AI Literacy — where are the limits?
The closing task: no build, but a conversation. Together you collect what the tools of this book can do — and above all what they cannot.
In a nutshell
What: You gather all the reading tools from this book — from face recognition to the digital twin — and for each check two questions: what does it really measure? And what does it not know? No computer needed, only heads, cards and a board.
The core idea: Whoever has built an emotion recogniser themselves no longer mistakes it for mind-reading. Exactly this sober clarity is the most valuable thing you take from the book — and it can only be sharpened in conversation, not in code.
You need: the book (or your notes on the activities), cards or sticky notes and a board. Time to argue helps.
What it's about
At the start of this book stood a vault: the fabled crypt of the Rosicrucians, in which a knowledge lay locked — with a wager that one day the instruments and the minds would come to read it. You now hold a part of this code in your hand. The last question is no longer how one reads, but what one does with it.
Four perspectives have accompanied you, and each leaves an attitude — Paracelsus: read nature itself; the Rosicrucians: share your knowledge openly. And one sentence runs through all the chapters: in a readable world you do not only read, you are also read. Literacy means knowing both — and knowing the difference between what a tool measures and what it only pretends to know.
Before you start
This is not a solo but a shared task — the best insights arise in disagreement. Form small groups, give each a few tools from the book, and let it all come together in the plenary at the end. There is no "right" answer to tick off here, but a shared map that you draw yourselves.
A little background
Two limits, always the same. Through the whole book run two sentences that mark every boundary. First: expression is not experience. A model reads a smile, a tone, a posture — not the feeling behind it. Second: correlation is not understanding. That two things occur together does not mean one grasps the other. Whoever has internalised these two sentences is immune to most AI fairy tales.
Why building it yourself protects you. As long as an AI is a black box, it works like magic — and magic is easily believed too much. But you have built a spectrogram yourself, counted words, let a model guess and watched it be wrong. You know how thin the ground of some guesses is. Exactly this knowledge turns wonder into judgement.
How to proceed
- Collect. Gather all the reading tools of the book — one per card: face recognition (Ch. 9), voice (Ch. 10), Happimeter (Ch. 5), Symbiont Analyzer (Ch. 7), plant sensor (Ch. 11/13), animal emotions (Ch. 12), swarm mood and Tribefinder (Ch. 14), network analysis (Ch. 8/15), digital twin (Ch. 16).
- Take apart. Write two notes for each tool: What does it really measure? and What does it not know? Force yourselves to fill the second note as concretely as the first.
- Weigh. Add a benefit and a danger to each tool. Almost every one can be used for good or for harm — the purpose decides, not the technology.
- Turn it around. Finally the uncomfortable question: where in everyday life are you read with exactly these tools — by apps, cameras, platforms? Collect examples.
Where it all leads
When your map is finished, one thing stands out: none of these tools reads thoughts. All read traces — a face, a voice, a voltage curve, a word — and infer from them, carefully and fallibly, a state. That is powerful enough to be useful, and limited enough to demand humility. This double insight is the aim of the lesson.
Worksheet
The map of the tools
Fill this table together — one row per tool. Two examples are already entered.
| Tool | What it measures | What it does not know | Benefit | Danger |
|---|---|---|---|---|
| Face recognition (Ch. 9) | patterns of the face → estimated emotion | whether the feeling is real; the reason; the thoughts | alert on overload, accessibility | secret rating, surveillance |
| Plant emotion (Ch. 13) | correlation signal shape ↔ person's stress | whether the plant "feels" or "understands" anything | ambient well-being feedback | anthropomorphising, esoteric hype |
- Which tool is most readily mistaken for "mind-reading" — and how do you tell the difference?
- Find a tool that is care or exploitation depending on the purpose. What exactly decides which side it tips to?
- "You are read." Name three places in your everyday life where you are already read. Did you know it, and did you consent?
- Where does useful measuring end and surveillance begin? As a group, draw a line — and justify it.
- What do you personally take away: one rule you will hold to if you ever use these tools yourself?
Show discussion notes
1. Mostly the emotion and plant tools. The difference: they measure an expression or a correlation, never the content of a thought. Test question: "Could the person think the opposite and still show the same signal?" If yes, it is not mind-reading.
2. Almost all. Example animal emotion: for the early detection of pain it is care; for the performance exploitation of an animal, the opposite. It tips on consent, purpose and power imbalance — not on the technology itself.
3. Individual: face unlock, voice assistants, personalised ads, step and location data, typing behaviour. What matters is the honest double question: did you know about it, and was it a real choice?
4. No fixed point, but good criteria: transparency (does the person know?), consent, purpose limitation, proportionality, revocability. Surveillance begins where these are missing — especially with a power imbalance.
5. Individual — and the most important part. Good rules are concrete: "I analyse no chat without everyone's consent", "I say so when a guess comes from a machine."
For the facilitator
| If … | … then |
|---|---|
| the group only fills "what it measures" | Insist on the second column. "What does it not know?" is the actual learning core — if need be demand three not-knowing points per tool. |
| everything stays too abstract | Demand a concrete example from a built activity (12.2, 13.2, 14.2 …) for every claim. |
| an "AI reads thoughts" statement appears | Take it as a gift: let the group apply the test question from note 1 rather than simply correcting it. |
| the ethics debate tips into black-and-white | Introduce "purpose, consent, power imbalance" — the same technology lands on different sides depending on these three. |
| time runs short | Five tools are enough. More important than completeness is that each row honestly fills both columns. |
Food for thought
- The old Rosicrucians dreamed of a fellowship that cracks the code of nature and shares its knowledge for the good of all. You now hold a part of this code — earlier and more powerful than they dared to hope. What you do with it is not written in this book. It rests with you.
- You read — and you are read. That is no dark threat but an invitation to alertness: whoever knows how they are read can decide for themselves what to show.
- Do you remember the girl from the prologue, at the table with the kalanchoe? Now you know what happened there: no miracle, but an honest signal, read by a tool you can now operate yourself. The vault stands open. Go through — and listen well.
Extension
- A debate. Split into two camps: "these tools make the world better" versus "… more dangerous". Each side must fairly restate the strongest version of the other before it objects.
- Read for a week. For a week, note everywhere you are read — cameras, apps, recommendations. Bring the list; it grows longer than you think.
- A self-commitment. As a class, formulate three rules for the responsible use of these tools — short, concrete, checkable. Hang them up in the room.