Using AI-Generated Instructor Video for Multilingual Content in MOOCs – An Interview with Benedikt Brünner

Generative AI-tools can assist in creating storyboards, translating content, generating assets like background images, creating voiceovers, and translations into multiple languages. Major focus of the research is the ability to create multilingual educational videos using AI avatars that can reproduce content in various languages, including languages that the instructor does not speak.

The e-education team around Martin Ebner and Sandra Schön at Graz University of Technology in Austria has been at the forefront of experimenting with AI-generated learning content for massive open online courses (MOOCs). In spring 2023, the Austrian iMooX platform launched a course that was produced solely with AI. Presently, the team is working on multilingual MOOCs, that mirror the language diversity of the European Union.

Trailer of the MOOC “Open Educational Resources (OER) in Higher Education” | iMooX.at CC BY 4.0

The team utilizes AI extensively to enhance educational video production, making learning materials more accessible and adaptable for diverse audiences. Through AI-powered tools, they generate first drafts of storyboards that outline educational objectives and tailor content to specific audiences, ensuring clarity and relevance. Using AI text-to-speech systems like ElevenLabs, they produce multilingual voiceovers, allowing for video content in multiple languages without needing additional speakers. They employ tools such as HeyGen for creating human avatars and Adobe Character Animator for character animations, giving videos a more personalized feel. Additionally, AI-generated assets—like background images or music (with the AI tool udio) and sound engineering using platforms like Auphonic—contribute to the overall quality of the content.

• HeyGen: HeyGen is a generative AI platform that enables users to create lifelike AI avatars for multimedia applications.
• ElevenLabs: ElevenLabs is an advanced AI-powered text-to-speech platform that enables users to generate natural-sounding voiceovers in various languages and accents.
• ChatGPT: ChatGPT is an AI language model that assists in generating structured storyboards by providing prompts for video titles, objectives, target audiences, and durations, helping in the ideation process.
• DeepL: DeepL is a high-quality machine translation tool that supports the efficient translation of educational materials into multiple languages, enabling broader accessibility.
• Pictory: Pictory is a video generation tool that uses AI to create videos directly from scripts or storyboards, allowing for efficient content production that aligns with educational objectives.
• Udio: Udio is an AI-powered music creation platform that enables users to generate custom, royalty-free music tailored to specific video content, enhancing engagement.
• Auphonic: Auphonic is an AI-based audio engineering tool that optimizes audio quality by reducing background noise, leveling audio, and improving overall sound, making educational videos clearer and more professional.
• Adobe Character Animator: Adobe Character Animator is a tool that animates characters in real time using AI, allowing for engaging, animated educational videos.

Benedikt Brünner is a member of the e-education team at TU Graz, and facilitates the production of avatars, assets, audio and video. As part of his PhD project he has conducted initial survey research to gauge students’ reactions to AI avatars. While most students found the avatars to be authentic and natural, they still preferred real human teachers over AI avatars. However, many students indicated that they would not have realized that the videos were AI-generated, which shows the increasing sophistication of the technology.

In the interview, Benedikt Brünner discusses the potential of generative AI for creating educational videos.

Interview

Question: What ethical concerns exist around AI human avatars, especially in terms of ownership and consent?

Benedict Brünner: Ethical considerations are a major focus. For instance, we are currently uncertain about ownership of an AI avatar due to the cost and data involved in creating it. Our protocol is to only create and publish videos that have been approved by the person being represented. Every AI-generated video we produce is marked as AI-generated. Right now, we’re only creating avatars for specific projects, not enabling everyone at the university to generate an AI avatar, as it’s still unclear what happens if someone were to leave the university. Are we still allowed to produce videos with that avatar? These are questions without definite answers currently.

Question: Are you working to create empirical data or guidelines for others on how to build AI courses?

Benedict Brünner: Yes, absolutely. The feedback we’re collecting now will help us publish research on the topic, likely for next year’s conferences. I’m also working on a research paper from this data to help create guidelines for high-quality, multilingual educational content generated with AI. We’re developing frameworks for ensuring that AI-generated content is of high quality and effectively serves its educational purpose. We aim to share this research by mid-next year.

Question: Have you looked at the impact of AI on teaching practices and student experiences?

Benedict Brünner: In my view, it can make education more accessible to students, especially through open-licensed content that can be translated and used globally. We’re also looking into simplifying educational content using AI, which could make complex materials more accessible to those with varying levels of language or cognitive abilities. The key here is open licensing, open educational resources. This opens the door for educators and students for doing something with the content, translating it, transforming it with AI.

Question: How is interactivity integrated with AI in your educational content?

Benedict Brünner: We’re currently working with chatbots to create more interactive content for students. While these AI-driven chatbots are not as in-depth as personalized human responses from the teacher, they provide immediate answers at any time of day, allowing students to get a baseline of information and take small steps toward understanding. This is especially helpful when a teacher may not be immediately available.

Question:  How will AI impact the future of teaching and learning, and what are the long-term implications of using AI in education?

Benedict Brünner: I think it is changing the landscape. At least in my perception, k-12 students are already using smartphones on a day-to-day basis and consuming a lot of short videos. So I think we are moving in the direction that everything has a faster pace, and everything is shorter, and it has to be condensed more. That is a trend currently, and I don’t see a signal that is pointing in a different direction. So yes, AI-content has a tremendous impact, but still, the teacher, in our vision, has the important role as a guide on the side. The teacher tries to coach students to get the best learning experience, but not as the one and only source of information that has all the authority. The teacher tries to help the students collect all the information, maybe research the information themselves, and the teacher, as a person who is really well-educated and well-informed, is able to guide them on their personal learning path.

Further Information:

• Learning Video Canvas: Collection of Ideas for the Learning Video
• Interview with Martin Ebner on AACE Review
• Multilingual MOOC ‘OER in Higher Education’

About

Benedikt Brünner is a university assistant at Graz University of Technology in the research area of educational technology. He is part of the research project Future of Digital Education and Learning (FutureDEAL) at the Institute of Interactive Systems and Data Science of Graz University of Technology. He had previously taught computer science and biology in a secondary school in Austria for over two years.