��ƷSM����ӰƬ

Sidney D’Mello is a professor in ��ƷSM����ӰƬ’s Institute of Cognitive Science and Department of Computer Science. He is also director of the National Science Foundation AI Institute for Student-AI Teaming, which aims to develop artificial intelligence (AI) technologies to facilitate social and collaborative learning experiences for all students.

Can you talk about your research with AI and education?

There are five flagship NSF research institutes that focus on supporting learning with AI, and we were the first one. Schools right now haven’t changed in a hundred years — they’re focused on efficiency as an outcome. The use of AI in education has been to keep that vision of efficiency going where students individually work with computer programs powered by AI.

The vision I want for classrooms is a place where students are working together, being loud, and it’s a noisy, rambunctious environment where they’re challenging ideas, and they’re building social relationships. This vision is centered around a different perspective of learning, which is that learning is authentic to students’ interests and identities and so on. A key question we ask is how best to integrate AI within that vision.

How could your vision be implemented?

This is a vision that’s been articulated forever in the learning sciences, but it is difficult for teachers to implement because they can’t be listening in on several student groups at once. Some conversations may go off track, or perhaps some conversations are amazing, but the teacher is unaware of the discussion happening. So the question is, how can AI help support this?

Our idea is thinking of AI as this social collaborative partner immersed in these small groups to help them along. The AI is actually interacting with small groups, listening in on the conversations, analyzing nonverbal behaviors like pointing, and then figuring out how to facilitate those small-group conversations, but always by coordinating with the teacher. The teacher remains the centerpiece here. The AI is providing decision support around the teacher to help them orchestrate their classroom as they see best.

How are students responding to your research?

We know that with AI comes great responsibility. We didn’t want to build anything without first working with the students themselves. We organized workshops with them to be transparent, acquire their trust and have their voices heard. However, we soon realized that it was challenging for youth to imagine what good collaboration could look like beyond what they had experienced in school.

So we took a group of students to this cooperative house in Berkeley, Colorado, where they learned how house members had to live together and collaborate outside of schooling. They were introduced to the idea of community agreements, which are mutually agreed upon norms of behavior that the house members themselves co-negotiate and use to hold each other account-able. The youth wondered if an AI could help them to generate and maintain such agreements and developed a design sketch to embody their ideas.

What happened as a result?

This led to one of our AI technologies called the . With the help of the teacher, students work in small groups to input their examples of agreements into the CoBi interface. As students engage in collaborative learning, CoBi analyzes student discourse for evidence, or “noticings,” of these agreement categories. The results are visualized by way of a growing tree animation that everyone can see, where the noticings are shown as flowers that bloom. Teachers and students use these visualizations to reflect and make sense about their adherence to their agreements.

Now this is where privacy becomes really important. By working with students and getting their sense of comfort, we learned they are terrified of any of their individual talk being known to the teacher. So we do not give the teacher any information on who's speaking and saying what. We don’t even say which student group. It’s all aggregated in this class-level interface, which is the tree. 

Are students interacting with the AI interface?

We actually show them what the CoBi is doing, and then the idea is they can correct it if they see something off with its predictions.That gives us good data,and the students can help it improve itself. But more importantly, they can have a conversation about why it’s doing what it’s doing. Because, remember, we also want to teach youth how to learn and collaborate. And so they can have a conversation like, ‘Hey, CoBi, we thought this was an example of being respectful, but you missed it.’

What concerns do you have about ChatGPT right now?

I don’t think we should ban these tools, as that never works. But there's a lot of stuff that still needs to be addressed with these AI programs still. For example, when I’m talking, I’m gesturing and pointing. So I’m making meaning in that context. ChatGPT and all of those other programs are good for their language piece, but they are not grounded in the real world. That’s why we are working with foundational AI to integrate semantics of speech with gesture, gaze and social cues to make an understanding from multimodal, multi-party discourse.

How can teachers quickly be trained in AI?

You can get teachers caught up in AI, and we’re developing curricula for AI literacy for this very purpose. But it’s really a more foundational thing — how can we change learning? How do you design a curriculum that does 21st-century skills, collaboration, critical thinking, inquiry, disciplinary knowledge, experimentation, investigation and developing character at the same time?

You’ve talked a lot about ethical and equitable AI. What does that mean? Right now a phrase that’s used a lot is “responsible AI”. It’s not asking what AI can do, it’s asking what I should do, basically. We have a framework of responsible innovation that we implement in everything we do from the start, and it begins with ourselves — our values, our processes and our commitment to our students and teachers.

  Submit feedback to the editor

Photo by Patrick Campbell