Visible Judgment: Designing Responsible AI Use in a 120-Student Design Course

Project Overview

Project Description

A course support and responsible AI redesign project for a 120-student Human-Centered Design course at Columbia University. The project focused on improving instructional support, reducing course navigation friction, and redesigning AI-integrated assignments so that students’ judgment became visible and assessable.

Category

Responsible AI Use Design

My Role

Responsible AI Learning Designer

Audience

120 Columbia University students

Duration

Jan 2026 - May 2026

The Challenge

Course Support Friction

Students repeatedly asked about links, deadlines, resources, and assignment expectations. These looked like individual questions at first, but the deeper issue was fragmented course information across Canvas pages, announcements, reminders, and assignment materials.

Responsible AI Use Gap

Some assignments allowed AI use, but students could accept AI-generated outputs without showing what they evaluated, revised, rejected, or verified. This created a gap between polished submissions and visible design reasoning.

Evidence & Insights

Evidence Sources

  • Office hour and course communication questions

  • Repeated patterns from assignment grading

  • Midterm learning experience feedback

  • Student use of AI in problem statement drafts

Key Insights

  • Course support problems were often information architecture problems.

  • Students needed clearer, centralized access to deadlines, links, and resources.

  • Responsible AI use needed to be built into assignment structure, not only stated as a policy.

  • Students needed scaffolds to critique AI output and connect revisions back to research evidence.

Design Interventions

Intervention 1: Canvas Information Redesign

Goal: Reduce repeated navigation questions and make course expectations easier to find.

What I designed: A clearer Canvas homepage that centralized key links, deadlines, weekly resources, and reminders.

Why it mattered: The redesign treated course support as an information architecture problem, not just a communication problem.

Intervention 2: Judgment Note for AI-Integrated Assignments

What I designed: A short Judgment Note requiring students to identify one weakness in the AI-generated problem statement, explain what the team changed, rejected, or kept, and connect that decision to interview or background research.

Goal: Make students’ human judgment visible when using AI.

Why it mattered: This shifted AI use from output generation to evidence-based critique and accountable revision.

Intervention 3: AI Critique Class Activity

What I designed: A 10-minute class activity where students critique an AI-generated problem statement, identify one weakness, observe a modeled 3-sentence Judgment Note, and compare good vs. weak examples.

Goal: Teach responsible AI use through practice, not policy alone.

Why it mattered: Students practiced evaluating AI output before using it in their own assignments.

Outcomes

Reduced Support Friction

Centralized links, deadlines, resources, and reminders to make course information easier to access.

Improved Course Navigation

Optimized the Canvas homepage so students could locate key information without searching across multiple pages.

Made AI Judgment Visible

Redesigned AI-integrated assignments so students had to explain critique, revision, and research-based reasoning.

Supported Course Continuity

Created TA handover guidance to make support processes more consistent and sustainable.

Reflection

This project changed how I understand instructional support. In a large course, support is not only answering questions one by one; it is identifying patterns, redesigning the learning environment, and reducing friction for both students and instructors.

It also shaped my approach to responsible AI pedagogy. Responsible AI use cannot rely on policy language alone. It needs to be embedded into assignment design so students practice critique, evidence-based reasoning, and ownership of their final decisions.

What this demonstrates

  • Learning environment analysis

  • Responsible AI assignment design

  • Student feedback synthesis

  • Canvas information architecture

  • Scalable instructional support