AI can do a lot for instructional designers. Here’s what it can’t do.
In the past year, I’ve used ChatGPT to do some degree of instructional design work. I’m not alone, as one American study reports that roughly 57% of workers have tried ChatGPT and 16% use it regularly at work.
To see the potential of AI tools like ChatGPT, I developed a Brightspace course entirely with generative AI. I used a combination of ChatGPT, Bing’s Image Creator, and a prototype course builder developed in-house at D2L.
Using AI cut out hours of development time and created a usable course that could be deployed with minimal maintenance. However, I also learned about the limitations of AI tools in instructional design. AI struggled with some of the bigger picture course goals and required constant human intervention to fine-tune the intricacies of the content.
So, instead of writing yet another article about what ChatGPT can do for instructional designers, I wanted to write about what it can’t do. Here are some of my takeaways.
AI can’t do your needs assessment
My first mistake was conducting my needs assessment entirely in ChatGPT. I assumed that I could plug in my traditional needs assessment ‘script’ into ChatGPT and watch the business goals roll in.
What I missed was the important contextual information that you get from a real stakeholder. ChatGPT was an excellent starting point that produced quality course objectives and competencies, but it would hallucinate information about business goals and didn’t give me a satisfactory content outline. What I missed out on was the ability to clarify points or ask the important contextual questions that a real stakeholder or SME can address.
Which leads me to my next point …
AI can’t replace your SME
It is important to keep a ‘human-in-the-loop’ as you develop a course with AI. My SME was a valuable check on AI-generated content. They were able to point me towards learning points that were more aligned with their organization and provided useful context that ChatGPT just couldn’t give me.
AI can’t get it right on the first try
Be patient, it will take a lot of refining to get the content you need. Most of my time was spent specifying my phrasing and learning from previous prompts to guide ChatGPT. Even after you’re satisfied with the return, you’ll need to work with your SME to carefully proofread for incorrect information, cliched phrases, or repeated information.
12 lessons from my experience
This article just scratches the surface of my lessons learned from using AI for instructional design. Of course, there are a lot of ways that AI can optimize the instructional design process, but there are some definite blockers to look out for.
For a list of the pros and cons, check out the 12 lessons I learned below:
- AI isn’t the first step, always involve the stakeholder and SME
- Conduct your analysis to determine high level design approach
- Provide detailed parameters to AI such as keywords, existing outcomes
- Set many guardrails as AI will ramble (word count)
- Provide AI as much as you are willing to give up in IP
- Always assume you’ll be refining your prompts to get more out of AI
- Assume hallucinations, misinformation, made up references, etc.
- Review for cliched writing
- AI will repeat itself with phrasing
- Expect to write most transition statements
- Experiment with new design strategies (increased case studies, examples, branching scenarios
- Segment materials into:
• Content
• Assessments
• Discussions
I’d love to hear your thoughts. Have you designed with ChatGPT? What was your experience like? Leave any comments below.