Research map / Learning theory and algorithms / Prediction and decision losses

Learning About Algorithm Auditing in Five Steps: Scaffolding How High School Youth Can Systematically and Critically Evaluate Machine Learning Applications

AAAI 2025 · Learning theory and algorithms › Prediction and decision losses · approach: Data attribution and recourse · 10 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a five-step algorithm auditing scaffold (hypothesis, systematic inputs, running tests, analyzing data, reporting) for high school learners and illustrates it with a case study of two teens auditing peer-designed generative AI TikTok filters.

Problem

Youth learning about AI/ML lack systematic, empirical methods for evaluating how black-boxed ML systems work and their limitations; prior K-12 auditing activities rely on one-off observations rather than full systematic audits.

Approach

Adapts expert and user-driven algorithm auditing into a five-step sequential process with learner-specific scaffolds (open exploration, two-axis input organizer, test-recording tables, percentage analysis, peer-audience reports), and demonstrates it qualitatively with one youth pair.

Key results

• A single youth pair (ages 14-15) tested 33 inputs and found the peer-designed filter transformed all outputs into female-presenting people, with 75% of outputs having red hair and 26% of inputs having hair style changed.
• Youth followed all five steps, with scaffolds such as a two-axis input organizer and an input-output table supporting systematic testing.
• Youth audit reports were produced as TikTok videos aimed at the filter designers.

Measured effects reported in the paper

Surprising findings

• Generating a full set of inputs before testing was perceived as repetitive and challenging for youth, and their dataset remained small (33 images).

Limitations

• Case study covers a single group of two youth from one out-of-school workshop, limiting generalizability.
• TikTok and Effect House access restrictions make the approach hard to deploy in school settings.
• Input generation and data analysis were less systematic than expert audits, with small datasets and descriptive analysis only.

Experimental setup

Datasets: youth-created TikTok filter input and output images (33 test images) screen and video recordings of youth workshop sessions

Metrics: qualitative case study analysis of youth artifacts and recor

Builds on: algorithm auditing (Sandvig et al.) user-driven auditing evocative audits (Walker et al.) Morales-Navarro et al. youth auditing workshop

Closest papers by meaning

• What Can Youth Learn About Artificial Intelligence and Machine Learning in One Hour? Examining How Hour of Code Activities Address the Five Big Ideas of AI (AAAI 2025 · similarity 0.84)
• An XAI Social Media Platform for Teaching K-12 Students AI-Driven Profiling, Clustering, and Engagement-Based Recommending (AAAI 2025 · similarity 0.84)
• Word2Vec4Kids: Interactive Challenges to Introduce Middle School Students to Word Embeddings (AAAI 2025 · similarity 0.83)
• “AlphAI”: Teaching AI Algorithms to K12 by Training Learning Robots and Visualizing How It Works (AAAI 2025 · similarity 0.83)
• Smart Motor: A Low-Cost Hardware and Software Toolkit for Introducing Supervised Machine Learning to Elementary School Students (AAAI 2025 · similarity 0.83)
• Advancing Research on Equitable AI Education Through a Focus on Implementation: Insights from a Middle School Computer Vision Module Beta-Test (AAAI 2025 · similarity 0.83)