AI avatars as lecturers: possibilities and limits

An AI avatar is a digital figure with a face and voice that acts as the presenter. In the simplest case it speaks a fixed text, much like a video with a narrator, only without camera or studio. The figure is created from a photo or a short video, the voice from speech synthesis or a recreated voice, and mouth, eyes and head move in time with what is said.

More advanced systems go further: the avatar answers questions. A learner asks by voice or text, a language model formulates an answer based on the course content, and the avatar speaks it lip-synchronously. The result is closer to a lecture where you can ask questions than to a video.

Both forms have their place and are often combined.

The greatest advantage is availability. An avatar is as patient at three in the morning as at ten, answers the same question for the fifth time without impatience and speaks the learner’s language. For organisations with shift work, many sites or international teams, that is a real gain.

A second advantage is changeability. When a law, a product or a process changes, the narration is updated and the affected section regenerated. Nobody has to step in front of the camera again. And because voices can be carried over into many languages, further language versions are created without new recordings.

Finally, a familiar face can build trust. When your own lecturer or a well-known manager appears as an avatar, with their consent, a course feels more personal than with an anonymous voice.

An avatar is not a person. It builds no relationship, does not pick up on nuances the way an experienced lecturer does and can be wrong. Live answers can contain errors, especially if the system is not tightly grounded in checked sources. Some learners also find avatars strange, particularly when movements or voice do not seem entirely natural.

It follows that avatars are well suited to presenting, explaining, reviewing and follow-up questions on the material. For individual advice, assessment with consequences or sensitive conversations, people remain responsible.

The EU AI Act (Regulation (EU) 2024/1689) contains transparency obligations in Article 50. People should be able to recognise that they are interacting with an AI system, and AI-generated image, audio or video content that convincingly depicts real people must be disclosed. For learning platforms this means in practice: an avatar and its voice should be visibly labelled as AI-generated.

If a real person is recreated as an avatar or voice, their consent is required. Image and voice are personal data under the GDPR. Consent should be documented, tied to a purpose and revocable at any time, and revocation should take effect immediately.

In Universe, the lecture for a slide exists as a pre-produced video; for follow-up questions a lip-synchronised head answers live. Both run on self-operated GPU servers in the EU. Above every slide it says that presenter and narration are AI-generated, with “What exactly?” for the details. Custom figures are only created once signed consent has been recorded, separately for image and voice; revocation takes effect immediately, and every face speaks only with its own voice. Answers in the course go through a verification chain against sources.