Stanford and Google Launch AI Quests to Teach Responsible Artificial Intelligence

Added context helps to communicate the importance of various steps and the care with which certain parts of the system need to be handled. Speaking broadly, stories are one of the ways in which new knowledge can be stored and retained mentally.

Victor R. Lee, Faculty Lead for the Stanford Accelerator for Learning’s Initiative on AI and Education

Schools add AI ethics units to curricula, but students struggle to move beyond memorizing principles. Lessons on fairness, privacy, and accountability remain abstract. Students learn that bias in AI systems causes harm and that data collection requires consent, yet traditional instruction rarely shows how these concerns arise when building and deploying actual systems or what decisions address them in practice.

AI Quests, a platform developed through collaboration between the Stanford Accelerator for Learning and Google Research, takes a different approach. The game-based system places students inside realistic AI development scenarios where they make choices about data collection, model training, privacy safeguards, and system deployment. Ethical and governance questions emerge through gameplay rather than through separate ethics modules disconnected from technical content.

In the latest quest, students help a fictional doctor develop an AI screening tool for eye conditions. They encounter decisions about patient data privacy, training dataset diversity, and trade-offs in real-world deployment. The scenarios mirror challenges that AI systems face in healthcare, criminal justice, and hiring contexts. The platform treats responsible AI as inseparable from the technical work of building systems.

To examine how game-based simulations change what students understand about responsible AI development and how they retain these lessons compared to conventional instruction, we spoke with Victor Lee, Associate Professor at Stanford University’s Graduate School of Education and Faculty Lead for AI and Education at the Stanford Accelerator for Learning.

Victor R. Lee is an associate professor in the Graduate School of Education at Stanford University and is faculty lead for the Stanford Accelerator for Learning’s initiative on AI and Education. Through his research, he asks what future-facing STEM knowledge, tools, and practices are important to know to enable active participation and critical engagement with our increasingly digitally-infused lives. He then uses the tools of educational design research to create examples for how we could get there.

Currently, this involves researching and designing learning experiences and resources for data literacy, K-12 data science education, and artificial intelligence literacy for both students and teachers. He also has maintained a portfolio of research related to elementary computer science education, maker education, and science cognition. His research is most often conducted through research-practice partnerships and involves designing, implementing, analyzing, and continually revising new learning experiences in real learning settings (such as schools, districts, or libraries).

Lee has co-authored multiple national reports for the National Research Council related to computing and data in education. His work has been featured in The New York Times, CNN, Forbes, Politico, and other national media outlets.

Lee completed his undergraduate studies at UC San Diego with an emphasis in cognitive science, human-computer interaction, and mathematics. He earned his doctorate in Learning Sciences at Northwestern University, where he was supported for several years through a fellowship with the NSF-funded Center for Curriculum Materials in Science.

Since leaving the Midwest and beginning his professional academic career, he has received the National Science Foundation CAREER award, the Jan Hawkins Award, a post-doctoral fellowship from the National Academy of Education and the Spencer Foundation, and various best paper awards. His book, Learning Technologies and the Body (published by Routledge), is the first compendium of current research of embodied technologies for learning. With Abigail Phillips, he published Reconceptualizing Libraries: Perspectives from the Information and Learning Sciences (2018). His most recent book, released in 2025, is Advancing Data Science Education in K-12. He is a past-president and elected fellow of the International Society of the Learning Sciences.

AI Quests walks students through the complete development pipeline. Each quest begins with a real-world problem and moves through data collection, model training, testing, and deployment. The structure differs from conventional AI education, which isolates technical concepts from application or treats ethics as a final checkpoint rather than an ongoing concern.

The eye screening quest puts students alongside a fictional doctor as they build an AI tool to detect eye conditions from images. Students select training datasets, establish patient consent protocols, test whether the model performs equally across patient populations, and determine deployment conditions for clinical settings. They encounter decisions about image quality standards, the size and diversity of training sets needed, and how to validate model accuracy before clinical use. Each decision produces consequences within the game’s narrative that students must address.

When students choose a training dataset that lacks diversity, for instance, they later discover the model fails to accurately diagnose certain populations. When they skip consent protocols to gather more data quickly, they face questions about patient rights and regulatory compliance. The quest structure makes visible the connections between technical choices and their ethical implications.

“AI Quests is unique in that it contextualizes AI from real-world problem identification to deployment – so this added context helps to communicate the importance of various steps and the care with which certain parts of the system need to be handled,” Lee says. “Speaking broadly, stories are one of the ways in which new knowledge can be stored and retained mentally, so this approach is meant to be in line with that capability.”

The story framework helps students understand why decisions matter and what happens when developers overlook critical considerations. Students see development not as a series of isolated technical steps but as interconnected choices, where early data decisions shape later fairness and performance outcomes. The quest format allows them to experience these consequences in compressed time rather than as abstract warnings.

The ethical questions students encounter through AI Quests emerge from the development process itself rather than from supplemental case studies or hypothetical scenarios. Privacy concerns surface when students must decide how to collect and store patient medical images. Data diversity becomes concrete when they see their model struggle with certain demographics. These moments arise naturally from the technical work, not as separate ethics lessons.

“The importance of privacy and diversity of training data is illustrated through the game with specific examples that are in the context of the problem that motivates the quest,” Lee says. “This approach lets students see what privacy would look like on medical records, for example, or see how the model performs poorly with certain populations.”

Traditional AI ethics instruction often presents privacy as a principle to memorize and bias as a problem to avoid. Students learn definitions without encountering the specific moments where developers must choose between competing priorities—more data versus stronger privacy protections, faster deployment versus more thorough testing across populations, or broader access versus safeguards against misuse.

AI Quests positions students at these decision points. They see privacy not as an abstract concept but as concrete choices about who can access patient images, how long data is retained, and what information requires explicit consent. They encounter bias not as a general problem but as measurable performance disparities when their model analyzes images from underrepresented patient groups.

The gameplay reveals governance questions that traditional instruction misses. Students must determine when their model is accurate enough for clinical deployment, who bears responsibility if the system produces incorrect diagnoses, and how to communicate the AI tool’s limitations to doctors who will use it. These questions lack clean answers, mirroring the ambiguity developers face in practice.

The contextualized approach changes what students internalize. Rather than learning that “AI systems can be biased,” they experience building a system, discovering its performance gaps, and deciding whether to delay deployment to gather more diverse training data. The lesson becomes operational knowledge rather than theoretical awareness.

The platform’s design draws on research in learning science on how students retain and apply new knowledge. AI Quests incorporates specific pedagogical strategies that conventional AI instruction often overlooks.

“We drew from findings and recommendations from the learning sciences – such as prompting moments of reflection and explanation,” Lee says. “The story structure is quite important as well, as it helps to motivate the various decisions and considerations in context. This gives students a chance to draw on prior knowledge as introduced by the story and framed by the problem they are trying to solve. Also, students have a chance to see the actual cases on which these were based, helping them to make connections and revisit some of the ideas and encounters from the game that are related to AI literacy.”

The quests prompt students to explain their reasoning at key decision points rather than simply making choices. When selecting a training dataset, students must articulate why they chose certain data sources and what tradeoffs they accepted. These explanation prompts force students to make their thinking visible and connect decisions to principles they learned earlier in the quest.

The story framework serves multiple pedagogical functions. It motivates why certain technical steps matter by grounding them in a problem students understand—helping a doctor improve patient care. It provides context that helps students draw on prior knowledge about healthcare, privacy, or fairness. It creates a coherent structure that students can remember and reference later.

The platform also connects game scenarios to real cases. After completing a quest, students can examine actual AI systems that faced similar challenges. A student who struggled with data diversity in the eye screening quest can read about real medical AI systems that performed poorly across different patient populations. These connections help students transfer what they learned in the game to understanding how AI operates outside educational contexts.

Building the platform presented significant challenges. “Designing robust and broadly usable educational resources that can be used effectively under real classroom time constraints is always challenging,” Lee says. “Some things took longer in development time because we wanted to make sure we could balance a number of priorities, including ease of use, high-quality representation of ethics and AI concepts, and research-informed pedagogy, and had to go through a few versions to strike the right balance. It is also tricky to show and depict some of the technical aspects of AI in a way that makes enough sense and is accurate but doesn’t get bogged down in details.”

The team iterated on the amount of technical detail to include. Too little, and students miss how AI systems actually work. Too much, and the content becomes inaccessible or consumes classroom time that schools cannot spare. The final design aims for technical accuracy without overwhelming students with implementation specifics, so they do not need to grasp responsible AI principles.

AI Quests addresses a gap in students’ understanding of AI’s scope and applications. Most students encounter AI through generative chatbots, like ChatGPT, Gemini, and Claude, and assume these tools represent the field. This narrow view misses the breadth of AI systems operating in healthcare, criminal justice, hiring, financial services, and scientific research.

“Generative AI chatbots are just part of the AI landscape,” Lee says. “AI Quests shows how AI that doesn’t look like a chatbot is actually used, and we think this better mirrors what professional scientists do when they build and use AI. Those applications still require careful consideration about responsible use and development. Ultimately, we hope that with AI Quests and other AI literacy efforts, the image of AI broadens beyond chatbots and everyone recognizes that no matter what the AI looks like or the context in which it is used, the idea of responsible AI applies.”

The platform demonstrates that responsible AI education requires more than teaching principles. Students need to see where ethical decisions occur in development pipelines, experience the tradeoffs developers face, and understand how early choices shape later outcomes. The eye screening quest gives students operational knowledge they can apply across AI contexts.

As AI systems expand into more domains, the gap between those who understand how these systems work and those who simply use them widens. AI Quests positions students to critically evaluate AI applications they encounter as future professionals, citizens, and decision-makers. The platform’s approach suggests that responsible AI literacy belongs in foundational education, equipping students with the judgment to ask meaningful questions about the systems shaping their world.