# Bok AI Lab Reports: Spring 2025 Experiments and Projects **What We’ve Been Up To** Below are detailed summaries of our recent weekly lab sessions, outlining the challenges, discussions, and experiments we’ve conducted. --- **Week 1: Moving Beyond the "AI Tutorbot"** We began by rethinking AI's role in education, moving away from the idea of a single "tutorbot" and towards creating agentic, recursive workflows that augment existing classroom dynamics. Key experiments included: - **Exploring AI "Knowledge"**: Participants prompted ChatGPT to perform basic math first with its LLM alone, then with coding tools, to explore the heuristics of LLMs, exposing certain illusions of competence to better demonstrate what LLMs do well-- and what they don't. - **Slack Glossary Bots**: A trio of Slackbots that responded to user-submitted terms with glossary entries categorized as neutral, positive (“friend”), and negative (“foe”). This method introduced recursive and dialectical approaches to definitions and helped surface varying perspectives on these concepts. --- **Week 2: Polling as Pedagogical and Computational Practice** We examined classroom polling from a pedagogical and computational perspective, exploring new AI-driven interventions in polling workflows. Faculty Prompt: How can AI enhance polling activities in a General Education course? Our Approach: - Reconceiving polling as a computational concept (i.e., polling as repeated querying, http requests) - Mapping how prolific polling workflows are, to expose the mechanics of polling and identify points where AI interventions could occur Experiments Included: - Paper prototyping to rapidly test user interfaces and student experiences - Python notebooks demonstrating: - Camera vision for capturing polling data visually - CSV parsing to manage poll responses - Front-end mockups for responsive polling interfaces --- **Week 3: Automated Quiz Generation from Classroom Audio** We tackled quiz generation from audio input, turning lecture videos into structured assessments. Participants received Google Colab templates and explored how conversational data can directly inform assessment design. Our Workflow: 1. Save audio files with ClipGrab 2. Transcribe speech using OpenAI’s "transcribe" endpoint within a Python notebook 3. Generate quiz questions from the transcription using GPT prompts tailored to specific learning outcomes 4. Output these quiz questions in json using structured outputs Participant Projects Included: - Quizzes teaching undergraduate Latin grammar - Assessments for children's vocabulary development --- **Week 4: (Upcoming) Listening Agents and the Management of Ambient Classroom Data** This week we’ll prototype listening agents that interpret live classroom communication—via audio, Slack conversations, and collaborative documents—to facilitate real-time pedagogical interventions. Our goals are to understand: - The nature of meaningful "units" of classroom interaction - Ways to identify and highlight key instructional moments, such as confusion or breakthrough insights - Ethical implications and practical strategies for classroom use of these tools