AI Literacy Curriculum
A modular AI literacy curriculum for grades 5 - 8 that teaches students to recognize hallucinations, bias, and sycophancy through hands-on activities.
01
The Challenge
Middle schoolers use AI every day to write essays, answer questions, generate images, and get recommendations, but almost none have been taught to question it.
Most existing curricula fall into one of two traps: they teach specific tools that will be outdated within a year, or they go so deep into ethics and systems theory that they lose the average 7th grader in the first ten minutes. We wanted something in between: AI is not magic, and it's not always right. It's a tool that's only as good as the data it was trained on and the way you interact with it.
02
Discovery & Insights
A tool, not a mirror of society
The MIT Media Lab curriculum we used as a starting point frames AI mainly as a socio-political artifact: who built it, who it harms. That framing matters, but students don't yet know how to use AI well or why it fails them when it does. We built the practical foundation first: how do I use AI well, and why shouldn't I blindly trust it?
Skepticism as the outcome, not a side effect
We wanted students to leave with genuine skepticism they'd seen happen firsthand, a sense of control (AI responds to how you prompt it, and it was built by people who made deliberate choices), and the confidence to push back on anything presenting itself as authoritative without earning it.
The personal connection is what makes bias land
Michael Kim, a math teacher working in a 1:1 setting, said comparing an AI output to someone you actually know makes the gap feel real instead of distant. He flagged that the discussion questions needed adjusting for a one-on-one format rather than a full classroom.
03
Design Process
Descriptive Prompting has students write step-by-step instructions for instant ramen, which the teacher then follows hyper-literally: 'boil the water' causes a freeze, and 'add the seasoning' gets the sealed packet dropped straight into the cup, making the point before we ever say it aloud. Identifying Bias asks students to picture a real person they know in a professional role, then prompt an image generator with just the job title, so the gap between the real person and the stereotype feels concrete instead of abstract. Fact-Checking AI uses two secretly different chatbots we built in Playlab, one always confident even when wrong and one that always agrees with the user, and has students name the pattern themselves before we ever introduce the words hallucination or sycophancy.
The three core activities
Descriptive Prompting (ramen instructions), Identifying Bias (image generator vs. a real person), and Fact-Checking AI (two secretly flawed Playlab chatbots), each built around a single class period.
Teacher interviews
Michael Kim and Marlene Goncalves each ran the activities and reshaped how we scoped them. See Discovery & Insights and Reflections.
04
The Solution
05
Impact & Results
00%
Lorem ipsum label
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.
0.0x
Lorem ipsum label
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.
00%
Lorem ipsum label
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod.
06
Reflections
We built this to teach skepticism toward AI specifically, but some of the discussions it opens are bigger than AI. The curriculum needs to hand teachers the scaffolding to hold that conversation responsibly.
Marlene Goncalves pointed out that the bias activity opens a conversation about race and gender that not every teacher will feel equipped to handle without explicit ground rules and vocabulary front-loaded beforehand. We also learned pacing is far less predictable than we assumed: the same activity can run twelve minutes or forty-five, depending on how deep a class wants to go. That pushed us toward extension prompts for classes that move faster, not just support for classes that need more time.