critical ai handbook

The Critical AI Handbook is Human Restoration Project's guide to using AI in the classroom without losing what makes teaching human. Co-authored with Trevor Aleo, it treats AI as a tool rather than a replacement for teachers, and asks students to look closely at how these systems work, who they leave out, and what they cost. Grab the printable PDF on the left, or work through the prompts and activities on the right.

We do our best to keep these ideas current, and every interactable carries a review date marking when we last checked it for accuracy.

Start here

Using AI Without Losing Ourselves: A Critical Media Literacy for the 21st Century

We must be proactive in teaching students how and when to use AI while taking a critical lens to how it works and its potential pitfalls. This is the essay the whole handbook grows from, rooted in Paulo Freire's critical pedagogy.

read the essay
1

Foundations

What these tools actually are, what they can't do, and why a critical eye comes first.

How a model guesses the next word

The "black box" isn't really a black box. Step through a language model one word at a time and watch it run on plain probability, not understanding.

open
coming soon

Tools, not systems

A calculator never replaced math teachers. Wikipedia never replaced research. So where does AI fit, and where does it quietly fall short? A discussion to work through with your class.

coming soon

Using AI without losing ourselves

The handbook opens with Paulo Freire and critical media literacy. This is the thinking everything else is built on.

coming soon

Proactive, not reactive

Banning a tool doesn't teach anyone to use it well. Students are already using AI; the real question is whether we help them do it thoughtfully.

2

Multimodal Thinking

Move one idea across modes, genres, languages, and reading levels, and bring more voices to a topic than any single text could.

Text curation

Four frameworks for building rich, interdisciplinary text sets: Sunburst, Duet, Tree Ring, and Solar System. Map your core text, then generate around it.

open
coming soon

Make a standard student-friendly

Standards are written for teachers. Rewrite one in language an eighth grader would actually understand.

coming soon

Single-point rubric maker

Turn dense standards into a clean single-point rubric students can read.

coming soon

Translate and keep the shape

Translate a poem into another language and ask it to keep the structure, then talk about what gets lost and found along the way.

coming soon

Tailor a message's tone

Take the heat out of a hard email, or model how the same message shifts for parents, admins, and colleagues.

coming soon

Pair writing with sound

Ask for songs that match a poem or essay, then weigh whether the suggestions actually fit, and where the model's ear falls short.

coming soon

Visuals that match your writing

Adobe Firefly trains on licensed, compensated work, which makes it a more ethical place to generate images that pair with student writing.

3

Transformation

Tasks that weren't really possible in a classroom before. These open up creativity and inquiry instead of shortcutting them.

coming soon

Remix for a new audience

Watch how audience and purpose reshape a text by remixing it for a completely different reader.

coming soon

Breathe life into a fan text

AI can copy the surface of a fandom but misses the in-jokes and references. Have it draft, then ask students to fix what only a real fan would catch.

coming soon

Going on an adventure

Turn ChatGPT into a storyteller for branching, choose-your-own creative writing.

coming soon

Read across texts like a researcher

Trace a theme across several books, or compare how two eras handle the same idea. It's the kind of pattern-finding that used to take weeks.

coming soon

Research with Elicit

A Scholar-style assistant that summarizes findings. It's the same tool we use to build HRP's research database.

coming soon

Talk to a PDF

Upload a dense academic PDF and ask it questions, with each answer tied back to the page it came from. A way into hard texts without heavy scaffolding.

coming soon

Discussion: AI and academic research

Where does this information come from? How was the model trained, and why does that matter when you're citing it? Questions to work through together.

4

Ethical Dilemmas

The largest part of the handbook, and the reason we don't just ban these tools: bias, deepfakes, labor, privacy, and the environment are conversations students need to have out loud.

coming soon

Whose great works?

Ask for the ten best books of all time, then examine whose canon the answer reflects, and who's missing from it.

coming soon

Whose wars matter?

The model echoes the textbook. Ask which U.S. wars matter most, then compare with the Philippine-American War it tends to skip.

coming soon

Make it hallucinate

Riddles, slang, niche facts: find the prompts that make a confident model invent things, then talk about what that means for medicine, news, and research.

coming soon

Spot the deepfake

Real AI-generated images and face-swaps set beside the genuine article. Can you and your students tell which is which?

coming soon

Deepfakes, consent, and harm

AI-generated explicit imagery is already turning up in schools. A hard but necessary conversation about consent, before it happens rather than after.

coming soon

The environmental cost

A calculator that weighs a class's AI use against a school-bus ride or a hot shower, so the energy question stays concrete instead of abstract.

coming soon

When AI detection fails

Detection tools flag real student work as fake and poison trust. Why do we second-guess students' AI use but take the AI detectors at face value?

coming soon

Cite the machine

If a student used AI, how should they say so? A conversation about attribution in a remix culture that rarely stops to cite anything.

coming soon

The AI Incident Database

A searchable, user-submitted wiki of AI gone wrong, with thousands of documented cases of harm. Good fuel for a class discussion.

coming soon

What counts as knowing?

A model can retrieve a fact in seconds. The handbook's five E's of sense-making (embedded, embodied, enactive, emotive, extended) argue that isn't the same as learning.