In the first three episodes of our AI Literacy series, we learned how to see AI as part of the environment kids grow up in, how to question what models “know,” and how to keep the human in the driver’s seat when learning with AI. This time we move from talking to making. Co-creation is the moment a child says, I have an idea – and a working game, song, or video appears on the screen.

There are two ways to experience this episode: watch our live conversation and screen share, or read this deep-dive version that distills the demos, the choices, and the lessons for families. Either way, the goal is the same: keep agency with the learner and creator while the tools do the heavy lifting they’re good at.

Watch it here → or read along

If this sparked something for you, pass it on, share it – the more we talk about AI literacy, the stronger our collective compass becomes.

Why Co-Creation – and Why Now

Stefania’s path into AI education and AI co-creation started years before generative AI. With Cognimates and Scratch at MIT, her focus was creative fluency – how quickly a young person can go from idea to execution – rather than memorizing what a variable or a loop is. 

That philosophy became the foundation for her 2021 research at Google’s People + AI Research group (PAIR). She recalled the early days of Imagen, one of the first text-to-image systems used internally at Google: “We found these power users who would spend more than three or four hours a day generating images. Some of them had never thought of themselves as artistic before.”

The paper, The Prompt Artists, captured a turning point: “People who weren’t good at drawing or Photoshop could now engage because they could just write a prompt and iterate. They began developing their own styles.”

These “prompt artists,” as she called them, had deep niche knowledge – one might know every type of train, another might visualize polar bears in deserts to spark emotion about climate change. “They were creating imagery you couldn’t find in the real world,” Stefania said, “but that would evoke strong feelings.”

But at some point, she said a lot of generated images became another media of AI slop.

“I’m curious,” she added, “if you get this feeling… when you look at a website or a deck and see an AI-generated image, you can immediately tell. It all has a similar aesthetic.”

I agreed: “It feels like in the beginning it was much more common – these illustrations everywhere, and they were just terrible. But people are more sensitive to visuals. And now I feel like a lot of people got back to buying images from photoservices.

But what’s fascinating,” I added, “is how kids respond. They use AI with much sharper imagination. They combine things that don’t exist together in reality – like creating a futuristic car out of impossible elements. For them, it’s not about saving money or making their own illustration. They’re truly experimenting.

That’s the heart of it – AI as a playground for impossible ideas. And kids know that intuitively.

Play with Code: The Family Game Project

Then Stefania suggested, “Let’s play with some tools and see what they can or cannot do – and if we think they are family-ready or not.”

We started on paper. Yes, paper. Thinking through a game or app before coding builds real agency – it teaches how systems work, not just how to use them.

And why this matters? Sketching screens and actions is already design thinking: what happens if you click here? what should the player feel? That’s where kids realize they can be creators, not just users – and where learning starts to feel like invention.

We decided to build a game. I said, “Can we build something like Pac-Man? For some reason I’m missing Pac-Man.”

Stefania lit up: “I have an idea – Pac-Man is a metaphor for the passing of time. He’s getting older as you play the game.”

From there, we began co-writing our concept live:

“The game starts simple, with a maze. Some ghosts appear and Pac-Man needs to avoid them.” “Ghosts look like flowers.”
“He can eat fruit and coins to gain points.”
“Let’s give it a twist – if he finds the exit before he gets old, he wins.”

Stefania pointed out, “This is important – most people know what Pac-Man looks like, but it’s helpful to write it down. Parents can sketch this with their kids on paper before going digital. It sets the creative vision.”

When she ran the idea through Replit (also possible to use Loveable or any other no-code AI platform), it returned a detailed plan: a maze, aging mechanic, 2D visuals, player goals.

But then she switched gears. “The problem with auto-builders,” she said, “is that they take away the agency. When kids create games, they want to say, ‘This is my game.’”

In her own tool, Cognimates (based on Scratch), the AI works differently:

“It doesn’t build for you – it asks, ‘How do you imagine the maze? Is it made of walls? What happens when you touch a color?’”

That Socratic design keeps the child in charge. If you remember, we discussed Socratic approach to learning in the previous episode. Math was our example. Creation is also learning, and socratic design sticks because the child makes the decisions.
Stefania: “I can tell it: my maze is made of purple bricks. Then the AI replies, ‘Have you ever tried using if touching color block before?’” – a question that nudges learning instead of replacing it.

“This is very different,” she said, “because now the child is the one deciding. They’re the one who drives.”

Remix Culture for Kids

She also mentioned the remix feature on Scratch: “Look, I can take any Pac-Man game another kid made and modify it. It’s like forking on GitHub.”

I laughed: “It never occurred to me that Scratch is basically open-source for kids.”

She told me the story behind that feature: “When we added remixing at MIT, the community split. Some kids said, ‘I don’t want anyone to change my game.’ Others said, ‘This is awesome – I want to see how many people modify it.’ But engagement skyrocketed. It lowered the barrier to entry because you don’t start from a blank page.”

That’s a perfect metaphor for AI co-creation: start with something existing, then remix it into your own.

Design First, Code Second: Using Google’s Stitch

Stefania also showed Stitch from Google Labs to generate UI designs. Even for a game, it helps to see screens before coding: splash, settings, pause, HUD. Pick mobile or web. Iterate on layout. Note the plan it generates; prune it to match your vision. Then implement.

Tip: Designs are suggestions. Keep your sketch nearby and reject anything that pulls you away from the idea your child cares about.

Play with Media: History That Sticks

On the media track, I decided to demonstrate something I was already working on – a short YouTube piece about the women who were the first “computers” and the first programmers. It actually dawned on me while I was preparing for the episode: I co-create with AI all the time, so why not share something real that any parent or child could use in their own projects?

I started in ChatGPT:

“Let’s start with the first computers. They were humans, mostly women. Build me a short 5-8 minute script about women as computers, as first programmers, and what else they pioneered.”

Then came the real work. “It gives you a great scaffolding,” I told Stefania, “but you have to check everything – every quote, every fact, every person.”

Once I had verified the script, I moved to Sora for video generation. “It’s super intuitive,” I said. “You prompt it, set how many scenes, and get a draft video in minutes.”

Stefania noticed, “I see Grace Hopper and the Hidden Figures women in there!”

“Exactly,” I said. “And I give them real quotes – fact-checked – so when the video plays, you feel their voice and their personality.”

It’s co-creation as storytelling: AI visualizes, but the narrative integrity stays human. “It’s such an amazing way to tell history,” I said. “Because that’s how we remember things – visually.”

Learning that every video generation takes about 5-7 minutes, Stefania suggested a tip for families: render short scenes while kids do another mini-task – writing a caption, collecting sources, or sketching the next costume. Waiting becomes part of the rhythm instead of a blocker.

Learning Through Music

Then I opened Suno, the AI music tool. “I didn’t understand why people made random songs,” I admitted, “until I thought about memory. My boys have to learn U.S. state capitals, and it’s hard to remember. So I made a rap – ‘North States Capitals Flow.’” →

Stefania laughed, “That’s so cool.”

“And you can use it for languages too,” I added. “I made a French alphabet song. It’s easy to remember – and takes seconds to make.”

Stefania linked it to cognitive science: “It’s proven that we remember better when we put information into rhythm. I use Suno to generate rap songs for learning Japanese. Like going straight from Romanian to Japanese and I would listen to it on my commute. I love that aspect of personalization.”

How about adding some friction?

The problem with GenAI is that sometimes the technology is so good, that can block creativity. A human thinks: “It’s so perfect, what can I contribute?”

Stefania told me about a study where students generated avatars. “One random chicken appeared in a picture,” she said. “Everyone asked, ‘Why is there a chicken?’ It added friction – and suddenly the room was full of ideas.”

I replied, “That’s why I’m not against hallucinations that much. They make you check, think, and redesign.”

“Totally,” she said. “But it only works if young people know hallucinations exist. Otherwise, they won’t question.”

And that led us back to literacy: “Reading and writing were once the basic skills you needed to succeed,” I said. “Now AI literacy is the new defense skill. You need it to stay safe and understand the world.”

Family Challenge: Create Your Digital Coat of Arms

This episode is our longest so far but we couldn’t leave you without some homework. We offer you a challenge that combines everything we tried.

“Think of your family as a creative studio,” I suggested. “Use these tools to make your own coat of arms.”

Here’s how:

Please send us your family’s coat of arms, and share this article and/or video with people who don’t yet understand AI – because they actually need to.

Closing Reflection: Where Co-Creation Is Going

Phones are becoming studios. On-the-go generation plus short reflection windows means more what-if loops: what if this scene runs backward, what if the main character ages in real time, what if the soundtrack is only body percussion. And that’s how for centuries people created stories. That’s a “what-if” path that AI can accompany you at. 

With images: kids already mix things adults consider “not combinable.” That instinct is the seed of scientific creativity – taking two domains that rarely meet and making them talk.

It’s important for parents and teachers to talk about the effect AI has on the environment and everyday ethics. When children learn to notice energy use, data waste, and how digital choices ripple through the physical world, ethical thinking becomes second nature. The kids who started counting how many images they generate each week because of the planet’s cost are already showing that shift. Families can set light, consistent rules: show sources, mark generated media, respect licensing, and talk about consent when people’s likenesses appear.

That’s the essence of AI literacy – learning to think, together, with our new machines.

👉 Next in the series: The next episode will be special: “AI in the Physical World” with LeRobot, Jetson Thor kit, and some surprises.

Resources and further reading

Play with AI and ML: