Google AI Quests Adds a Brain-Mapping Game

A cartoon character has lost their memory, and a class of 11 to 14-year-olds has to work out why by mapping the neurons in a brain. That is the premise of Polar Peaks, the third game in AI Quests, a series built by Google Research and the Stanford Accelerator for Learning to teach children how AI systems are actually made.

Google announced the new quest on 26 October. In it, students “scan and map the neurons in the brain to better understand the game character’s memory loss”, according to Ronit Levavi Morad, a senior director at Google Research. Along the way they clean data and test a model, tasks Google uses to make the point that “AI requires careful human oversight and critical thinking”.

In brief

Polar Peaks is the third AI Quest, aimed at ages 11 to 14, and is based on Google’s connectomics research.
Google says more than 400,000 quests have been completed in 10 languages. That counts plays, not pupils, and no learning results have been published.
The AI Quests site says the game is free, needs no coding and uses no generative AI inside the game.

Students clean data and test a model to map neurons

The quest draws on more than a decade of Google’s connectomics work, the effort to chart how every neuron in a brain connects to the others. Google points to its part in the complete map of the fruit fly brain, which it calls “the largest brain map by number of neurons to date”. At the end of the game, students hear from two researchers who work in the field. Google’s post does not name them.

The new quest follows two earlier ones. The first had students forecast floods, an area where Google has built its own forecasting models. The second dealt with diabetic retinopathy, the eye disease that AI screening tools are now being tested on in clinics, as TSN reported when AI eye exams closed a youth screening gap in a Baltimore study last week.

The design is deliberate. When the series launched in 2025, Stanford’s Kristen Blair, one of the game designers, said the team wanted students to be “making choices and decisions and seeing the impacts of those decisions” rather than just being guided through how AI works, according to the Accelerator. Associate Professor Victor Lee, a co-creator, said many students are “not very familiar with what data are or how exactly they are selected and used” to train an AI.

Teachers get a module next month, and the game is free

Google is also adding teaching material. Training modules for the first two quests are already in its Google AI Educator Series, and the module for Polar Peaks “will follow next month”, the company says. The AI Quests site already lists a Polar Peaks teacher guide, which describes how students prepare data and train a model to map the brain.

The AI Quests FAQ describes the game as “an open and free self-service resource” that needs an internet connection. It says Stanford is “the primary co-creator”, providing “the core pedagogical expertise”, and that there is “no use of Generative AI experiences within the game itself”. Google says it collects no personally identifiable information from players, only “anonymous aggregated gameplay metrics”.

The series has picked up awards, including a 2026 Webby in the AI education category and ETIH’s Best Global K-12 Learning Solution award. “We’ll continue working with educators and partners to bring more quests to more classrooms this school year,” Levavi Morad wrote.

Interest in this kind of material is broad. Governments are also paying for AI skills programmes, such as the US National Science Foundation’s AI-Ready America hubs, and TSN’s weekly tech for good round-up tracks which such projects have been checked by anyone outside the maker.

400,000 completed quests is a usage count, not proof of learning

The headline figure needs care. Google says “more than 400,000 quests have been completed to date”. That is a count of games finished, not of children reached. One pupil who plays all three quests adds three to the total, and a class that replays a quest adds more. Google has not said how many students or schools have used the series.

Nor has it published any evidence on what players learn. The Stanford team built the game on learning research, and the FAQ cites situated learning, authentic practice and game-based learning as its pillars. But neither the blog post nor the site reports a test of whether players understand data, models or human oversight better afterwards. The Webby and ETIH awards are not evidence of measured learning outcomes either.

That gap matters because the core claim is educational. The strongest parts of the design are easy to check: it is free, it needs no coding, it avoids generative AI inside the game and it collects no personal data, by Google’s account. Those choices lower the barrier for a teacher with one lesson to spare. Whether a 12-year-old leaves able to explain why cleaning data changes a model’s answer is the question the series has yet to answer in public.

A published evaluation from the Stanford co-creators would be the next useful thing to see, ideally alongside the teacher module due next month.