David Joyner recently wrote about his experience of creating a course with an AI avatar. He mentioned the advantages of being able to quickly and easily update course videos using AI rather than having to go to the recording studio and re-record and edit sections of videos. He also mentioned the drawbacks of uncontrolled use of AI to mimic real people.

Having taken over 200 online courses, I decided to try Foundations of Generative AI. I wanted to compare the experience from a learner’s perspective while learning more about the emerging AI field. How different would it be from learning from a real person? You can see an introductory video presented by DAI-vid here.

DAI-vid vs David

DAI-vid’s AI face and voice are realistic, very similar to David himself, who appears in the second video of the course. I thought the lip-synching of video and voice was well done. Whether or not it would pass muster with a lip-reader, I don’t know, but I found it believable. Better than some poorly-edited non-AI videos when the soundtrack plays ahead of the image.

The facial expressions and head movements were also well done, even at 1.25X speed, which I usually use for online courses and used for most of this one. If I hadn’t already known that DAI-vid was an AI creation, the voice and face might have fooled me.

But the hands! My first thought was that DAI-vid’s repetitive hand movements were rather irritating and very distracting. So distracting, in fact, that sometimes I caught myself watching DAI-vid’s hands rather than thinking about the narrative. It took me a while to realize that the main problem is that the hands tend to bounce around aimlessly at the ends of sentences. They emphasise important points, but the effect is ruined by those extra movements after DAI-vid stops speaking. And, noticeably, the hands remain locked together throughout every single video.

Just as imitations are rarely as good as the original, DAI-vid doesn’t display the full range of expressions and voice inflexions as the real David Joyner. In the course, DAI-vid wears a handmade bracelet, while David wears his watch and wedding ring. One learner confessed in the discussion forum that he thought DAI-vid was the real David.

Initially I thought the professor was the AI avatar, until he introduced DeiVid,(sic) who looked more real to me. Lol.

Course Structure

As well as introductory and wrap-up sections, the course is divided into three modules with three topics within each.

Each topic has five elements: a preflection that asks reflective-style questions about the upcoming material, the video lesson taught by DAI-vid, a reading list (optional, for those who want to explore the topic more fully), a 10-question quiz for paying learners, and a few discussion prompts that also work as revision of the topic. I found this format useful for learning: the preflection primes the brain for the subject, the video teaches the lesson, the reading list is there for more background, if you’re a paying learner the quiz tests your knowledge (you have unlimited attempts at the questions, which is a change from past edX quizzes I’ve done), and the discussion prompts give you the chance to put what you’ve learned into your own words.

You only get one attempt on the final exam, but if you’ve done well on the quizzes, you’ll have already passed the course.

If you audit the course for free, you can’t view the graded items and you are also subject to repeated pop-ups about the benefits of earning a certificate.

One aspect of edX that I like is the Progress page. This course mentions grades between A and F.

The passing grade is 61% (less than 61% is an F). 61-70% earns a D, with C, B, and A available in 10% increments. But the certificate does not include the grades. Incidentally, my certificate arrived as soon as I reached 61%, before I finished the last few modules.

The quizzes allow unlimited attempts (if you get it wrong, you have to wait at least 5 minutes), but you only have one attempt for each exam question.

Grady

Some optional steps in this course give learners the opportunity to help train an AI-powered assessment tool (named Grady) for open-ended answers. You’re encouraged to write essays on aspects of AI covered in the course modules, then you can provide feedback on the AI-powered assessments of your essays. I didn’t complete these steps, but might come back to them in the coming weeks so I can see how well Grady works.

Conclusion

An ongoing metaphor in Foundations of Generative AI compares AI with a car. Maybe you just want to be a driver and use the car to take you where you want to go. Or sometimes you’re a mechanic so you can fine-tune or upgrade the car itself.

Some topics in the course include a brief history of AI, why generative AI matters, patterns and probability, training and fine-tuning neural networks.

It was interesting and rather mind-blowing to learn about how generative AI works from AI itself.