Do you sometimes struggle to get the best out of AI tools? Or are you sceptical about how they can help you in your role? A ‘human in the loop’ approach can help bridge the gap between AI's capabilities and your needs.
What is ‘human in the loop’?
The term human in the loop (HITL) has been popping up a lot recently on social media. Put simply, HITL is ensuring human expertise and judgement are present in AI processes. Though AI has amazing capabilities and can surpass humans in a range of tasks, it’s important to understand the limitations and where human intervention is needed.
Being the ‘human in the Loop’
Being the 'human in the loop' is about using your expertise to complement AI’s vast knowledge and computational power. AI can help you with a range of learning design tasks from analysing learner data, to creating scenarios or designing assessments. Yet despite its range of applications, your involvement helps ensure learning experiences are accurate, authentic and relevant.
Working with AI tools, such as ChatGPT, is like having 24/7 access to a genius assistant, albeit one that lacks the understanding of the wider context or specific needs of your task/project. This is where your guidance comes in. Your research of the learner and organisation can help AI tools focus on what’s relevant, producing more tailored content. You can include details about the learner or organisation in your prompts, give it links to online information or upload files (such as documents or survey results).
AI can analyse data and generate content that would take humans days, weeks or even months! However, the quality of its output depends on the quality of your input. Vague prompts, inaccurate documents or incomplete datasets will not give you the complete picture or best results. Even with the right training and instructions, your judgement is needed to make sure AI content is appropriate to learners and upholds ethical standards. For example, a learning experience about communication skills should include diverse perspectives that represent your learners.
While AI can enhance the learning design process by offering new insights and efficiencies, it can sometimes provide information that might be misleading or incorrect. Although these ‘hallucinations’ can sometimes uncover interesting ideas, it may also land you in hot water — like the lawyers who were fined after submitting fake citations generated by ChatGPT in their court documents. Always check ai-generated content before sharing or publishing. My blog post, Responsible Learning Design with AI, explores important ethical considerations and tips on how to verify AI-generated content.
Once you have high-quality AI-generated content, it’s your expertise that decides how best to organise it. There are a number of decisions you need to make as a learning designer, such as the best medium to present the content, the way the content should be split up or how 'learning' should be measured. Though AI can assist with these decisions, such as recommending learner pathways, you are the ‘human in the loop’ who supervises how learning experiences are implemented and evaluated.
The Human-AI dream team
AI, with your guidance, can improve the efficiency and effectiveness of your learning design. But, you don’t have to be an AI expert to make the most of this AI-human partnership — you just need the right mix of human expertise and AI brains.
Below are two approaches to help you get the most out of AI tools, like ChatGPT or Google Gemini.
Approach one: AI as assistant
This approach is best for completing tasks, such as writing scenarios, assessments or scripts, especially when you have a good idea of what the final product should be like. To get the most out of the AI as assistant, follow these four steps:
Step one: Define what the task is and what ‘good’ looks like.
Take some time to consider the details of the task, the best way to approach it and what you'd like the final product to look like. Collate any useful data, documentation or information you could give the AI, such as good examples you've made for other projects.
Not only will this help you solve the task well, it’s useful information for the AI. This step doesn't need to take long and will guide the AI to better results, saving you time later.
For example, when creating an outline, you might define the learner, time/length, outcomes, etc., and find a good example.
Step two: Brief the AI.
Just like assigning a task to a colleague, you’ll get better results if the AI knows what it needs to do. This is where your information from step one comes in handy!
Your 'brief' could be a prompt, instructions for a GPT (learn more about GPTs here) or documents uploaded to train the AI. Include the following:
The role of the AI (e.g. ‘You are an expert instructional designer…’)
Context and details of the task
Instructions (what you want it to do)
It's worth spending a bit of time typing out this information in an online notebook or document before giving it to the AI, so it's easier for the AI to read and 'understand'.
You don’t need advanced prompting skills, just type it out clearly as if you were sending it to a colleague — though you don’t need to include niceties!
Check out my prompting guide for example prompts.
Step three: Iterate by reviewing and giving feedback
It’s common to include reviews and feedback in the learning design process, especially for complex tasks. Similarly, you may need to give feedback to the AI and regenerate until you get a result you're happy with.
Don’t give up on the AI after the first attempt or expect it to give perfect responses. As the ‘human in the loop’ you need to guide the AI. Then, check and edit its results.
Step four: Check the final product
As mentioned earlier, AI content can contain inaccuracies. So, it’s crucial you check the final product before sharing or publishing it. This is especially true for factual content.
If your published content is mostly AI-generated, you may want to inform the audience — my blog post, Responsible Learning Design with AI includes advice on labelling AI-generated content. I've included a note at the end of this post about how I used AI for this post.
Approach two: AI as brainstorming buddy
This approach is incredibly simple — just chat with the AI like you would a human! This approach is great when you need help generating or discussing ideas.
Try the prompt: “You are an expert…. I want to brainstorm/chat about ….”
Why not try out these approaches and create your own Human-AI dream team?
Below are the two approaches summarised on a downloadable image, to keep for later or share - right click or press and hold the image to download.
Want to learn more?
The You Are Not So Smart podcast episode 281 by David McRaney with Jeremy Utley, Kian Gohar, and Henrik Werdelin is an interesting discussion about using chatbots to generate ideas.
Credit to David Hopkins who I’ve seen championing the ‘human in the loop’ on LinkedIn.
Don't forget to subscribe by selecting the button in the footer below. And, follow me on LinkedIn for more tips and insights on using AI in learning design.
ChatGPT was used to help edit parts of the Being the ‘human in the loop’ section.
コメント