Debates about AI in education often stay at altitude: policy frameworks, competence models, ethical principles. The third webinar held in April 2026 under the Hub's AI, data and security theme landed somewhere more hands-on. Two practitioners came with working tools, live demos, and hard-won experience of what actually happens when you put generative AI in front of students - and one deceptively simple question: how do we equip teachers with the tools they need?
Coding is not dead - it has changed
Leo Briand, founder and CEO of Vittascience (Paris), opened the webinar with a provocation: in 2024, Nvidia's CEO publicly argued that children no longer need to learn to code. Briand disagrees. The foundations that coding builds - understanding how algorithms decide, how data shapes what we see, how machines can manipulate - are civic skills as much as technical ones. Over 50 countries have integrated programming into school curricula. The challenge is that teachers often lack training, equipment is unequal, and the technology moves faster than any curriculum can follow.
Vittascience's browser-based platform, used more than one million times a year across Europe, lets learners move between block-based and text-based coding while connecting to over 30 hardware devices. Its AI training module - used 365,000 times annually - allows students to train models on images, sounds, and sensor data, making the mechanics of AI visible rather than mysterious. It supports multiple language models, so students can compare outputs and begin to understand how choices about technology shape what they get back. Privacy by design, not by policy
Martin Sønderlev Christensen, PhD, Head of Institute for AI and Technology in Education at University College Copenhagen and co-developer of SkoleGPT, described what happened when ChatGPT arrived in November 2022: Danish students used it in secret, guilt-ridden, while teachers had no compliant tool they could legally put in front of a class. SkoleGPT was the response - a free, open-source chatbot built on open-source language models, hosted entirely on national infrastructure. Its GDPR solution is radical in its simplicity: no login, no logging, no user data collected or stored. Christensen called it "security through purity." SkoleGPT now serves between 3,000 and 5,000 users daily and won the Danish Language Award in 2024. Several participants noted that its relative simplicity is itself a pedagogical advantage: a tool that sometimes fails is one students can interrogate rather than simply defer to.
Productive discomfort
A live whiteboard activity asked participants when generative AI works best in schools. Responses clustered around personalisation, critical thinking, and time savings for teachers. The discussion that followed was sharper. How do we prevent AI tutoring from removing the productive discomfort that learning actually requires? Both speakers converged on the same answer: the question is not whether AI supports learning, but whether it supports a student's own thinking or quietly replaces it.
A related question proved equally pressing: since AI makes it easy for any teacher to generate lesson materials quickly, how do we ensure those materials genuinely contribute to learning rather than merely keeping students occupied? Both speakers agreed: the teacher still needs to read and verify what AI produces before it reaches students.
Both were also candid about a harder, still-unresolved question: when AI generates harmful content, who bears responsibility? Neither platform nor regulation has fully settled this, though both tools deploy guardrail models that screen outputs before they reach users.
From discussion to action
For educators: resist the impulse to start with the tool. Christensen's advice was direct - identify a real problem in your current classroom first, then ask whether AI can help solve it. Once you have found your use case, experiment with compliant, free tools and treat AI output as raw material, not a final product.
For institutions: safe, GDPR-compliant infrastructure is the prerequisite for any serious AI pedagogy.
For policymakers: open-source, nationally hosted solutions like SkoleGPT show that Europe can take meaningful control of the tools used in its classrooms - and that doing so is a condition for trust, not a constraint on quality.
Explore and connect
The session recording and speaker presentations are available on the European Digital education Hub. Not yet a member? The Hub is open to educators, researchers, policymakers, and anyone working in digital education across Europe. Joining is free.
Connect with Leo Briand and Martin Sønderlev Christensen on LinkedIn.
Resources
- Vittascience - open platform for teaching coding, robotics, and AI
- SkoleGPT - free, open-source, GDPR-compliant chatbot built for Danish schools
- SkoleGPT Science - subject-specific RAG extension for science education, in development
Please log in or sign up to comment.