Who’s Doing the Thinking?

Introductions Hi, I’m Bodong

Bodong Chen

Associate Professor, Learning Sciences & Technologies, Penn GSE

Wonder Lab

I study how people build knowledge together, and I build tools that keep humans in charge of that work.

InkSpire Reading scaffolds for instructors

WonderBits AI agents propose, students decide

DataX Data investigations of justice issues

“I’m less interested in whether students use AI than in who is doing the thinking, and who gets to shape the environment.”

You, right now Most of you wear three hats at once

Learner

Taking courses, reading, writing

What am I trying to learn here?

TA / Instructor

Leading sections, designing tasks, giving feedback

What am I trying to teach here?

Researcher

Collecting and analyzing data, building an argument

What counts as my judgment here?

01 Where we are How AI has been implemented and experienced by students, educators, and researchers

Where we are Adoption came bottom-up and uneven

1

Students moved first

Often privately, with some anxiety about whether it’s allowed. Real help getting unstuck, and a pull toward offloading the part that was the learning.

2

Institutions followed with policy

Detection and bans came first. Course-by-course rules still leave students guessing. Transparency norms help more than bans.

3

Pedagogy is still catching up

Shifting toward “purposeful engagement,” but unevenly. AI on the teaching side (prep, scaffolds, feedback) gets far less attention.

Where we are For researchers, AI is both topic and tool

Researcher

What’s happening

Coding qualitative data, literature work, discourse analysis
In my lab: LLMs to evaluate the promisingness of ideas in knowledge-building discourse

Still open

What isn’t settled

Validity and reproducibility
What counts as the researcher’s judgment
Norms for AI in dissertation work are still forming, and advisors vary a lot

Most of what’s happened so far is AI being dropped into existing practices. The more interesting question is what practices we design around it.

02 Same principles, different outcomes Where the benefits are real: when AI fits what the discipline does and what the course is trying to teach

My groundings Two principles I start from

Education as life itself

“Education is not preparation for life; education is life itself.”

John Dewey

Authentic disciplinary practice

What do professionals in your field actually do?

How are those practices changing?

Let students practice what their field is becoming, not an outdated version of it. That includes research.

Two of my courses Same principles, different design outcomes

Objectives changed

Programming Fundamentals

Before Learn Python syntax and concepts With AI Work with GitHub Copilot to solve authentic problems

Copilot explains and generates code. The focus shifts to problem-solving and reading code critically.

Objectives unchanged

Learning Theories

Goal Understand foundational theories and apply them to real problems With AI ChatLab helps students connect theories to their own interests

AI supports the goal instead of replacing it. Engagement with hard concepts goes deeper.

Both sides of the classroom AI for learners, and for teachers

Learner

ChatLab

Students pick theorists to “join” a conversation about a problem they care about.

TA / Instructor

InkSpire

Generates reading scaffolds tied to specific learning goals and disciplinary practices. Something I value but rarely had time to do well for every reading.

Making thinking visible AI proposes. Students decide.

In WonderBits, a knowledge-building space, AI assistants suggest connections between ideas. Students accept or reject them.

Each decision becomes a record of student judgment that teachers and researchers can look at.

TA / Instructor Researcher

Student idea Groups that build on each other’s ideas seem to trust each other more.

AI assistant · suggestion This looks related to the note on shared roles. Link them?

✓ Link ✕ Not quite

→ “Not quite: that note is about roles, mine is about trust.”

Illustrative sketch

Where the benefits are real Five benefits, when purpose comes first

Learner

Access to authentic practice

Professionals in many fields already work with AI. Practice the field as it is becoming.

Learner

Personal connection to hard ideas

Link abstract theory to what you care about.

TA / Instructor

Scaffolding at scale

Goal-aligned support for every reading, not just some.

Instructor Researcher

Making thinking visible

Accept/reject decisions leave a record of judgment.

Everyone

Students as builders, not just users

It’s now realistic for students and teachers to prototype their own tools. That changes who gets to shape learning environments.

03 Tensions The central tension is about agency

Who is doing the thinking?

Who designs the environment where thinking happens?

Tensions What pulls against what

Offloading Learning If the struggle was the point, removing it removes the learning. Which struggle matters is a case-by-case call.

Efficiency Epistemic agency Tools optimize for fast answers. Learning often needs slow questions.

Surveillance Trust Detectors are unreliable. Policing pushes use underground.

Homogenization Diversity of ideas Same model, similar ideas. Idea diversity is a resource in knowledge building.

Speed Rigor For researchers: what does “your own work” mean in a dissertation?

Tensions Two structural questions

Equity

Who can pay, and who knows how?

Access to paid tools is unequal. So is know-how about using them well. Name it. Where possible, use tools the institution provides.

Design

Who builds the tools?

Most AI in schools is built by companies with their own goals. Educators and learners are rarely at the design table.

The question isn’t AI or no AI. It’s architecture: we are always building environments for thinking, and AI is now part of the material.

04 What to do Be purposeful, be transparent, and put yourself in the designer’s seat

For everyone Three habits

1

Start with the goal, not the tool

Ask “what am I trying to learn or teach here?” before “which AI should I use?”

2

Be transparent

Say when and how you used AI. Transparency builds trust; policing drives use underground.

3

Keep the judgment calls

Let AI propose; you decide. Practice rejecting outputs, and saying why.

Transparency in practice What it looks like on the page

From my syllabus

Generative AI tools can enhance your learning when used thoughtfully. AI should supplement, not substitute for, your own critical thinking.

If you use one of these tools you will not be penalized. Instead, acknowledge when and where you used it.

For your research: keep a short AI log

When	Task	What AI did	My call
Wk 2	Literature search	Suggested search terms	Kept two; one was off-topic
Wk 4	Interview coding	Proposed initial codes	Merged and renamed; memo on why
Wk 6	Chapter 2 draft	Flagged unclear transitions	Rewrote them myself

It will help you answer methods questions later, while norms keep shifting.

By role Wearing each hat

As a learner

Use AI to get unstuck, to argue with you, and to test your understanding
Be careful using it to produce the thing you’re supposed to learn to produce

As a researcher

Keep a log of how you use AI
Talk with your advisor early about expectations
Treat validity and reproducibility as open questions, because they are

As a TA / instructor

Revisit objectives: some should change, some shouldn’t. Decide on purpose.
Model it: show how you use AI, and where you don’t
Different courses need different approaches

Build, don’t just adopt.

With today’s tools, a doctoral student can prototype a small learning tool in a weekend. That’s a research opportunity and a teaching one.

↻ Roundtable 30 minutes: talk, see, try, commit

Roundtable · 30 min How we’ll spend the time

5

Quick round

Your hat right now. One use, one worry.

10

Show and tell

A quick look at one or two tools, with a screenshot backup if the wifi doesn’t cooperate.

10

Try a prompt together

Pick one and run it live on something someone brings.

5

Wrap

One change you’ll try this semester.

Quick round

Which hat are you wearing right now?

One way you use AI. One worry.

Learner TA / Instructor Researcher

Show and tell Tools on the table

TA / Instructor

InkSpire

Reading scaffolds tied to learning goals. Good for preparing discussion.

Learner Researcher

WonderBits

Knowledge-building canvas. AI proposes, students accept or reject.

Learner

ChatLab

Connect learning theories to your own interests.

Course design

Copilot in intro programming

Changing learning objectives on purpose.

Not AI

Hypothesis

Annotate readings together, then compare your annotations with an AI’s reading.

General

Claude, ChatGPT, Gemini

Set up with custom instructions as a course tutor or writing partner.

Try a prompt Prompts that keep you thinking: learning

Reading partner Learner

I’m reading paper. Don’t summarize it. Ask me three questions that test whether I understood the main argument, one at a time, and push back on my answers.

Theory connector Learner

I’m learning about theory. I’m interested in personal interest. Help me find one connection between them, then ask me to find a second one myself.

AI use disclosure Learner

Help me write a short, honest note describing how I used AI on this assignment: describe. Keep it factual.

Try a prompt Prompts that keep you thinking: research and teaching

Devil’s advocate Researcher

Here is my research question and design: paste. Act as a skeptical committee member. Give me the three strongest objections, and don’t soften them.

Scaffold builder TA / Instructor

I’m teaching topic to audience. The learning goal is goal. Draft three discussion questions that move from comprehension to application to critique. Explain what each is meant to surface.

Assignment stress-test TA / Instructor

Here is my assignment: paste. How could a student complete this with AI without learning goal? Suggest two changes that keep the goal while making AI use more purposeful.

If things go quiet Questions to sit with

Which part of your own learning would you not want AI to do for you? Why?

Should your field’s doctoral training change because of AI? What should stay the same?

If you could design one AI tool for your students, what would it do, and what would it refuse to do?

Treat AI as part of the environment you design for thinking, and make sure you, and your students, stay the designers.

cbd@upenn.edu · bchen.net · Wonder Lab

Backup Quick answers

Should I use AI to write my dissertation? Use it to think with, not to think for you. Be transparent, check program and advisor norms, keep a log. Your contribution is your judgment and argument.

Is AI detection reliable? No. It produces false positives and erodes trust. Design assignments and norms instead.

How do I write a course AI policy as a TA? Align with the instructor, then state the purpose, what’s allowed, and how to disclose. Short and clear beats long and legalistic.

Which tool should I use? Depends on the task. Start from the learning goal. The prompt and purpose matter more than the brand.

Won’t students just cheat? Some will, as before AI. Transparency norms and assignments tied to process, reflection, and authentic problems reduce the incentive.

What about equity? Name it. Where possible, use tools the institution provides so access isn’t tied to who can pay.