La
Rochelle
31/07/2026
Rochelle
31/07/2026
👾
Pacman's · France
27 → 31 July 2026 · intensive week
Three causes, one week, a chatbot that had to prove it.
Five days at the Vent des Îles youth space in La Rochelle, with one sealed mission envelope opened each morning. Three teams chose their cause through a Sustainable Development Goals bingo, then turned a general-purpose AI into a local expert.
5Days
17hin the lab
11Researchers
The week, day by day
WikiYouthBot in the Local OuestMonday
Choosing a cause.
Tuesday
The heart and the sources.
Wednesday
The ears and the voice.
Thursday
The memory.
Friday
The personality.
01
Monday
Done
Done
SDG bingo · choosing a cause worth a week.
▸
What we ran
The week opened with no subject decided. The youth played a bingo built for the workshop on the seventeen Sustainable Development Goals: printed grids on the table, numbers drawn one by one, and each number bringing up a goal and a concrete situation behind it. Thirty situations went through the room that way, debated one after another, before a show of hands settled which ones stayed. Three causes came out of the vote and became the three teams of the week: spotting wildfire outbreaks from satellite images, mapping coral reefs, and peer mediation against bullying at school.
Who
11 researchers
Mood
Noisy and competitive, the way a bingo should be. Thirty situations, thirty arguments, and no adult deciding the winner.
Skills
SDG literacy
framing a problem
collective decision
teamwork
Best practice
A game that puts thirty situations on the table before any vote is what makes the choice real. At twelve to fifteen, the value lay as much in the arguments over each situation as in the three causes that emerged, and the youth defended them all week because they had chosen them.
02
Tuesday
Done
Done
Mission 01 · the heart and the sources.
▸
What we ran
Each team gave its chatbot an identity: a name, a mission in one sentence, and the exact territory its cause covers. They then wrote two or three rules of conduct, what it always does and what it never does, plus their own quality requirements: dated figures, named actors, precise places. In the afternoon, a voice-driven press conference put a general-purpose AI on the spot: the whole room questioned it aloud on their subject, rumours and fake news included. The bullying team came out of it with a rule written in their own words: stop inventing sources.
Who
11 researchers
Mood
Sceptical, then loud. Catching the AI citing a source that did not exist turned the room against it.
Skills
system prompt
source criticism
hallucination spotting
writing rules
Best practice
An AI answering out loud, to the whole room, gets caught much faster than the same AI typing on a private screen. At this age the collective reaction is what turns a vague suspicion into a rule the youth are willing to write down and enforce.
03
Wednesday
Done
Done
Mission 02 · the ears and the voice.
▸
What we ran
A day spent on the people who would actually open the chatbot. Each team drew and wrote persona cards: a first name, an age, a job, a family situation, favourite apps, and sliders for how much they know about the cause, how patient they are, how they feel about technology. Then, for each persona, the question they arrive with and the tone the chatbot owes them. The day closed with reference question-and-answer sheets: the question, the answer the team considers correct, the sources it must cite, and what it must never say. Seven personas came out of the day, including a conspiracy theorist written by the fire team to trap their own chatbot.
Who
11 researchers
Mood
Focused and playful. The personas got backstories, pets and terrible taste in apps before they got a single question.
Skills
persona design
audience awareness
tone of voice
reference answers
Best practice
Asking the youth to write a persona who disagrees with their chatbot is what pushes the work beyond decoration. Building their own opponent forced them to decide, in writing, how the chatbot answers someone it will not convince: never confirm without proof, never be brusque, keep asking questions.
04
Thursday
Done
Done
Mission 03 · the memory, built in the neighbourhood.
▸
What we ran
The field enquiry day. Colour-coded cards, one colour per type of information: a figure, an actor, a place or fact, a useful gesture. Every card carries the answer and a line where the source goes. The teams searched online and, above all, went out to question the people around the youth space. The fire team came back with thirty-five filled cards, from the role of the forest office to the four fire-risk sectors of the Charente-Maritime. The bullying team counted how many of five people knew the 3018 helpline, and what five adults would tell a young person to do. Those cards became the memory of the chatbots.
Who
11 researchers
Mood
Out of the room and back again. Interviewing strangers was the hard part; comparing the answers was the fun part.
Skills
field enquiry
dataset building
sourcing
data literacy
Best practice
A source line on every card is what makes the difference. Several cards came back marked "source: AI" in the youth's own handwriting, next to others marked with a ministry website or an interview, and that visible gap is far more convincing than any lecture on verifying information.
05
Friday
Done
Done
Mission 04 · personality, red lines and the reveal.
▸
What we ran
The last day gave the chatbots a face and a voice. Mascots were drawn and coloured: a smiling flame with smoke for hair for Feurisson, a character carrying a red, blue and yellow shield for Superbot. Each team fixed its tone and red lines, what the chatbot will never say whatever it is asked, then assembled an identity poster with the QR code that opens it. The fire team kept a constraint it refused to drop: Feurisson answers only in verse, because advice that rhymes is remembered better, and rhyming does not excuse being wrong. The week closed with each team presenting its chatbot to the others and to the adults present, live questions included.
Who
11 researchers
Mood
Last-minute and proud. Colouring the mascot while the poster was still drying, then defending the whole thing out loud.
Skills
character design
red lines
AI configuration
public presentation
Best practice
Drawing the mascot is not decoration, it is when the rules become concrete. At twelve to fifteen, giving the chatbot a face is when the youth start saying "it would never answer that" and write the red lines themselves.
What the week shipped
Published as the teams left themFire prevention
Feurisson.
A chatbot that answers only in verse, cites three sources before answering, and covers the fire-risk sectors of the Charente-Maritime.
Open the chatbot ↗
School bullying
Superbot.
A chatbot answering from a sourced base, which puts personal distress ahead of every other rule and points to a trusted adult.
Open the chatbot ↗
Robot prototype
Robo-Caillou.
A neighbourhood watch robot born from the enquiry, whose eyes follow the official fire vigilance level and which never sprays a person, only an ember.
Open the prototype ↗