Under the bonnet, we simulate a true end-to-end machine learning lifecycle: defining a community problem, collecting and cleaning data, training a model and seeing its impact in the real world. But to make it understandable for 11- to 14-year-olds, we had to give room for creativity and anchor everything to tangible, experiential learning:
We use gameplay to make the algorithms intuitive. By testing different data choices, students see how their flood prediction model evolves in River Fair. If they make wrong data choices, their model will perform poorly. They learn through experimentation, refining their approach for better results, mirroring the iterative and creative process of real-world research.
Stanford’s research showed us that ‘data’ as an abstract concept simply does not work with this age group. So, we represented the data within the game using physical SD cards. It’s a simple, tangible example, and the kids understood it instantly.
In our medical mission, Twilight Canyon, we introduce deliberate friction. Students build a model to diagnose an eye disease, only to realise that it works perfectly for one group of patients but badly for another. This ‘conflict’ forces them to actively understand why it is ‘unfair’ and how to solve it.
As for privacy at Google, this is non-negotiable, so we made sure to integrate it into AI Quests from day one. In the health mission, for example, the very first task for the students is to clean the dataset of names and personal details to protect the privacy of the patients.
