AI in learning
Artificial intelligence is changing how courses are made and how people learn. Avatars present, tutors answer questions around the clock, language models draft courses from existing material, and learning paths adapt to each individual’s level.
The possibilities raise questions: How reliable is AI-generated content? How do you prevent invented statements? What applies to the voices and faces of real people? What do the GDPR and the EU AI Act require?
The articles in this section answer these questions factually, with references to the official texts. Where it fits, a short section shows the rules Universe has set itself.
AI avatars as lecturers: possibilities and limits
An AI avatar as a lecturer is a digital figure that presents course content with a synthetic or recreated voice and, in more advanced systems, answers learners’ questions live. It makes talks available at any time, in many languages and without filming, but it must be recognisable as AI, may only recreate real people with their consent and does not replace human subject-matter responsibility.
ReadAI tutors as learning companions: what they can and should do
An AI tutor is a learning companion based on a language model that answers questions about the course, checks understanding through follow-up questions, spots gaps and suggests suitable next steps. Good AI tutors stay on the course topic, rely on checked content, guide learners to the solution through questions rather than giving it away, and are protected against misuse by guardrails.
ReadAI-assisted course authoring: faster, but checked
AI-assisted course authoring uses language models and other AI services to draft outlines, narration, graphics, exam questions, translations and voice-overs from existing material. This shortens course creation considerably; the results nevertheless remain suggestions that subject experts check, correct and approve before they reach learners.
ReadFact-checking AI-generated course content
Fact-checking AI content means holding a course’s verifiable statements, such as figures, dates, legal provisions and definitions, against reliable sources and classifying them as supported, refuted or unverified. Good procedures archive the sources permanently, flag uncertainty openly instead of smoothing it over and document every check in a traceable way.
ReadAI voices in learning: speech synthesis and consent
Synthetic voices turn text into natural-sounding speech and make narrated courses quick to produce and easy to change in many languages. Recreating a real person’s voice requires their explicit, documented and revocable consent; a voice is personal data, and AI-generated voices should be recognisable as such to listeners.
ReadData protection for AI in learning: GDPR and AI Act
When AI is used in learning, the GDPR applies to all personal data and the EU AI Act applies to the AI systems themselves. What matters are data-minimal flows with no names in model calls, processing in the EU where possible, transparency towards learners, human decisions on assessments and avoiding prohibited practices such as emotion recognition in educational institutions.
ReadAvoiding AI hallucinations in learning
AI hallucinations are convincingly worded but false or invented statements from a language model. In learning they can be greatly reduced by grounding answers in checked course content and sources, putting verifiable claims through fact-checking, flagging unsupported statements openly and having the system decline rather than guess outside its topic.
ReadAdaptive learning paths: learning that adjusts
Adaptive learning paths adjust the order, content and method of learning to each learner’s level: what someone can already do is skipped, identified gaps are worked on specifically and then checked again. They are based on ongoing data from questions, exercises and tasks; AI can spot gaps and suggest suitable steps, while people set the goals and rules.
Read