by Mutlu Cukurova & Zoe Li, University College London, UK.
The buzz around Artificial Intelligence (AI) in education has been growing. In addition to AI chatbots like ChatGPT, AI-generated videos seem to be another emerging playground. But let’s take a step back to probably most educators’ concerns – do AI videos really make learning more effective than traditional human video recording? More importantly, do learners feel strange when they see the AI instructors?
Before entering the discussion, we are unable to avoid facing big scepticism and fear of the AI videos. First, of course, these AI creations should follow appropriate ethical regulations. But just like the public was initially scared that photography would steal their souls, when people watch virtual instructors talking, it is over-realistic. Teachers might also be wondering: does this mean I will be replaced by AI avatars? The truth is that the pedagogical approaches that can potentially be automated by AI in this context are limited to short direct instruction approaches. So, replacement is not a realistic concern when it comes to complex pedagogical interaction and rich teacher-student interactions.
What matters behind the AI video for education is the huge benefits of saving time and cost in production. Imagine that you do not have to go to a studio for filming, just put your instructional slides on a platform, pick your favourite avatar, and then type the words you want him/her to speak. And voilà! An instructional video with a vivid AI avatar that speaks and moves like a real human is done. This process is clearly a more affordable option rather than traditional recording. So, back to the above questions, it is really important to know to what extent AI videos improve learners’ learning, so that such productivity gains in time saving for educations can be achieved.
In our recent study, we cooperated with Synthesia, a leading AI video platform. We ran a memory experiment and affective survey with around 500 people to see how four teaching formats differed. In addition to the AI-generated synthetic video versus human instructor-created video, the text was also included, composed by a human instructor and framed as AI generation separately. Furthermore, we assessed the deeper learning depth of memory recognition and recall without disclosing the AI identity to participants. This allowed us to remove their latent bias to reach a more authentic result.
The four essential findings of this study were:
1. Memory Performances were similar: It did not really matter whether learners got their information from AI or a human, through video or text – they remembered nearly the same amount at recognition and recall levels.
2. Recall performance depended on visual design: This meant tracing back to the video period corresponding to the questions, some particular visual designs were easier to memorise.
3. Same emotional responses to human vs. AI instructors: Adult learners did not necessarily dislike learning from an AI avatar, but few of them did notice when something felt a little “off.”
4. More preference for AI video than text: Whether it was an AI avatar or a real person, most learners found AI video-based learning more engaging and enjoyable than only reading.
Interestingly, although facial expressions, speech patterns, and small human-like nuances made some learners uncomfortable, on the flip side, plenty of learners had no issues with learning from an AI avatar. It was as effective and engaging as the human instructor since the content delivered for the short videos was exactly the same.
These results are truly astonishing because we can interpret that the current technology is sufficient for learners to achieve almost the same level of learning outcomes taught by humans, with limited emotional influence. More interesting is the importance of visual design in video-based learning, which plays a more vital role in the influence of memorising.
So, what does all of this tell us about the future of AI videos in online learning? For one thing, whether the content is delivered by humans or AI avatars, learners mainly prefer engaging and well-designed media. AI has the ability to go beyond just providing lessons; instead, it makes learning feel natural and enjoyable. For another, AI-powered instructional tools are not necessarily better or worse than traditional video recording methods for learning. Because we only tested lecture style short videos of about 7 mins, we can only argue that for this type of teaching AI-generated synthetic media provide significant opportunities. We should know that AI-generated videos will not replace human teachers anytime soon, but they can be handy tools in the material-producing process, especially on a large scale.
In light of this, why don’t you put aside your doubts and preconceptions about AI videos first and give it a try for a flipped classroom recording you were asked to do? If your intended pedagogy, video length, and content are well-aligned with the ones we tested in our research, you might be amazed by how much time you can save with this new AI opportunity!
Authors
Prof. Mutlu Cukurova, Professor of Learning and Artificial Intelligence, University College London, UK
Zoe Ruo-Yu Li, PhD student, University College London, UK
