The first time my partner Brian typed something into ChatGPT, he decided to see if ChatGPT could generate a lesson plan for a 60-minute training session on how to sell a specific product. He had just heard about this new AI engine, and wanted to see just how smart it was. What kind of lesson plan could this thing generate? And would it actually know what he was asking for?
Almost instantly ChatGPT generated a full lesson plan, with specifics about the product. At first glance, he was impressed. Weren’t we all?
As he scanned the lesson plan, it seemed instructionally sound, but a bit hollow. It suggested a case study, but he would need to go back in to ChatGPT to actually generate a case study. When he did, it generated a story about John selling the product to Sarah, and it wasn’t in the type of case study format he’d design for an Endurance Learning training program (not to mention that John and Sarah seemed to be the default names every time he asked ChatGPT to generate a case study).
Merging AI and Instructional Design
ChatGPT is pretty cool. But can it give you everything you need to create a training plan? Maybe, but there is still a learning curve for learning to write the best prompt to generate the content for all the pieces needed to deliver a training.
Before the launch of ChatGPT I had been learning how to get more out of Soapbox using the OpenAI API for GPT. We wanted to see what would happen if we were to use the power behind ChatGPT’s model and combine it with the instructionally-sound structure that our tool, Soapbox, was already using.
I started my tests by plugging activities into OpenAI’s Playground (sounds fun, right?) and seeing what I could control for an output. Just like with the development of Soapbox, my goal was to see how much we could reduce the burden on a user while producing high quality activities.
There’s a longer, nerdier story involving spreadsheets, Zapier,
CloudConvert and a bunch of other small tools that I’ll tell at some point.
Soapbox Plus GPT (not ChatGPT)
We’ve been pleasantly surprised by what we could create. We’re still very much in development but we wanted to a share a small sample of a lesson plan focused on writing an effective business email. This is what the AI behind ChatGPT plus the instructional strategies in Soapbox yielded:
It didn’t just generate these activities. It created alternatives for the Anchor, Content and Application activities. It also created activities if you were delivering this in person or virtually, with a really large group, with a really small group, using the Chat feature, using Breakout rooms, using only the Chat feature.
In other words, it isn’t just creating an activity. Soapbox was able to generate all of the activities that are required for Soapbox to cover every possible delivery scenario. Soapbox users depend on us for reliable design. What we’ve proven to ourselves is that we can reliably update activities to provide the same experience as Soapbox with more content.
The Future of Instructional Design and AI
The end goal is that Soapbox continues delivering training sessions built on a proven structure with trusted activities, slides, a facilitator guide and participant materials while making it dramatically easier to see content specific to the designer’s topic.
We’ve been testing this for over a month to see what’s possible and will be excited to share more as we get further in our tests. In the meantime, if you use Soapbox today and happen to design a training session involving better communication in email, you may be pleasantly surprised by how complete the content is!