How can I audit the algorithms behind everyday social media filters?

Originally published on January 8, 2026

This lesson is part of a broader set of AI Auditing for High School materials developed by UPenn GSE. CRAFT provides this page as a curated entry point, linking educators directly to the original lesson resources for classroom use.

Overview

Social media platforms increasingly rely on AI-powered filters and recommendation systems that shape how people see themselves and others. In this lesson, students learn a systematic, five-step process for conducting an algorithm audit by examining TikTok’s AI Manga filter. Using a curated database of inputs and outputs, students form hypotheses, collect and analyze data, and evaluate how algorithmic behaviors may reflect bias or unintended impacts. The lesson emphasizes evidence-based reasoning and helps students see how everyday users can meaningfully question and assess AI systems.

AI & Society
120 minutes

Originally developed by the University of Pennsylvania Graduate School of Education as part of the AI Auditing for High School (beta) curriculum

Digital Materials

Objectives

By the end of this lesson, students will be able to:

Describe the key steps involved in conducting an algorithm audit
Develop and test hypotheses about how an AI system behaves
Systematically analyze inputs and outputs from an AI-powered filter
Identify potential social impacts and fairness concerns in algorithmic systems
Communicate audit findings through structured reports and discussion

Questions explored

How can everyday users systematically test whether an AI system is fair or biased?
What kinds of patterns emerge when AI systems process different inputs?
How do design choices and data shape the behavior of social media algorithms?
What responsibilities do humans have when deploying or using AI systems?

Key Terms

Algorithmic Bias

When AI produces repeatable errors that create unfair outcomes, favoring some groups over others.

Training Dataset

The set of data used to train a machine learning model. The quality and size of the training dataset can significantly affect the model's performance.

How do algorithms reflect bias in everyday search results?

Search engines and other AI systems shape how people see the world, often in ways that feel neutral or objective. However, these systems can reflect and amplify social biases that are embedded in data, design choices, and societal patterns. In this lesson, students investigate algorithmic bias through image search results, using everyday search terms to uncover patterns related to race, gender, and representation. Through discussion, annotation, and reflection, students develop a sense of agency by learning how everyday people can question, audit, and respond to biased algorithmic systems.

Originally developed by the University of Pennsylvania Graduate School of Education as part of the AI Auditing for High School (beta) curriculum

How can AI help a community prepare for a flood?

Floods are among the world’s most devastating natural disasters, affecting millions of people each year. Today, scientists are using AI flood forecasting models to predict when and where floods may occur, giving communities more time to prepare and stay safe. But these predictions are not perfect. Data can be messy, and humans have many important steps to take in order to build trustworthy AI models that can help with decision-making. In this lesson, students will play Market Marshes, an AI Quest to explore how AI predicts floods, consider challenges with data accuracy, and reflect on the human responsibilities involved in making decisions with AI.

Created by Google Research and the Stanford Accelerator for Learning

How can we mitigate bias in AI design?

This lesson introduces students to the concept of bias in AI training data and its societal implications. Using a real-world example of Google’s Gemini AI, students will explore how overrepresentation and underrepresentation in datasets affect various groups.

Created by Chelsea Dixon

Adapted by Chris Mah