Grounding Air Castles of GenAI In Education

Let’s reflect whether EdTech is slowly reclaiming breathing room in the world of generative AI and Silicon Valley cults of personality.

Stanford researchers have conducted a study published this week that suggests a way of restructuring and rethinking how we might design EdTech applications so that they let students do the work instead of showing them how it’s done and hoping they’ll learn along the way. In this specific case, the research focused on students who are preparing for 中考 (Zhongkao), a school-leaving exam for junior high schoolers.

The study began with qualitative research, followed by a proof-of-concept app tailored for exam cramming. This tool fosters student agency and respecting students’ learning patterns that the qualitative part discovered.

It would be a disservice to parrot the whole study. I highly recommend reading it if you have 20 minutes. However, this study hints, whether intentionally or unintentionally, at a powerful criticism of pathological usage patterns of generative AI we’re seeing today. Just showing that GenAI is capable of solving everything for the student doesn’t mean we should encourage students to engage with answer dispensers.

I would even argue it is our duty as adults to educate children and young adults that using chatbots without any guidelines and pedagogical alignment has a negative impact on their ability to retain and reuse skills, which is backed up by current research that labels this unprecedented phenomenon as “vaporized learning”. Not to mention the other mental health risks and pathologies that are also under heavy scrutiny.

We should abandon the use of unnecessary and rote-learning online services that revolve around spewing out multiple-choice tests, reading long, meaningless prose, clicking the self-help button “explain like I’m 5” resembling slot machines in Las Vegas, or interacting with a function’s graph just for the sake of it and calling this “an interactive environment.”

It is undeniable that generative AI is becoming a part of our daily lives, for better or worse. That does not imply, however, we should accept using sloppy GenAI chatbot wrappers, especially in high-stakes contexts, one of which is pedagogy. Having a perfect system prompt or a suite of agents won’t cut it.

I’m not alone in thinking this. There are people who have already proposed frameworks that mandate how AI should be used in schools. Even though the practical adoption of such frameworks in traditional educational institutions is vastly unexplored, we should let schools and other educational institutions take back control and ownership of the tools they’re using. As is the case with other industries already, machine learning technologies should be easy to audit, predictable, consistent, affordable and, most importantly to me, serve end users—including children and teachers—rather than Big Tech corporations that use them for highly questionable purposes.

I would like to show Czech high schools that generative AI doesn’t have to be just a nightmare, but can be a tool that serves the needs of schools, teachers, and students. That’s why I started developing Heuri a few months ago and have been working intensively on it since. Currently, I’m in the middle of crafting a high school math glossary that will be the backbone of this EdTech platform and will reflect my individual tutoring practice, supported and enhanced by the insights of seasoned educators. Crafting deterministic resources onto which we can then try to graft non-deterministic tools should be the standard, not an exception of modern EdTech.

Happy learning!