through an in -depth investigation of what she calls Sean numbers.
Until Sean himself realizes that the real meaning of odd and even is something different than he had imagined.
And now, he gets it.
I don't want to focus just on how little Sean finally made his own way to the right answer.
I'm interested in what his teacher did to get him there.
Debra Ball worked magic.
She never told Sean the right answer.
She just led him to a place where he could discover it for himself.
My name is Malcolm Globwell.
This is season 6 of SmartTalks with IBM, where we offer our listeners a glimpse behind the curtain of the world of technology and artificial intelligence.
In this season, we're going to visit companies as varied as L 'Oréal and Ferrari and tell stories of how they're using artificial intelligence and data to transform the way they do business.
This episode is about the promise of a radical new idea called responsive teaching, the kind of teaching that took place that day in Sean's classroom, and whether artificial intelligence can help us train the next generation of teachers to be as good as Deborah Ball.
Before we talk about how AI could transform the way we train teachers, I want to go back for a moment to the famous video of Sean.
In the video, the teacher, Deborah Ball, doesn't have a predetermined plan that she's imposing on the class.
She's improvising, making up her approach as she goes along, responding to her student's odd theory about the number 6.
Second, she's taking Sean seriously.
She's not dismissing his theory.
She's listening to him and trying to understand the problem from his perspective.
And thirdly, and most importantly, she's not force feeding him the right answer.
She's being patient.
She's waiting to see if, with just the right subtle hints, he can get to the right answer on his own.
Improvisation, empathy, patience.
That's responsive teaching.
What I think about in terms of responsiveness is more like, I think that students need to have a sense of agency in what happens in the classroom and, like, authentic agency where they can be legitimized as knowers.
I spoke to a physicist at Seattle Pacific University named Amy Robertson, a longtime advocate for responsive teaching.
She uses the Sean video in her classroom.
You have to trust that kids have a way of doing that.
And what she mostly did was to facilitate a conversation and to say, you have to listen did them talk!
No one told him he was wrong.
That's right. And then he goes, he goes, I didn't think of it that way again.
I've been thinking that way, thank you for bringing it up.
You've expanded my understanding.
Thank you for bringing it up again!
It's like this kid, I love this kid!
I know, responsive teaching as I think about it is kind of rooted in this, um, like Eleanor Duckworth's work around the having of wonderful ideas where she says, Like the goal of education is for students to have wonderful ideas and to have a good time having them.
What a way, I love that.
I've never heard that.
What a beautiful succinct way of summing up the purpose of education.
Responsive teaching is beautiful.
It's rare to find a new teaching idea that everyone loves.
This is one of those rare ideas.
Watching the Deborah Ball classroom, all I could think was, I really, really hope my daughters get to experience a math class like that.
Far too many kids are convincing themselves at far too young an age that math isn't for them, and responsive teaching is a way to solve that problem.
But here is the issue.
It's really, really hard to teach responsive teaching.
Robertson says that teaching exists in a cultural environment where the teacher is expected to be the source of truth, that teaching is about the immediate correction of error and not letting a wander down the pathway of their own misunderstanding.
Responsive teaching is deeply counter -intuitive and the only way to understand its beauty is to do it over and over again.
Aspiring teachers need a way to practice.
For as long as there has been technology, people have turned to digital machines to solve problems. My father was a mathematician and I remember him coming home in the 1970s with a big stack of computer cards in his briefcase that he used to program the mainframe back to the office.
Today with the rise of artificial intelligence, the scale and complexity of the problems technology can help us solve, has jumped by many orders of magnitude.
You must have worked with a million customers who are experimenting with LLMs. Has there been one use case that you were like, whoa, I had no idea, or just simply that's clever?
I'm speaking to Brian Bissell, who works out of IBM's Manhattan office.
He helps IBM customers discover how best to get AI to work for them.
There is one, but I don't think I can talk about it, unfortunately.
Wait, wait, wait. You can't tease me like that.
Can you disguise it for me?
Just give me a general - It was about the ability to pull certain types of information out of documents that you wouldn't think you would be able to get the model to do and be able to do that at a very large scale.
Bissell's point was that we are well past the stage where anyone wonders whether AI can be useful.
The real question now is what problems do we want to use it to solve, where it can make the biggest difference.
And Bissell saw lots of opportunities in education.
I have two kids, one in middle school and one who just graduated high school.
And I'm well aware of students using things like ChatGPT to do their homework and it's very easy to take tools like that, and even IBM's own large language models, and just take a problem, a piece of homework, something you want written, and drop it into that and have it generate the answer for you.
And the student, the user in that case, hasn't done any work.
They haven't put any real thought into it.
To Bissell, that's the wrong use of AI.
That's technology making us dumber.
What we really want is technology that makes us smarter.
Bissell explained to me that there are now two big tools being used for AI productivity.
AI agents and AI assistants.
Let's start with AI agents.
AI agents can reason, plan and collaborate with other AI tools to autonomously perform tasks for a user.
Bissell gave me an example of how college freshmen might use an AI agent.
as a new student, you may not know, how do I do with my health and wellness issue?
How many credits am I gonna get for this given class?
You could talk to someone and find out some of that, but maybe it's a little bit sensitive and you don't wanna do that.
Bissell told me you could build an AI agent, a resource for new students that helps them navigate a new campus, register for classes, access the services they need, and even schedule appointments on their behalf, which in turn buys them more time to focus on their actual schoolwork.
We can see patterns of how agents and assistants can help employees and customers and end users be more productive, automate workflows so they're not doing certain types of repetitive work over and over again, and streamlining their lives and making data more accessible to them 24 hours a day.
But Bissell says you can also use AI assistants in the education space.
AI assistants are reactive, as opposed to AI agents, which are proactive.
AI assistants only perform tasks at your request. They're programmed to answer your questions.
And as it turns out, AI assistants are now being used to further the responsive teaching revolution, which is why I found myself on a beautiful Georgia spring day, not long ago on the campus of Kennesaw State University.
Sitting in the classroom with two researchers, one of them, Professor Da -be Lee.
Let's go into the journey of building this thing.
You started Da -be by taking a course.
What was the course you took?
Yeah, so it was offered by Coursera.
It was designed by IBM.
It was AI Foundation for everyone.
In her AI Foundations course, Lee learned how to build an AI assistant using IBM WatsonX.
That course took how long to take?
It was not too long.
It was, like, 14 weeks.
Lee's idea was to train an AI assistant on classroom data to play the role of Sean, a digital persona of a nine -year -old who likes math but doesn't always understand math.
And that AI assistant, she thought, could be used to train preservice teachers or teachers in training, Who are preparing to enter one of the most challenging professions in the modern world.
So when you think about the teacher education and the major challenge that teacher education face is that we need children to practice with.
We need instructors who will give the instruction on the pedagogical skills.
So when you look at the Teacher Education Program, we have coursework and field experience.
And in those two areas, there's something missing all the time.
Lee says that pre -service teachers often lack access to both students and experienced teachers during their education.
So, what we try to resolve is that we have this virtual student for pre -service teacher to work with so that they can practice their responsive teaching skills.
The first AI assistant Lee created is Ji Woo.
Ji Woo emulates the persona of a nine -year -old third grade girl.
Then, with the help of one of her collaborators – a researcher at Kenazan named Shaun English – she created two more AI assistants, Gabriel and Noah, each of which have their own distinctive characteristics.
So, how are Gabriel and Noah different from Ji Woo?
Gabriel, my first one, he's very short answer.
If you ask open -ended question, he will answer it in a close way.
So I use that characteristic.
And that's the problem that most teachers actually face.
There are children who are shy, who are rich, and who would not share much of their thoughts.
So we wanted that characteristic in some characters and we used Gabrielle to have that characteristic.
And Noah, what's Noah's personality?
Happy, playful, cheery, bright and energetic.
That's Sean English, Professor Lee's fellow researcher.
And Jiwoo? Jiwoo is articulated and kind of smart.
But she has her own way of thinking.
I would end up spending a lot of time with Jiwoo.
She's something of a character.
I asked Sean about the process of creating these AI assistants.
What does building the content side of the AI assistant entail?
Sean, what do you think?
It sets up those series of actions, effectively, which are response cases, you can kind of think of them as.
You have a series of questions that you tie to an intent, and then that intent has reactions from the bot.
And so effectively, if we were looking to say, make a hello action, we would have all the different ways that people could say, hello, hello, what's up, how are you doing and all that kind of stuff.
Shaun says the longer the list of potential responses, the better.
But AI responses don't just follow the list. The AI assistant uses those suggested responses to come up with a universe of other responses.
And in that process, sometimes it comes up with things that just don't make sense.
And from a technological standpoint, While AI is a fantastic tool, AI can hallucinate which means just give things that it's just straight up made up.
There is a famous example of this called the Three Rs is where you ask a popular large language model how many Rs are in strawberry and it gives you a wrong answer and he repeats that result repetitively.
You always want to have a human interacting with the system to be able to go, Hey, that's a little crazy.
I don't think that's exactly what we're going for here.
That's why it's good to have someone like Shaun English around to step in and get the model back on track.
and over time, when the model has enough training, it's ready for the teachers -in -training.
One of the rollouts of Jiwoo, Gabriel, and Noah was with the teacher -training program at the University of Missouri.
I was just kind of excited to see what the program was and what it was gonna be doing.
This is Logan Hovis, a junior at Missouri, on the path to becoming an elementary school teacher.
Obviously a little skeptical when he said it was supposed to, You know, be like talking to a student, you're like, there's no way this AI thing is going to totally sound like a second grader or a third grader.
Like it's going to sound like an adult or it's going to sound like a robot that knows all the answers, and it really didn't.
It really was like talking to a child.
It was very, very well developed in the way that you really sit there and you feel like you're talking to a kid.
Her point wasn't that Jiwoo and her fellow avatars were equivalent to real kids.
Of course not. But for someone starting out, someone who was already nervous about being plunged into a classroom of nine -year -olds, Ji -woo was like a warm up before a baseball game.
What I can think of is like you know, how when you're at batting practice for baseball or softball, you have those automatic pitchers that throw them because you're working on your skill as the hitter.
What can I do differently?
What am I doing wrong?
But that doesn't replace the game and what you do in a game, but this is you getting to practice your own skills to be better when you go in a game.
And I think that's kind of what the AI software feels like for us.
In batting practice, the pitches don't come as hard and fast as the pitch is in a real game, but you get to stand at the plate and the pitcher throws you dozens of balls over and over again in a concentrated block that allows you to work on your swing closely and carefully.
There's a lot less stimulus going on around because the classroom is very, very busy.
It's wonderful, it's beautiful, but it's very, very busy.
So sometimes it's hard to keep that focus in on the tasks that they're doing at hand.
And also, in the teacher setting, you're also always looking around, making sure that other students are doing what they're supposed to be doing, but also if they need any help, if everything's going okay in the classroom.
So being on the GWU chat, it was just nice that you didn't have to do any of the extra work to keep the focus on there.
And it also felt You didn't have to feel the students' nervousness of being one -on -one with you.
And also as a teacher, it was a lot less pressure too because I was like, okay, I'm taking this serious.
This is a student I'm questioning, but I also know I'm probably not going to hurt someone's feelings right now and that's terrifying to think I'm going to ask the wrong question and upset the child because I've done that.
We think of the typical use of AI as a tool for speeding things up.
That's what we always hear, that the introduction of AI to problem x gave an answer in minutes when solving problem x used to take weeks.
But we shouldn't forget another use, that it allows us to slow things down.
Hovis if she wanted to could spend a whole weekend practicing with Jiwoo.
A real nine -year -old will get frustrated and bored with the fumbling novice after 10 minutes, but Jiwoo?
Jiwoo will happily answer questions for as long as it takes for the people who want to learn to be responsive to learn how to be responsive.
At the end of my time at Kennesaw State Sean and Dabe led me to a small table where Dabe had set up her laptop.
In the corner of the screen was a chat box of the sort we've all seen and used a thousand times.
Jiwoo began. She had been given a math problem.
When Matthew was making play -doh, He added three -fourths cups of flour to the bowl.
Then he added another three -sixths cups.
It's the total amount of flour he used greater than or less than one cup.
How much flour did he use?
That's a simulation of Gee -Woo speaking.
We pause it for a second.
So Ji -woo is trying to solve this problem.
And the first thing she does is she draws a rectangle on the screen.
This is a common tactic of nine -year -olds.
Try to visualize the fractions.
And she divides it into 4 pieces.
And now she's going to color in 3 of the 4 pieces.
So she's representing, this is quite good, she's representing 3 quarters on the screen, okay?
Yeah. And this is a three sixes.
So now Jiwoo does another rectangle with six boxes and colors in three of them.
Okay. And when you add them together, that makes six out of ten.
So then she counts up all the colored boxes and that's her numerator.
And counts up the total number of boxes and that's her denominator Jiwoo had counted the colored boxes and landed on an answer when you add three quarters of a cup and 3 sixths of a cup you get six tenths of a cup So according to Jiwoo Martin has less than one cup and she thinks she solved the problem.
Yes Okay, so it's less than one cup.
Yeah, so she says it's less than one cup Now oh my god, this is hard So the question is, what do I, as a teacher, say to Jiwoo?
We were off. The rules were simple.
I couldn't give Jiwoo the answer or explain to her what she was doing wrong.
I had to be Deborah Ball.
I had to help her find the way herself.
The chat box in the corner of the screen was waiting for my first question.
I thought for a moment and started typing.
Do you think the boxes in the red rectangle are the same size as the boxes in the blue rectangle.
Then I turned to Sean and Daabae.
Is that a good question?
Seriously? Yeah that's a good question.
Jiwoo doesn't mess around.
She answers immediately.
So she says the blue and red pieces are not the same sizes.
Oh. So you understand now Jiwoo knows the size differences.
So she's pretty smart here.
Yeah. Then I asked, if they are not the same size do you think you can add together.
Ji -woo answered right away.
Ji -woo says, I have learned that I could add any numbers in grade two so three plus three is six and four to six is ten.
Yeah, so she's using the knowledge of adding integrals into adding fractions.
Now I'm stumped. So now I have to somehow lead her to figure out a way to get her to understand that we're dealing with a different kind of problem.
That's a harder problem.
Amy Robertson had told me that learning how to do responsive teaching properly was really hard. And now I understood why.
I had to put my mind inside the mind of a nine -year -old.
I had to internalize her knowledge base and assumptions, and keep in mind, I haven't been nine for a very long time.
I honestly had no idea what to say next.
I thought for a moment.
I asked what I quickly realized was a hopelessly convoluted question.
Daabae and Sean had built a mentor into the system, an experienced, responsive teacher who supervises the session and offers advice.
My mentor noticed that I was struggling.
Told me to simplify my question.
Remember, she's a third -grader.
Daabae was trying to help me too.
She suggested, why not just ask Ji -woo if three -quarters is bigger or smaller than one -half?
So we are trying to help her to think about fraction in a more conceptual way.
This time, Jiwoo understood.
She wrote back, three -quarters is larger than one half.
I wrote back, is three -sixths of a cup bigger or smaller than one half?
Jiwoo said, I'm confused.
Oh no! I've confused Hwan Jiwoo!
But it's good. She's understanding, she is realizing her misconception.
So she's getting confused.
She says, I'm confused, three -quarters is pretty close to one, and adding three -sixths would make it go over one.
Oh, so she's got the answer.
Yeah. But then she says, But there are six pieces out of ten, which is less than one, so I don't get it.
So, she's the point that, oh, this, I have something wrong here.
That's a good sign.
She's getting there.
Yeah, she's getting there.
But I still have to get her to… She has to get the six pieces out of ten out of her head.
Yeah. I have no idea how to do that.
What? And she thinks she's confused when she has actually...
She's figured out the answer.
Yeah, she did. So we have advanced, even in my stumbling and bumbling, we've made some progress.
We have very notable progress here.
My conversation with Jiwoo went on for some time, and eventually I got there.
Jiwoo found her way to the right answer.
She said, I have more than one cup of flour.
The mentor chimed in, I got a little emoji that made me feel good.
And when it was over, I realized two things.
The first was, I needed more batting practice, much more.
And that batting practice was really, really easy to do because someone has gone to the trouble of building me my very own baseball diamond and given me a pitcher who would throw me baseballs all day long.
My second thought was that I'd been thinking about AI all wrong.
I have interpreted a lot of the talk about the promise of AI to be about replacing human expertise.
I had actually thought when I first heard about Dabe's project that that's what Dabe and Sean were doing, creating an AI to teach students by passing the teacher all together.
But if you did it that way you would miss the magic of the classroom.
Remember Eleanor Duckworth's quote, the goal of education is for students to have wonderful ideas and have a good time having them.
I think we often focus on the first part of that formulation, the wonderful ideas, but neglect the second, the good time having them.
Real learning is born in pleasure, in community, in playful discussion.
In a group of kids coming together to solve a problem.
And all of that magic only comes from human interaction, from a teacher who is skilled enough to inspire a class of 9 -year -olds.
We don't want AI assistance to replace the teacher.
We want AI assistance to help teachers turn themselves into even better teachers.
Smart Talks with IBM is produced by Matt Romano, Amy Gaines McQuaid, Lucy Sullivan, and Jake Harper.
We're edited by Lacy Roberts.
Engineering by Nina Bird Lawrence, Mastering by Sarah Bruguier, music by Gramascope.
Special thanks to Tatiana Lieberman and Cassidy Meyer.
Smart Talks with IBM is a production of Pushkin Industries and Ruby Studio at iHeartMedia.
To find more Pushkin podcasts, listen on the iHeartRadio app, Apple podcasts, or wherever you get your podcasts.
I'm Malcolm Glapa. This is a paid advertisement from IBM.
The conversations on this podcast don't necessarily represent IBM's positions, strategies, or opinions.
