"Bespoke Bots": Diverse Instructor Needs for Customizing Generative AI Classroom Chatbots
Through an analysis of existing resources and interviews with STEM instructors, this study reveals that while educators prioritize customizing AI chatbots to align with course materials and pedagogy over persona, their specific needs vary significantly by context, suggesting that modular design is essential for effective educational AI systems.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Idea: One Size Does Not Fit All
Imagine you are a chef. You have a massive kitchen (the classroom) and hundreds of hungry guests (students). You want to hire a robot assistant to help you cook.
The researchers asked: "What kind of robot assistant do teachers actually want?"
They found that teachers don't want a generic, off-the-shelf robot that does everything the same way for everyone. Instead, they want a "Bespoke Bot"—a custom-built, tailor-made assistant that fits their specific kitchen, their specific menu, and their specific cooking style.
How They Did It
The researchers didn't just guess. They did two main things:
- Scavenger Hunt: They looked at 182 different "instruction manuals" (prompts) that people were already using to teach robots. They found 10 common ways people try to tweak these bots.
- The Card Game: They sat down with 10 university professors (mostly in science and math) and played a sorting game. They gave the professors 10 cards, each representing a different way to customize a bot (like "Change its personality," "Teach it our textbook," or "Make it strict"). The professors had to sort these into "High Priority," "Medium Priority," and "Low Priority."
What They Found: The "Must-Haves" vs. The "Nice-to-Haves"
🏆 The Top Priorities (The "Must-Haves")
The teachers agreed on three things that are absolutely non-negotiable. Think of these as the foundation of a house.
The Textbook (Course Materials):
- The Analogy: Imagine a tour guide who knows the history of the world but doesn't know the specific museum you are visiting. The teachers said, "Our bot must know our syllabus, our slides, and our assignments."
- Why: If the bot doesn't know the specific class material, it gives generic answers that might be wrong for that specific course.
The Teaching Style (Pedagogical Strategy):
- The Analogy: Some teachers are like Socratic coaches (asking questions to make you think), while others are like Sage on the Stage (giving clear examples).
- Why: Teachers want the bot to mimic their style. If a teacher hates giving direct answers, they want the bot to say, "Try to figure it out yourself," not "Here is the answer."
The Safety Rules (Guardrails):
- The Analogy: This is the fence around the playground.
- Why: Teachers want to make sure the bot doesn't lie, doesn't give away homework answers, and doesn't say anything mean. However, they felt this should be "built-in" by default, not something they have to spend hours tweaking.
📉 The Low Priorities (The "Nice-to-Haves")
Surprisingly, teachers didn't care much about two things that tech companies often focus on:
The Personality (Tone/Persona):
- The Analogy: Do you want your robot assistant to be a joking clown, a strict drill sergeant, or a friendly neighbor?
- The Verdict: Most teachers said, "Just be neutral and professional." They worried that if the bot is too "personable," students might get too attached to the AI and stop talking to real humans. They didn't want a "character"; they wanted a tool.
The Formatting (Content Format):
- The Analogy: Does the bot need to speak in poems, tables, or code blocks?
- The Verdict: Teachers said, "Just give me the answer." They care about what the bot says, not how it looks on the screen.
The Twist: Where Teachers Disagreed
While everyone agreed on the foundation, the teachers were polarized on the rest. This is where the "One Size Fits All" idea breaks down.
- The "Personalization" Debate:
- Teacher A (Large Class): "I have 4,000 students! I need the bot to know that Student X is struggling with math so it can give them extra help."
- Teacher B (Small Class): "No! I have 40 students. If the bot treats everyone differently, it's chaos. I want them all to struggle a bit and learn together. Don't personalize it!"
- The "Logistics" Debate:
- Some teachers wanted the bot to act as a Secretary (answering emails, grading papers, organizing groups).
- Others said, "I can handle my own emails. I'd rather the bot focus on teaching concepts."
The Solution: "Teachable Teammates"
The paper suggests we stop trying to build one giant, super-smart robot that does everything. Instead, we should build a team of specialized robots (or "Teachable Teammates").
- The Grader Bot: Only knows the grading rubric and safety rules.
- The Lab Coach Bot: Only knows safety protocols and lab equipment.
- The Tutor Bot: Only knows the textbook and how to ask Socratic questions.
Why is this better?
Imagine you are building a LEGO castle. Instead of buying one giant, pre-built castle that you can't change, you get a box of modular bricks.
- If you are teaching a huge lecture, you snap on the "Logistics Brick."
- If you are teaching a small workshop, you snap on the "Personalization Brick."
- If you are teaching a strict math class, you snap on the "No-Hints Brick."
The Takeaway
Teachers don't want a magic wand that solves all their problems with a single button. They want a Lego set. They want the ability to pick the specific pieces they need for their specific class, snap them together, and share their unique creation with other teachers who might want to use it too.
In short: Customization is key, but it needs to be flexible. What works for a massive calculus class won't work for a small design workshop. The future of AI in education isn't a single "perfect bot"; it's a toolbox of customizable parts that teachers can mix and match.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.