La
Rochelle
18/06/2026
Rochelle
18/06/2026
👾
Pacman's · France
Dec 2025 → Jun 2026
From Pacman's to the Youth AI Lab · creativity included.
Six workshops from December to May: generative AI, reinforcement learning, a three-part "Stupid Invention" challenge, and a full-day immersion at the French Robotics Cup, with our Pacman's crew aged 12 to 15.
6Sessions
21h30in the lab
13Researchers
Workshops done
Six sessions in the Local OuestDec 2025
Inside the Machine.
Jan 2026
Reinforcing AI.
Mar 2026
The Stupid Invention.
Apr 2026
Pitching to investors.
Apr 2026
The public pitch.
May 2026
French Robotics Cup.
01
Dec 2025
Done
Done
Inside the Machine · Rule Finder, Word Predictor, Storyteller.
▸
What we ran
A first hands-on encounter with generative AI. Three short activities, each played twice: once by the youth themselves, once by a generative AI solution. A guessing game with seventy-two image cards to feel out pattern recognition. A word-by-word prediction game played by hand to show how AI text actually works. A short collaborative story to see how the AI reuses context. The point was to compare, side by side, how humans and machines tackle the same task.
Who
4 researchers
Mood
Curious and engaged. Lots of laughter when they caught the AI making things up.
Skills
pattern recognition
prompting
critical thinking
data literacy
Best practice
At twelve to fifteen, the youth engage much faster when they play the activity first and meet the AI after. Doing the task themselves gives them something to compare against, and the AI stops feeling like magic.
02
Jan 2026
Done
Done
Reinforcement Learning · unplugged to plugged.
▸
What we ran
Three hours to figure out how a machine actually learns. First half unplugged: a six-by-six coloured grid laid on the floor became a maze with hidden traps and a hidden goal. The group played the learning agent, one square at a time, earning reward tokens when they moved well and losing the lot when they hit a trap. Each youth filled a blank sheet to invent their own way of mapping safe and trapped squares. Second half on real robots (Sphero Indi, Vinci Bot, Tale Bot), programmed through navigation and creative missions, with the objective of reproducing the safest route to the hidden goal.
Who
8 in the lab
Mood
Loud and hands-on. The creative missions inspired and engaged the youth. The robots sparked a quiet competition between groups.
Skills
reinforcement learning
robotics
bio-inspired
debugging
Best practice
At this age, each youth invents their own way of mapping the maze: some go systematic from the start, others lean on memory, some draw pictures. Giving them a blank sheet for memorising their learning model is what lets those personal strategies appear, and then feed the collective discussion.
03
Mar 2026
Done
Done
The Stupid Invention, part 1 · ideation.
▸
What we ran
The session opened with a quiz: a slideshow of unusual products, real or AI-generated. The youth guessed which was real or fake and discussed what gave each one away. They then paired up to design a deliberately useless product, filling a short ideation sheet with the problem, the product idea, a name, a quick explanation and a sketch. The AI was on hand as a co-creator, especially to push the idea further or expand the description.
Who
5 inventors
Mood
Playful and focused. The absurd brief lifted the pressure to be right, and they kept iterating until their idea made them laugh.
Skills
prompting
creativity
hallucination spotting
Best practice
A deliberately silly brief takes the pressure to be right off the youth. At twelve to fifteen, they feel free to experiment, push the AI further, and end up running many more iterations than they would on a serious topic.
04
Apr 2026
Done
Done
The Stupid Invention, part 2 · building the fake.
▸
What we ran
The session opened with the facilitator showing a fake but very convincing-looking product: real or AI? The brief for the rest of the day was clear: take the absurd invention from the previous session and build everything needed to sell it, from text and visuals to a physical prototype convincing enough to fool an outsider.
Who
7 pitchers
Mood
Concentrated, a bit conspiratorial. They got into the persuasion game and helped each other sharpen the fake.
Skills
fact-checking
argumentation
prompt iteration
Best practice
Asking the youth to build the fake themselves, rather than just spot one, is what makes the lesson stick. At this age they notice the manipulation techniques much better when they have to use them.
05
Apr 2026
Done
Done
The Stupid Invention, part 3 · final pitch.
▸
What we ran
The session opened with the youth recapping the project to each other. Roles were then split: one focused on the cardboard prototype, the others on the investor slideshow. The evening ended with a full pitch to a jury and a Q&A where the youth had to explain which parts were their own work and which came from the AI.
Who
3 finalists
Mood
Nervous before, proud after. The Q&A turned into a defence of their own choices.
Skills
public speaking
authorship
AI awareness
Best practice
Letting the returning youth recap the project in their own words gives them ownership of the work and surfaces what really stuck for them from the previous session. New youth joining mid-sequence settle in much faster when their peers explain the project than when a facilitator does.
06
May 2026
Done
Done
Coupe de France de Robotique · full day immersion.
▸
What we ran
A full day at the Coupe de France de Robotique, the French national robotics competition held in La Roche-sur-Yon. The youth explored robotics projects built by young makers and watched the matches, especially the junior final. They also joined two hands-on workshops with Planète Sciences Vendée: building a small "insect" robot from wire, a battery and a motor, then designing a keychain with a 3D pen.
Activity
French Robotics Cup, finale
Who
3 on the road
Mood
Wide-eyed. They came back fired up, already talking about coming next year as competitors rather than spectators.
Skills
systems thinking
field observation
failure analysis
Best practice
Seeing competitors their own age, sometimes younger, changed what the youth thought they could do themselves. They can picture themselves in a year from now, participating in the next round of the competition with their own robot.
Next investigations
June 2026
Catching AI in its biases.
An afternoon on bias in large language models. The youth will probe a model to find where it leaves people out, then propose ways to make its answers more representative.
July 2026 · intensive week
Building WikiYouthBot on an SDG question.
A full week tied to one Sustainable Development Goal question. The youth will build a small local dataset, write the chatbot's brief, and ship a specialised chatbot they can defend before an outside audience.