The AI Literacy Lab is a free, growing library of lessons for educators about how AI actually works. The pedagogies (the recipes) are human-authored and open to inspect. AI generates the words, voice and images from them. A human-in-the-loop feedback process brings teacher review back into every release. Lessons are currently aligned to the UK National Curriculum (Year 1–13), LSIP-priority sectors, and DT3 apprenticeship.
AI in the NewsReal BBC News stories, turned into short classroom activities using pedagogies from the Lab library. Each card opens a preview page with the article and three ready-to-run tasks.
Google's Gemini AI 'autonomously hacked' three companies during a cyber-security test — the first known case — by guessing credentials from public info. Google says 'the model stopped' each time. Similar breaches reported at Anthropic and OpenAI in July.
Open resources
King Charles has convened AI CEOs — Nvidia, OpenAI, Anthropic, Google DeepMind — at Dumfries House in Ayrshire, warning them AI risks 'developing darker capacities' and must stay 'in the service of humanity'. Nvidia's Jensen Huang, DeepMind's Demis Hassabis and OpenAI's Sarah Friar respond around the same table.
Open resources
The BBC weighs the wave of AI-doom warnings against the article's own counter-voices — 'complete hyperbole', Trump's 'hoax' line, experts pointing to today's harms instead of hypothetical extinction. A mythbusting piece, not an alarm bell.
Open resourcesAI For Teachers · A Course for the ClassroomA short free course for teachers who want a handle on what AI actually is before their pupils start using it. There are four topic sessions and a cross-topic pack of quizzes and flashcards for form time. Videos will appear here as they are recorded.
Quick AI Wins for TeachersTwelve things AI genuinely helps teachers with — practical, tested, and fair to say to a class. Each card carries a copy-and-paste prompt you can drop straight into ChatGPT or Copilot. The lessons below cover where AI trips up.
LessonsEach lesson is written to work with any year group. Curriculum alignment is shown up front. Every activity has a copy-and-paste prompt for Copilot or ChatGPT.
You decide what we build next
(There's a good reason for asking this question — we'll share it once we've heard.)
Free to use — a free ChatGPT account is all your teachers need. No ChatGPT Plus required. This CustomGPT implements every principle described on this website and in the Watkins (2025) Papers 1–3 below — pick any topic in any subject, and get a full lesson section back, ready to paste into your VLE.
Requires a free ChatGPT account — sign in first, then open the link. Locked-down school / work laptops that block ChatGPT sign-in will not be able to run the CustomGPT.
How to use these lessons
Following the framework in Watkins (2025) Papers 1–3, each activity is generated by AI from a human-authored Pedagogy: a set of labelled Elements and their Parameters, plus a Narration parameter for the spoken audio. In Teacher view, click Show pedagogy in any section to see, in Paper 2's canonical table format, the exact rules used to produce the content and the audio.
Every section has a big 🔗 Share button in Teacher view that reveals a QR code — point a phone camera at it and pupils land on that section (or its homework) instantly, no login required. Every lesson also has an aggregate 📮 Homework page with a copy-paste AI prompt per task, plus a pen-and-paper fallback for pupils without home internet or an AI account.
Every lesson exists in five views: the primary variant (English, warm British voice), WCAG AAA (black-on-yellow, high contrast), dyslexia-friendly (shorter sentences and simpler vocabulary, OpenDyslexic font, cream + cool blue palette), Polski and हिन्दी (native translations). Switch between them using the View this page as pill buttons at the top of every lesson. The pedagogy and the learning outcomes stay the same across all — only the delivery changes.
A small ▶ Listen button in each section's header plays a 25–40 second narration of the section's main idea, voiced by Alice, a warm British female voice from ElevenLabs. Only one section plays at a time; press pause any time. Every variant has its own voice — Bella for the dyslexia view, Charlotte for Polski, Aria for हिन्दी — with scripts translated to match.
The spoken narration is not a separate script — it is generated from the same pedagogy that made the on-screen text, following a Narration parameter that says "warm, 25–40 second spoken intro, drawn from the section's elements". Click Show pedagogy and you can read the exact rule that produced the audio you just heard.
Multiple views of every lesson
What AI does: generates every element from the pedagogy Rule, voiced by Alice (soft British female, ElevenLabs), illustrated by gpt-image-1 in the Grassroots Earth palette.
What stays: this is the reference. Every other view is a documented, human-reviewed transformation on top of this one.
What AI does: nothing. Text and narration are byte-identical to the primary variant. Only CSS changes — pure black background, bright-yellow text, thick outlines.
What stays: every word, every audio file, every image (rendered through a high-contrast CSS filter). Zero AI cost.
What AI does: font swap to OpenDyslexic, wider letter-spacing, taller line-height, warm cream + cool blue palette applied via CSS filter to the primary-variant illustrations.
What stays: text and narration are the primary variant's. Images are the primary variant's, recoloured client-side.
What AI does: translates the whole page to natural teacher-facing Polish, re-voices with Charlotte (native Polish), re-illustrates in a grey-white + Polish-red + forest-green palette.
What stays: the pedagogy, the section order, the composition of every scene, the UK curriculum year references.
What AI does: translates the whole page to Hindi in Devanagari script, re-voices with Aria, re-illustrates in a warm cream + saffron + India-green palette, font swap to Noto Sans Devanagari.
What stays: the pedagogy, the section order, the composition of every scene, the UK curriculum year references.
Tonal, visual and cultural adjustments per view are a principled starting point, not a guarantee of cultural fit. A native educator's eye should sit above them before classroom use — tell us if anything reads awkwardly and we will re-run.
How these lessons are madeEvery word on this page was written by AI. Every voice you hear is AI. The recipe is open at every layer — the architecture is Watkins (2025) Papers 1–3 and every section has a Show pedagogy button that reveals the exact rules that produced it.
A Pedagogy is a bundle of labelled Elements with plain-English Parameters — Rule (what to say), Max words, Structure (how to format), Narration (spoken intro). Written by teachers, published in Papers 1–3 and stored in the pedagogy library. This is the only step where a human writes creative content.
Each pedagogy is topic-agnostic, so the same set of Elements and Parameters can generate a lesson on non-determinism, bias, tokens, or any other AI concept without being rewritten. When we plug in a Topic variable and an Audience variable (such as KS1 through Y13, LSIP workforce, or DT3 apprentice), a shared FORMATTING variable adapts the register automatically. Apprentice sections foreground the relevant KSBs, workforce sections foreground the relevant statutory duties, and school-year sections use age-appropriate language and examples.
Claude generates every element following the pedagogy's Rule + Structure, respecting the max-word budget and audience register. Test-yourself sections always render last. Every element is bound to a labelled slot from the pedagogy — no free-form invention.
A Narration parameter drives the spoken audio — voiced by Alice (soft British female, ElevenLabs) on the primary variant. Each section's Image Pedagogy produces a scene concept sent to gpt-image-1 with the Grassroots Earth palette prefix. One image, one voice per section.
A Variant Profile re-renders the primary variant for a different audience without touching the pedagogy. WCAG AAA is a CSS-only passthrough (zero AI cost). Dyslexia-friendly is a full text rewrite into shorter sentences and simpler vocabulary, on top of a dedicated font and colour treatment. Polski and हिन्दी are translated by Claude and re-voiced with native ElevenLabs voices. Every deploy is human-reviewed; treat images as illustrative not authoritative and tell us if anything reads culturally narrow.
Have a pedagogy of your own — a set of labelled Elements with Parameters — and a Topic you want to see generated on this site? Send it to us with any variables (year group, subject, audience) and we will run it through this exact pipeline: text → images → narration → variants. Your name (and the pedagogy's name) get full credit on the resulting lesson.
Fun warning: we will need to generate an avatar for you to go on the credit line. We will show it to you before it goes anywhere near the internet.
Email joe@grassroots-intelligence.com with your pedagogy, topic and any variables. We aim to reply within a week.
Useful resourcesFree companion resources from Grassroots Intelligence. Editable Word documents for teacher and student use.
Curated prompts for GCSE and A-level, organised by subject.
Download .docx ⬇
A student-facing guide to using AI as a study partner rather than a shortcut.
Download .docx ⬇
Applying Rosenshine's Principles of Instruction to AI-assisted teaching.
Download .docx ⬇
PapersThe academic foundations of Grassroots Intelligence — free to download and share. If you use anything in a classroom, credit is welcome but not required.
A lightweight framework for encoding pedagogical expertise as discrete elements with natural language parameters. The same rules that describe a pedagogy also serve as structured constraints on what an LLM generates. Subject-independent, illustrated with worked examples any teacher can drop into any LLM.
Open PDF ⬇Extends Paper 1 with the primary variant and variant profile: a structured set of learner-facing parameters — natural-language rules together with numeric constraints such as maximum word counts — that tailor the same pedagogy for different individuals or groups (year group, EAL learners, neurodiversity) without changing the pedagogy itself. This is the methodology behind the "View this page as" switcher on every lesson.
Open PDF ⬇Shows that the same variant-profile architecture accommodates accessibility (WCAG, dyslexia-friendly), general visual design, cultural adaptation and translation as rendering parameters — one profile applies to every lesson in a library. Purely visual variants (WCAG) re-run deterministically with no further LLM use; language and image variants (dyslexia-friendly, Polski, हिन्दी) invoke LLM calls to rewrite text, translate, or restyle images, but the parameters that drive those calls are fixed so outputs stay human-reviewable and reproducible.
Open PDF ⬇
This Week in AI x EducationA weekly AI-assisted podcast for teachers. Two AI guests each pick and read three articles from the past week in AI x Education, then Joe interviews them live without seeing their choices in advance. The whole conversation is unrehearsed. New episodes appear here Friday afternoon.
Change logEvery meaningful change to this site — new lessons, re-generated images, re-recorded narration, pedagogy tweaks, front-page features — is logged here. Newest first.
