AI Quests has a new adventure, and this one goes inside the human head. On Monday 26 October 2026, Google Research and the Stanford Accelerator for Learning launched “Polar Peaks”, the third quest in the free series. Students aged 11 to 14 play as researchers who scan and map neurons to work out why a game character is losing their memory. Along the way they clean data, test a model and learn that AI still needs people checking its work.
The quest is modelled on more than a decade of real work at Google Research in connectomics, the science of mapping how every cell in a brain connects to the others. That work produced a 1.4 petabyte map of a fragment of human cortex in 2024 and, in September 2026, the complete wiring diagram of a male fruit fly’s brain and nerve cord, the largest brain map by number of neurons so far.
This article explains what Google announced, how the AI Quests series works, and what students actually do in Polar Peaks. We then look at the science the game is based on, with the real numbers, and set out what teachers, parents and businesses should take from it.
Table of contents
- What Google Announced: The Third of the AI Quests
- How the AI Quests Series Works
- Inside Polar Peaks: Mapping the Human Brain in a Game
- The Real Science Behind AI Quests: A Decade of Connectomics
- Why Mapping the Whole Human Brain Is Still Out of Reach
- What AI Quests Teaches About Responsible AI
- AI Quests for Teachers: Training Modules and Classroom Fit
- AI Quests Timeline: From Floods to the Brain
- What AI Quests Means for Schools, Parents and Businesses
- AI Quests FAQs
- References
What Google Announced: The Third of the AI Quests
The announcement came in a Google blog post by Ronit Levavi Morad, Senior Director at Google Research. It is short, but it carries three pieces of news: a new quest, a usage figure for the series, and new training for teachers.
Polar Peaks in brief
Google describes Polar Peaks as a journey to “the frontiers of neuroscience”. With help from in-game mentors, students scan and map the neurons in a brain to understand a character’s memory loss. Their tasks include data cleaning and model testing, which Google says teach them “firsthand that AI requires careful human oversight and critical thinking.”
When students finish, they hear from two of the researchers behind Google’s real brain-mapping results, including the recent map of the complete fruit fly brain.
Who built AI Quests
AI Quests is a joint project. Google Research supplies the science and the stories, which are all based on its own published projects. The Stanford Accelerator for Learning, an initiative at Stanford’s Graduate School of Education, is described in the official FAQ as “the primary co-creator”, providing the teaching design, the story-based challenges and the active learning parts of each quest.
400,000 quests and 10 languages
Google says more than 400,000 quests have been completed since the series began, and the game is now available in 10 languages. That figure counts completed quests, not individual students, so one pupil who finishes all three counts three times. Even so, it is a large number for a series that launched in English only, just over a year ago.
The three quests now available are listed below.
| Quest | Name | Real problem | Google Research project behind it |
|---|---|---|---|
| 1 | Market Marshes | Predicting floods | Flood Forecasting initiative |
| 2 | Dusky Dunes | Detecting eye disease | AI model for diabetic retinopathy screening |
| 3 | Polar Peaks | Mapping the brain | Connectomics research |
How the AI Quests Series Works
Each quest is a browser game set in a fantasy world, but the problems in it are real ones. Students take on the role of Google researchers and use AI to help the people in the story.
A fantasy world with real problems
The first quest set the pattern when it launched on 9 September 2025. In Market Marshes, students collect data such as rainfall and river flow, judge the quality of that data, then train and test a model to warn the game’s characters about the next flood. Google said at launch that students learn “how factors like human judgment and data quality shape AI’s outputs and impact.”
The second quest, Dusky Dunes, uses the same structure for medicine. Students gather data and train a model to help detect diabetic retinopathy, an eye condition that can lead to blindness. Polar Peaks applies the structure to neuroscience.
Professor Skye and the no-code design
Students are guided by an in-game mentor, Professor Skye. No coding is needed, and the official FAQ is clear that AI Quests uses no generative AI inside the game. Google says that choice is meant “to maximize access to this game and to make it adaptable for all curricula.”
That matters for schools. A game with no chatbot inside it avoids the questions many schools now ask about pupils typing into AI systems. It also means the model students train is a simple, visible one, so they can see cause and effect.
Three teaching pillars from Stanford
The FAQ names three principles behind the design: situated learning, authentic practice that mirrors real AI and data work, and game-based learning to keep students motivated. The teacher resources page says the quests are designed so students learn “through reflection and even productive failure.”
In practice, that means students are allowed to train a poor model, see it fail and go back to fix their data. That is closer to how AI work really goes than a lesson that only shows the right answer.
No accounts and no personal data
AI Quests does not ask for a name, a login or an email address. Google says it collects only anonymous, aggregated gameplay figures, such as the average time taken to finish each quest. For schools working under strict data protection rules, that removes one of the usual barriers to trying a new online tool.
Inside Polar Peaks: Mapping the Human Brain in a Game
Polar Peaks is the most scientific of the AI Quests so far. Floods and eye disease are problems most students can picture. Mapping the wiring of a brain is harder to imagine, so the game turns it into a mystery about memory.
The story: a character with memory loss
The quest begins with a character who is losing their memory. To understand why, students have to look inside that character’s brain at the level of individual cells. This mirrors a real goal of connectomics. Scientists hope that detailed wiring maps will one day help explain how memories form and what goes wrong in conditions such as Alzheimer’s disease.
Scanning and mapping neurons
Students scan brain tissue and map the neurons in it. Real brain mapping starts the same way: tissue is cut into thousands of very thin slices, each slice is imaged under an electron microscope, and computers stitch the images back into a three-dimensional volume. The game simplifies those steps, but the order is the same.
Data cleaning comes first
Before any model can be trained, the data has to be prepared. The official teacher guide says that “by preparing data and training an AI model, students experience some of the decisions and processes that scientists use to accelerate discoveries in neuroscience and medicine.” Cleaning data is the least glamorous part of AI work and often the most important. Polar Peaks makes students do it before they are allowed to see a result.
Model testing and human oversight
Once the model is trained, students test it and check its output. This is where the quest’s main lesson lands: the model makes mistakes, and a person has to find and correct them. Google’s blog post says the tasks teach students that AI needs careful human oversight.
That lesson is not invented for the classroom. In real connectomics, human experts spend years proofreading what the AI produces. We return to that point below.
Meeting the real researchers
At the end of every quest, students watch a recorded message from the real scientists behind the work. For Polar Peaks, those are two of the Google researchers behind the brain maps. It is a small touch, but it links a game to named people doing the job.
The Real Science Behind AI Quests: A Decade of Connectomics
The connectomics team at Google Research was formed in 2014. Its job has been to build the machine learning tools that turn huge stacks of microscope images into maps of individual brain cells and their connections. Polar Peaks is a simplified version of that work.
What a connectome is
A connectome is a complete wiring diagram of a nervous system: every neuron and every connection, or synapse, between them. The first one was published in 1986 for the 302 neurons of a tiny worm, Caenorhabditis elegans. Google notes that it took 16 years to make, and that researchers coloured in cells by hand from one cross-section to the next.
From 302 neurons to 166,000
AI is what took the field from hundreds of neurons to hundreds of thousands. The chart compares the neuron counts of five landmark maps. Each bar is that map’s neuron count divided by the 166,000 neurons in the 2026 male fruit fly map.
The human cortex fragment sits low on that chart only because it is tiny: about half a grain of rice. It is still enormously data-hungry: 1.4 petabytes for a speck of tissue.
How AI turns slices into 3D maps
Google’s best-known tool is the flood-filling network. It is a type of convolutional neural network that starts at a single pixel and finds every other pixel that belongs to the same cell, tracing it through layer after layer of tissue. This replaced the manual colouring-in used for the worm. It is computer vision on an extreme scale: the network has to follow one cell through thousands of images without jumping to its neighbour.
Other tools followed. SegCLR identifies parts of cells and cell types. TensorStore is an open-source library for storing massive multi-dimensional datasets, now used well beyond neuroscience. Neuroglancer is the browser viewer researchers use to explore the maps. Google’s state-of-the-art reconstruction system, PATHFINDER, improved its speed and accuracy by adding synthetic neurons to its training data.
The human cortex fragment
In May 2024, Google and Jeff Lichtman’s lab at Harvard published a map of one cubic millimetre of human cortex in the journal Science. The tissue came from the temporal lobe of a person with epilepsy, removed during surgery. The team imaged more than 5,000 slices, each about 30 nanometres thick. Image collection alone took 326 days.
The finished map needs 1.4 petabytes, or 1.4 million gigabytes, of storage. It contains about 57,000 cells and 150 million synapses. The chart shows how those cells divide, using Google’s own counts divided by the total of 57,000.
So most of the cells in that sample are not neurons at all. Glia, the support cells, outnumber them by two to one.
The surprises in the sample
The human map turned up things nobody expected. Most contacts between an axon and its target, 96.5%, involve a single synapse. But a rare few are far stronger: 0.092% have four or more synapses, and some pairs of neurons are joined by more than 50. Google’s statistics suggest these strong links are not chance, though their purpose is unknown.
The team also found a mirror symmetry in the deepest layer of cortex, where 77% of a type of triangular neuron tilted one of two mirror-image ways. And it found “axon whorls”, axons wound into tight knots. Google is careful to add that because the tissue came from a person with epilepsy, some of these features might be linked to the disease or its treatment.
The complete male fruit fly
The quest’s finale points to the newest milestone. On 3 September 2026, Google described a complete map of the male fruit fly’s brain and central nervous system, published in the journal Cell. The project was led by HHMI’s Janelia Research Campus with the MRC Laboratory of Molecular Biology and the University of Cambridge. Google segmented the microscope images; experts at Janelia and Cambridge proofread them.
It contains more than 166,000 neurons and 125 million synaptic connections. It also includes the ventral nerve cord, the fly’s equivalent of a spinal cord, so it links the senses to the muscles. As Cambridge neuroscientist Greg Jefferis put it to Janelia, “It basically lets us get from eyes to legs in one go.”
The table compares the main maps behind AI Quests and its brain-mapping story.
| Map | Year | Neurons | Connections | Led by |
|---|---|---|---|---|
| Worm (C. elegans) | 1986 | 302 | About 7,000 | Manual tracing |
| Fly hemibrain | 2020 | 25,000 | 21 million | Janelia with Google |
| Human cortex, 1 cubic mm | 2024 | About 16,000 | 150 million | Harvard with Google |
| Female fly whole brain | 2024 | 139,255 | About 50 million | FlyWire (Princeton, Cambridge and others) |
| Male fly brain and nerve cord | 2026 | More than 166,000 | 125 million | Janelia, Cambridge, Google |
Why Mapping the Whole Human Brain Is Still Out of Reach
The title of the new quest promises a map of the human brain, and the game delivers one in story form. The science is not there yet, and Google says so plainly.
86 billion neurons
Google’s fruit fly post states that “mapping the 86 billion neurons in a human brain is not yet possible.” The male fly map has about 166,000. Dividing 86 billion by 166,000 gives roughly 518,000, so a human brain has more than half a million times as many neurons as the largest map made so far.
The storage problem
Data is the other wall. One cubic millimetre of cortex took 1.4 petabytes. An adult human brain has a volume of roughly 1,200 cubic centimetres, or about 1.2 million cubic millimetres. Multiplying 1.2 million by 1.4 petabytes gives about 1.7 million petabytes, or roughly 1.7 zettabytes, at the same resolution. That is a rough illustration, not a Google estimate, but it shows why the field is working on smaller brains first.
Mice and fish come next
Google says it is working to map the hippocampus of the mouse with partners at Harvard, Princeton and elsewhere. It has also helped map part of the hindbrain of the elephantnose fish, in a Columbia University study in Nature, and is preparing a whole-brain dataset for the larval zebrafish with Harvard. Larval zebrafish are among the few vertebrates small enough to map from end to end, and they are see-through, so scientists can record brain activity in the same animal.
Human proofreading is the bottleneck
The hardest part is not the AI. It is checking the AI. Google writes that “mapping the fruit fly brain requires years of human effort just to verify and annotate the neural shapes.” Every reduction in that manual error correction lets research groups attempt bigger brains within realistic budgets.
That is exactly the lesson Polar Peaks teaches. The model in the game makes mistakes, and the student has to catch them. In the real lab, teams of experts do the same thing for years.
What AI Quests Teaches About Responsible AI
The science is the setting. The real subject of AI Quests is how AI systems are built, and where they go wrong.
Data quality shapes the output
Across all three quests, students learn that a model is only as good as its data. In the flood quest, they choose which data sources to use. In Polar Peaks, they clean the data before training. This is the same point that comes up in almost every failed AI project in business: messy, biased or incomplete data produces a confident model that is wrong.
Human oversight is part of the system
Many AI literacy lessons treat human checking as a safety add-on. Polar Peaks treats it as part of how the science works. That is accurate. The male fly connectome was “annotated and verified, or proofread, by a team of human experts at HHMI Janelia”, according to Google. Students see that the person and the model work together.
Learning from a model that fails
The Stanford-led design deliberately lets students fail. A pupil who trains a model on poor data sees the result, reflects and tries again. That builds a habit of testing AI output, rather than trusting it, that will be useful long after the game is forgotten.
Linking the game to the real world
The closing video from the real researchers makes the point that AI is built by people, for reasons, with trade-offs. For a 12-year-old, hearing a scientist explain why the fruit fly map took years of human work is a good counterweight to the idea that AI simply knows things.
AI Quests for Teachers: Training Modules and Classroom Fit
Alongside Polar Peaks, Google is rolling out teacher training modules inside the Google AI Educator Series. The modules for the first two quests are available now. Google says the module for Polar Peaks “will follow next month”, which means November 2026.
What teachers get
Each quest comes with a full lesson plan and a teacher guide with activities for before and after the game. There is also a teacher’s handbook, a document explaining the teaching approach and, for the flood quest, optional student notes. All of it is free on the AI Quests teacher resources page.
Fitting AI Quests into a timetable
AI Quests is designed to fit into existing lessons, not to replace an AI course. The FAQ suggests subject classes such as geography or science, and the quests were written for that kind of use. Stanford’s CRAFT programme hosts a 50-minute lesson built around the flood quest. There is also a Slides-based version of Market Marshes for classrooms with poor computer or internet access, aimed at ages 9 to 15 and lasting 30 to 45 minutes.
Partners already using AI Quests
The Raspberry Pi Foundation added AI Quests to Experience AI, its AI literacy programme built with Google DeepMind, on 30 September 2025. It sits in lesson 6 of those resources. CRAFT at the Stanford Accelerator for Learning also uses it, and Google invites other AI literacy providers to embed the quests in their own courses.
The table summarises what a school needs to know before trying AI Quests.
| Question | Answer |
|---|---|
| Cost | Free and open to all |
| Target age | 11 to 14, adaptable for others |
| Coding needed | None |
| Accounts or personal data | None; anonymous aggregate gameplay figures only |
| Chatbot inside the game | None |
| Minimum computer | Intel i5-5250U, 4GB RAM, Chrome 139 or later |
| Internet | Required (web game) |
| Polar Peaks teacher module | Due November 2026 |
AI Quests Timeline: From Floods to the Brain
The series has moved quickly. Here is how it got from a single English-language flood game to three quests in 10 languages.
September 2025: launch
Google launched AI Quests on 9 September 2025 with the flood forecasting quest. Its launch post promised two more quests “in the coming months”: one on diabetic retinopathy and one inspired by connectomics “to map and understand the human brain.” Polar Peaks keeps the second of those promises.
2026: awards and the second quest
Dusky Dunes, the eye disease quest, followed. Earlier in 2026, AI Quests won a Webby Award for AI Education, Learning and Training, and the award for Best Global K-12 Learning Solution at the ETIH Innovation Awards.
September and October 2026: the science and the game
The complete male fruit fly map was announced on 3 September 2026. Polar Peaks launched on 26 October, with students now able to hear from researchers behind that map. The Polar Peaks teacher module is next.
What AI Quests Means for Schools, Parents and Businesses
AI Quests is a small product in Google’s terms. The ideas behind it reach further.
For schools
AI Quests gives teachers a ready-made, free, data-light way to teach how AI is built, not just how to use it. That is a useful complement to the stricter rules some school systems are adopting; New York City, for example, has banned AI use for students until they reach high school. A game that teaches about AI without putting a chatbot in front of pupils sits comfortably alongside rules like that.
For parents
The quests run in a browser and need no sign-up, so a child can play at home. Polar Peaks is also a gentle way into a serious subject. Asking a child what the model got wrong, and how they fixed it, is a good test of whether the lesson stuck.
For businesses
The skills Polar Peaks teaches are the same ones that decide whether AI projects succeed at work: clean data, careful testing and human review of the output. Most failed AI projects stumble on those basics, not on the choice of model. Organisations planning their own systems can read our AI strategy guidance, and those wrestling with messy inputs can start with data management and analytics.
Connectomics is also a useful case study in AI at scale. A single cubic millimetre produced 1.4 petabytes. The lesson that storage, tooling and human review dominate the cost applies to any large data science project.
Limits and open questions
Three caveats are worth keeping in mind. First, AI Quests is made by Google about Google’s own research, so it shows the company’s work in a good light. Second, Google has not published independent learning outcome data; the 400,000 figure measures completed quests, not what students learned. Third, the brain in the game is a story. The real human map covers one cubic millimetre, from one patient.
None of that makes the quest less useful. It does mean teachers should frame it as one example of how AI research works, alongside others. For more on how AI and the brain are compared, see our explainer on why AI behaves more like a brain than a database, and our look at Meta’s TRIBE v2 brain foundation model. Our AI models and tools hub tracks the wider field.
AI Quests FAQs
What is Polar Peaks?
Polar Peaks is the third of the AI Quests, launched on 26 October 2026. Students aged 11 to 14 scan and map neurons to solve a character’s memory loss, learning data cleaning, model testing and the need for human oversight.
Who makes AI Quests?
AI Quests is made by Google Research with the Stanford Accelerator for Learning. Each quest is based on a real Google Research project: flood forecasting, diabetic retinopathy screening and connectomics.
Is AI Quests free?
Yes. AI Quests is free and open to all, with no account needed. Teacher guides and lesson plans are free on the official site.
Do students need to code?
No. AI Quests is a no-code game. It also contains no generative AI, which Google says makes it easier to fit into any curriculum.
What is connectomics?
Connectomics is the science of mapping every neuron in a nervous system and every connection between them. Google Research has worked on it for more than a decade, building AI tools that turn electron microscope images into 3D maps.
Can scientists map a whole human brain yet?
No. Google says mapping the 86 billion neurons in a human brain is not yet possible. The largest map so far, of the male fruit fly, has about 166,000 neurons.
When will the Polar Peaks teacher module be ready?
Google says the training module for the third quest will follow next month, so November 2026. Modules for the first two quests are already in the Google AI Educator Series.
References
Our newest AI Quest: Mapping the human brain (Google)
AI Quests: Bringing AI literacy to the classroom (Google)
AI Quests Teacher Resources (Google Research)
A connectomics milestone: Mapping the complete male fruit fly brain (Google Research)
Ten years of neuroscience at Google yields maps of human brain (Google Research)
A petavoxel fragment of human cerebral cortex reconstructed at nanoscale resolution (Science)
Sexual dimorphism in the complete connectome of the Drosophila male central nervous system (Cell)
Researchers reveal connectome of the male fruit fly central nervous system (HHMI Janelia)
Neuronal wiring diagram of an adult brain (Nature)
Introducing AI Quests within Experience AI (Raspberry Pi Foundation)
