Can AI design your course?

In this new, AI-driven world, a curriculum based purely on knowledge is not going to prepare learners for the complexities of the modern workplace. In response, we are transitioning towards a competency-based pedagogy, where a blend of in-person classes, asynchronous e-learning and generative AI play important roles. 

This shift toward mastering relevant, real-world skills and equipping graduates with demonstrated capabilities requires agile and speedy responses. Without it, we risk the curriculum becoming irrelevant to our graduates’ lives. 

A pedagogical pivot like this involves multiple stakeholders in the university’s ecosystem. The largest impact involves academic staff with disciplinary and classroom management expertise, and curriculum and content developers, who bring learning and instructional delivery expertise. Everyone wants speed and accuracy but no one can accept misalignment or the sort of hallucinations and “plausible nonsense” AI often generates, especially when AI-generated drafts are introduced without a clear governance routine. 

A new division of labour

Our team of instructional designers and media producers, using frontline AI and AI-enabled authoring tools, had accelerated our learning design and courseware development. Before AI, it took at least seven working days to draft the script of a video lesson. 

Now, a good draft can be created within half a day, using the right data inputs and prompts. We follow the same principles of content development as in the pre-AI era – human oversight is still involved in the quality and accuracy of content produced.  

• The coordinates that reorient learning in an AI-enabled world
• ‘The more I improved my instructions, the better they worked as AI prompts’
• A ‘buffet’ model of faculty development to support evolving use of AI

Our model divides the labour of two distinct activity systems, academic staff and instructional designers, and uses the learning content created as the mediating artefacts. Our role as instructional designers means we maintain oversight of the learning design, and content production and academic staff, who are the subject matter experts, validate the content produced.

Speeding up course mapping without losing rigour

The backward instructional design process begins with our understanding of the intended graduate outcomes, as articulated by competency statements. With the aid of AI tools, we analyse, deconstruct, resequence and chunk existing content, constructing the learning outcomes the course design will work towards. The tool identifies any content gaps. These are then produced as a report and course map for human validation. 

In an engineering bachelor’s programme that was recently transformed from a traditional curriculum into a competency-based, microcredentialled, modular, stackable programme, we took existing content such as presentation slides, scripts of video lectures, PDFs of notes, handouts, journal articles and book chapters, and loaded them into the AI tool, grounding it with relevant data. 

Next, we prompted it to analyse and check for alignment to the new course profiles, again loading data that contain revised enabling and terminal learning outcomes (or competency statements), to produce the first draft of a report and course map for stakeholder analysis. The report and course map is the blueprint of the programme, containing key information such as learning outcomes, prior knowledge requirements, syllabus, learning activities and assessment structure. The draft is then validated and, if necessary, corrected by our team for pedagogical alignment and the academic staff for relevance to the subject matter.  

Before AI, this process was manual and gruelling, taking at least three weeks of sequential tasks and multiple iterations between the instructional designers and academic staff. In the AI era, we can perform sequential tasks simultaneously and draft a report in less than 30 minutes, significantly reducing the iterations and time taken. This ground-prompt-produce-human validation process is a workflow that benefitted our engineering programme. It can be applied effectively to other fields of study such as humanities, social sciences or business. Humans always validate the output. 

Human judgement in the loop

Based on the validated course map, we ideate and draft storyboards. AI can integrate these tasks, enabling us to speedily tweak, enhance and better align the learning activities. We then send the storyboards to the subject matter experts for validation and final corrections before the courseware production commences. Because the academic staff own the final mile course delivery to the learners, they must know the direction of their courseware and this iteration also keeps them informed of the progress of development.

What instructional designers do well

At the course-mapping and storyboarding phases, instructional designers design the pedagogical or andragogical structure, focusing on critical elements that AI cannot fully conceptualise or align. We check and tweak AI output, ensuring the course design and learning activities are scaffolded and constructively aligned to the learning outcomes. This ensures that content, activities and assessments lead to the competencies being taught.

We also use our human empathy to exercise judgement and mediate the cognitive loads AI places on learners. After all, AI is a “word calculator” that gauges the next most likely word to appear, leading to the output of plausible nonsense. It has no human empathy nor sensing of emotions experienced by our learners or instructors.

Bringing relevance forward in the design process

We use higher-order questions and real-life examples to relate the learning content to the workplace in the real world. Traditionally, we would conduct research for realistic exemplars during the production phase, and deem it optional learning. A competency-based pedagogy must be relevant to the workplace and our students’ future working lives. 

With AI, our research is brought forward to the storyboarding phase, providing stakeholders with a more holistic, detailed and competency-based view of the course early in its developmental phases. This frees up time, which is then is put to better use for both instructional designers and academic staff. Both teams can sharpen the final learning experience and go deeper into application-based activities – improving the quality of the teaching and conditions for learning.

Producing e-learning objects with AI-enabled tools with human oversight

When everyone involved agrees on the storyboard, we proceed to producing e-learning content or e-learning objects. Traditionally, we used authoring tools such as Articulate Storyline, H5P and Adobe Premier Pro to create content from scratch.

Today, these authoring tools are augmented by AI and complemented by an ever-increasing range of text-to-graphics, text-to-voice and text-to-video tools, reducing time taken to create, produce and edit. Staple tools such as Microsoft Office 365 (Enterprise version), also augmented by Copilot, have enabled us to create narrated video slideshows from source MS PowerPoint files, with a few clicks of the mouse. 

But AI hallucinates, so human oversight and quality checks are built in to the process. Often, this involves editing output errors: from typographical errors in AI-generated graphics, mispronounced words in AI-generated voices and most importantly, ensuring no imagined facts.

During our engineering courseware development, the instructional design team spotted mispronounced scientific names made by the AI voices in videos and corrected them. Academic staff spotted wrong engineering calculations and imagined scientific concepts, and corrected them. Both teams essentially collaborated and co-created the content. This forms the basis for our ground-prompt-produce-human validation workflow, where the fidelity of courseware is governed – this is what we call our GAI + Human Touch model.

Our biggest takeaway: AI enhances productivity and improves the content creation process. The time saved is redistributed to enhancing the value of what we’re teaching. All content must be validated by human subject matter experts because of the hallucinating nature of AI. The GAI + Human Touch model is our university’s way of delivering accelerated, high-quality, competency-based education.

Seet Kim Hwee, Monika Shukla and Nguyen Thi Tu Linh are instructional designers in the Digital Learning Experience team at Singapore Institute of Technology.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.