I spent 2024-2025 year doing a school project with my 14yo teenager studying how #genAI works and their bias. It was an eye opener for him, and also for the teachers to whom he gave the sumup talk
I hope it will help him take some second thought with AI, because it's sure he will be confronted to using it.

(Reposting here a thread from blsky. https://bsky.app/profile/fanf42.bsky.social/post/3lyt3fy3lqk2g
It was about teaching / explaining gen AI to our children.)

And we need to remember that our teens never got a working internet search. They only have a broken SEO optimised search of useless content. They never get the chance to learn what good search results can look like.
For most of my son friends, AI is (the think) a far superior replacement for search.

But yes, school need to learn to teach critical thinking and curiosity and source validation more than ever, else we will just have a young army of brainless zombies as students

Did you just create such a project yourself, or did you use some existing curriculum? I'd be interested to see how this could be done :)

From scratch and fully tailored to the student, with perhaps 20h of working together on it, and much, much more on his side.
So, for context, he has to endure living with me, so he likely has a better understanding of floss, systemic, power structures (esp in late capitalism context), commons, ...

brain bias, privileges and impacts, etc than many school boy of his age (for ref, he runs archlinux on his laptop since he is ten, and I barely helped him for 3y). Talking about privileges, hey?
Plus, the idea of that school project in the French curriculum is to make children of fourteen discover

... how a long project is run, how you need to work ahead of the June deadline and regularly, etc. and a/b gender equality.
He had the IA idea, and I had some about project management, teaching him with practice to 1/ rely on the scientific method loop, 2/ build is project management tool box and

rituals/practices, 3/ progress iteratively and continuously, setting new milestones with the acquired knowledge and skills, 4/ start as simple and cheap as possible.
Given all that context, he already had his idea: demonstrate how much gender biased genAI tools are. Of course that wasn't that simple

@fanf42.bsky.social

From scratch and fully tailored to the student, with perhaps 20h of working together on it, and much, much more on his side. So, for context, he has to endure living with me, so he likely has a better understanding of floss, systemic, power structures (esp in late capitalism context), commons, ...

Bluesky Social

The whole progress in 7 month went roughly like that. My main tasks were helping him choose tooling (notion was a good fit for semi structure document oriented project, among others), remind him to find something simpler to try, and helping him being working regularly.
He wandered like that:

  • what kind how genAI tools exists? (As of 2024 Chatbots, image, music, slides)
  • OK, are chatbots biased? Start with 4 or 5 available online. Discover how to use it, prompt. Try simple things, not much success to expose biases.
  • read research on the subject (summary of papers, popularization post,

blog, etc - even ask genAI bots and notice they lie about reference, or what the seem authorized to tell)

  • new test with refined prompt: some gender biases exposed. As soon as things need to be bit astute, protection layer unravel
  • wonder if there is other biases than gender? Be horrified by the

privilege, rich, western culture centered biase
(All along, he was dutifully copying all his test prompt with the answers in notion, also noticing that sometimes, some not trap that used to work didn't anymore)

  • switch to image generation and notice that OMG, image generation is a whole other of

horror show than text: not only the basic anti stereotype protections that text bots had seem missing, but you can even choose your preferred stereotypic flavor. You get things like that for "draw me a 8y girl/boy in her/his bed room"

  • from here, wondering how genAI works and learning about self learning, training/labeling/rewarding, master prompt, defense side process, prompting.
  • wondering what is actionable, and master prompt seems very much so if you run your own genAI bot
  • learning about olama and deepseek,

  • and with tries and errors, building his own (very limited, given his laptop 8y old middle-entry video card) white macho bot with the good old time where white men where respected and women knew what jobs where ok

  • then, wondering what kind of artifacts he could do about that (producing an artifact other than slides is part of the school curriculum, some children choosing art like a painting, etc)

  • discovering that autonomy is also about knowing to ask for help, and emailing his referent teacher

  • with teacher advice, choosing to do a video explaining how genAI are stochastic parrots without mental model, nourished with the worst of internet including a kind of canibalism, built with the bias of programers and the will of corporations, and the danger of confort versus critical thinking

  • finally, create the slide deck presenting the meta learning, I'd roughly what I told above (project management, etc). That part took a lot of time, since I used it to help him create a good slide deck (not doing it for him, but saying a lot "too much text" "the transition 8s not flowing"...)

And yep, that was a really cool trip, for him and for me.
He got the highest mark, and everyone who had access to his slude and video learned something (teachers, his grand parents, his brothers, and even my coworkers).
(Yes, i'm proud and bragging, and I stop now)