Enhancing Student and Teacher Agency in Knowledge Building with AI

Bodong Chen
cbd@upenn.edu • bchen.net

Associate Professor, University of Pennsylvania

Knowledge Building

A constellation of perspectives involving students working as a community to continually improve ideas

  • Agents with epistemic agency
  • Artifacts created & refined
  • Practices for idea improvement

[30-45 seconds]

I want to start by grounding us in Knowledge Building. At its core, Knowledge Building is about students working together as a community to continually improve ideas.

There are three key elements in KB: Agents who have real epistemic agency over their learning, Artifacts that capture and evolve ideas, and Practices that support idea improvement in a community.

(Chan & Aalst, 2018; Chen & Hong, 2016; Scardamalia & Bereiter, 2022)

Technology for Knowledge Building

Knowledge Forum
Digital environment for continual idea improvement

KB Wall
Physical manifestation of collaborative knowledge work

[20-30 seconds]

“Technology has been an integral part of Knowledge Building. On the left is Knowledge Forum—a digital environment designed specifically for Knowledge Building. Students post notes, build on each other’s ideas, and create knowledge artifacts together. On the right is a KB Wall, a physical space achieving the same purpose – making ideas visible and to be continually improved.

These have been our tools for decades. But now AI has entered the picture. The question is: How do we integrate it thoughtfully?

Essential KB Principles

  1. Epistemic Agency: Students have high-level agency over what they want to know and how they arrive at knowledge

  2. Knowledge-Creating Dialogue: Carried out through dialogues that differ from teacher-centered discourse

  3. Transformative Assessment: Assess state of the art and feed-forward to potential progress—like people working at the knowledge frontier

How can AI uphold these principles rather than undermine them?

[60-75 seconds]

Answering this question requires us to consider essential Knowledge Building principles.

First, Epistemic Agency. This means students have high-level agency over what they want to know and how they arrive at knowledge.

Second, Knowledge-Creating Dialogue – dialogue that seeks to create and improve ideas, not just answering questions.

Third, Transformative Assessment. We assess not just to grade, but to understand the state of the art and feed forward to potential progress—much like people working at the knowledge frontier.

So a central question for us is: How can AI uphold these principles rather than undermine them? This is what I want to explore with you today.

(Scardamalia, 2002)

Two Examples: AI Supporting Agency

1: Integrate AI in student dialogues
to support student epistemic agency

2: CraftPad for teacher design
to support teacher agency in design

[20-30 seconds]

Here, I’ll briefly share two examples – one about integrating AI into student dialogues in a Knowledge Building classroom—supporting student epistemic agency. The second is about a tool called CraftPad my lab created to help teachers design Knowledge Building lessons with AI.

In each case, I’ll briefly walk you through our goals, our design, and what we learned.

Case 1: AI in Student KB Dialogues

Context

  • High school World Religions course
  • Taught by one experienced KB teacher
  • Multiple digital tools (KF, Miro, ChatGPT)

Design Question How to distribute agency between students and AI?

Co-Design with Teacher

Situate AI in Knowledge-Creating Dialogue Moves:

  • Problem definition
  • New ideas
  • Promisingness evaluation
  • Meta-dialogue
  • Comparison
  • Critical discourse
  • Higher-level ideas

[60-75 seconds]

Let me start with students. The context here was a high school World Religions course taught by an experienced Knowledge Building teacher in Colombia. Students were exploring religious literacy using multiple tools: Knowledge Forum, Miro for visual organization, and ChatGPT.

The central design question we faced was: How do we distribute agency between students and AI? We didn’t want AI to replace student thinking.

One important strategy we took was to – work with the teacher to carefully situate AI within what we call Knowledge-Creating Dialogue Moves.

These moves are defined by Bereiter and Scardamalia. You see these listed here—problem definition, generating new ideas and assessing their promisingness, comparison and critical discourse, meta-dialogue, and higher-level synthesis. These are the moves that drive knowledge creation. Our design approach was to position AI as a participant in these moves, but always with the students maintaining control.

(Bereiter & Scardamalia, 2017)

Two Patterns of ChatGPT Integration

Pattern 1: Explore Problem Space

Pattern 2: Build Through Discourse

[45-60 seconds]

Based on this idea, we came up with two main patterns, shown in this figure.

Pattern 1, on the left, is about exploring the problem space. Students would use ChatGPT early in their inquiry to generate initial questions, uncover complexity they hadn’t considered, and identify gaps in their own knowledge. It was like using AI as a thinking partner to open up the problem.

Pattern 2, on the right, is about building through discourse. Here, students used ChatGPT later in their work to generate alternative perspectives that challenged their emerging ideas, to push back on their thinking, and to deepen the collaborative dialogue they were having with each other.

I won’t have time to unpack details. The key thing to notice is that in both patterns, students are in the driver’s seat. They’re using AI strategically, at different points in their inquiry, to support their knowledge-building work.”

Key Finding: AI Made Learning Harder and Deeper

Students Demonstrated AI Literacy

  • Understood mechanisms & limitations
  • Recognized need for fact-checking
  • Saw risks of over-dependence
  • Maintained critical stance

Students Took High-Level Agency

  • Used AI for inspiration, not answers
  • Engaged in prompt design
  • Collaborated more, not less, with peers

🦉 By maintaining epistemic agency, students learned about AI, and learned how to learn with AI

[90-120 seconds - THIS IS YOUR PUNCHLINE]

From data analysis, the most interesting finding from the students was: AI made learning harder, not easier.

When we analyzed student reflections and their work, we found that students making efforts to develop AI literacy. They understood the mechanisms and limitations of large language models. They recognized the need for fact-checking. They saw the risks of over-dependence. They need to work on maintaining a critical stance when using AI.

In the process, students took high agency over their learning. They used AI for inspiration, not answers. They engaged deeply in prompt design – learning how to ask better questions. Also crucially, they collaborated more with their peers, not less. AI didn’t isolate them—it gave them more to talk about with each other.

[Pause for fragment to appear]

In this case…

(Tan et al., 2025)

Case 2: CraftPad for Teacher Design

Goals

Support teachers as designers with professional agency

  • Flexible workspace with pedagogical scaffolding

[60-75 seconds]

Let me now shift to Case 2, which focuses on teachers. This is CraftPad—a tool my lab has been developing to support teachers as designers with professional agency.

We want teachers to be designers of Knowledge Building experiences, not just consumers of AI-generated lesson plans. So CraftPad is designed with three key features: 1) a flexible workspace with pedagogical scaffolding built in…

Case 2: CraftPad for Teacher Design

Goals

Support teachers as designers with professional agency

  • Flexible workspace with pedagogical scaffolding
  • Context-aware KB Coach

  1. a context-aware KB Coach that understands Knowledge Building principles

Case 2: CraftPad for Teacher Design

Goals

Support teachers as designers with professional agency

  • Flexible workspace with pedagogical scaffolding
  • Context-aware KB Coach
  • Rich opportunities for teacher decisions

  1. most importantly, rich opportunities for teachers to make decisions throughout the design process.

CraftPad: Preliminary Findings

Design Activity (from 52 users)

  • Shows deep iterative refinement
  • Some with up to 96 revisions

KB Coach Conversations

  • Most were consultations focused on refining lesson designs
  • Some focused on certain KB principles
  • A few went deep (e.g., 38 messages exploring SDG connections in a photosynthesis unit)

Teachers are designing with AI, not just consuming AI-generated content

[90-105 seconds]

“We’re still early in this work, but the preliminary findings are encouraging.

On the left, you see data from 52 teachers who used CraftPad to design lessons. What struck us was the depth of iterative refinement. Many teachers went beyond the initial version. Some created up to 96 versions of a single lesson! This isn’t AI spitting out one lesson plan that teachers just accept. This is teachers designing, revising, reconsidering, refining—engaging in iterative design work.

On the right, you see data from conversations between teachers and the KB Coach. Most conversations were consultations focused on refining lesson designs—short, focused interactions. Some zoomed in on specific KB principles. But a few went really deep—one teacher had a 38-message conversation exploring how to connect photosynthesis to Sustainable Development Goals.

The key point here is: Teachers are designing WITH AI, not just consuming AI-generated content. The metrics reveal agency in action—iteration, decision-making, sustained engagement with ideas. Just like with students in Case 1, AI is a tool that supports human agency rather than replacing it.”

Integrate AI in KB Classrooms

Students + AI

  • AI aligned with dialogue moves
  • Students maintained epistemic agency
  • Learning became harder (deeper)
  • More collaboration, not less

Teachers + AI

  • AI positioned as design coach
  • Teachers maintained professional agency
  • Design became more iterative
  • More reflection, not less

Design for high-level human agency. AI as flashlight, not magic.

[60-90 seconds - YOUR SYNTHESIS]

“Looking across these two cases, we can see that:

With students and AI: We positioned AI within dialogue moves. Students maintained epistemic agency. Learning became harder and deeper. And students collaborated more with each other, not less.

With teachers and AI: We positioned AI as a design assistant. Teachers maintained professional agency. Design is iterative. And teachers engaged in more reflection, not less.

To conclude: I encourage us to expand our thinking of AI in education to go beyond automation or efficiency. We should design for high-level human agency and position AI carefully so that AI enhances human capability rather than replacing it.

Thank you!

Bodong Chen cbd@upenn.edu
https://penn-wonderlab.github.io/

Acknowledgement Jiayu Cheng, Vivian Leung, Yue Zhao (University of Pennsylvania); Xinran Zhu (UIUC); Fernando Díaz del Castillo (Mentu/Gimnasio La Montaña)

[10-15 seconds]

“Thank you for your attention. I’m happy to take questions, and if you’re interested in learning more about this work, please feel free to reach out.”

[Leave slide up for questions. If questions touch on appendix material, advance to relevant appendix slides.]

AI for Education • 11/20/2025 • Singapore

1 / 14
Enhancing Student and Teacher Agency in Knowledge Building with AI Bodong Chen cbd@upenn.edu • bchen.net Associate Professor, University of Pennsylvania

  1. Slides

  2. Tools

  3. Close
  • Enhancing Student and Teacher Agency in Knowledge Building with AI
  • Knowledge Building
  • Technology for Knowledge Building
  • Essential KB Principles
  • Two Examples: AI Supporting Agency
  • Case 1: AI in Student KB Dialogues
  • Two Patterns of ChatGPT Integration
  • Key Finding: AI Made Learning Harder and Deeper
  • Case 2: CraftPad for Teacher Design
  • Case 2: CraftPad for Teacher Design
  • Case 2: CraftPad for Teacher Design
  • CraftPad: Preliminary Findings
  • Integrate AI in KB Classrooms
  • Thank you!
  • f Fullscreen
  • s Speaker View
  • o Slide Overview
  • e PDF Export Mode
  • r Scroll View Mode
  • ? Keyboard Help