AI Alignment with Education

Project Overview

It is a common practice for generative AI (GenAI) models to go through “red teaming” and “alignment” processes before they are released to the public. But they – the models and tools powered by these models – can somehow escape such processes before they are deployed in educational settings.

This research project aims to critically examine and advance how GenAI tools can uphold and promote fundamental educational values. These values, which include student agency, cultural inclusivity, and ethical engagement, are central to learning environments that empower all learners. By conducting both technological evaluations of existing GenAI tools and forward-looking design explorations, the project seeks to shape the next generation of AI-driven educational technologies.

This project has two thrusts:

Pedagogical Biases in AI-Powered Educational Tools

We examine pedagogical biases in AI-powered educational tools. In a recent study focused specifically on lesson plan generators, we investigate how these tools may implicitly embed outdated educational approaches that limit student agency and classroom dialogue. To mitigate this issue, we further experimented with intentional prompt engineering, which showed promise in significantly enhancing these dimensions in AI-generated lesson plans. We offer practical strategies for educators and developers to mitigate harmful pedagogical biases while promoting contemporary educational values.

Of course, pedagogical biases are not limited to these two dimensions or lesson plan generators. We are currently working on a broader framework to identify and mitigate pedagogical biases in AI-powered educational tools. We are especially keen in drawing on different epistemological and pedagogical frameworks to imagine possible futures—especially those are currently neglected.

CraftPad: A Workspace for Teachers to Sketch and Refine Knowledge Building Designs

CraftPad is presented as a counter argument against efficiency-centric dialogues on AI in education. It re-centers educator expertise and judgment, with a focus on supporting teacher design of Knowledge Building (KB) lessons or classrooms.

Designing effective environments for knowledge building represents a sophisticated craft that demands deep understanding of KB principles, careful balancing of structure and emergence, and a design-oriented mindset. While generative AI has gained popularity in supporting teacher lesson planning, current systems—even the most advanced ones—operate within rigid templates that constrain teacher agency. These limitations prevent educators from engaging in the skilled, iterative actions essential to the craft of teaching.

Our work adopts an alternative approach that positions teachers as designers—rather than mere end-users—of AI tools. This perspective necessitates “meta-design”—designing for design itself—which empowers educators to configure and adapt tools in service of their professional craft and educational aspirations. By centering teachers’ expertise and creative agency, we aim to develop AI-enhanced environments that genuinely support the complex, nuanced work of knowledge building design.

Departing from these perspectives, we design CraftPad as a flexible workspace that centers teachers’ expertise and creative agency. CraftPad features: (1) an infinite canvas (similar to Knowledge Forum) providing space for open-ended ideation and flexible organization; (2) an embedded prompt library informed by KB principles to guide various AI generation tasks in support of teacher design; and (3) a flexible workflow enabling teachers to quickly initiate designs, iteratively refine specific components with their professional input, and compile design artifacts (e.g., lesson designs) that incorporate AI-generated content while preserving the primacy of teacher judgment. This approach embodies our meta-design philosophy by supporting teachers as designers rather than mere consumers of AI outputs.

Publications