Artificial intelligence is being used extensively in learning design workflows. You might be familiar with AI through the widespread use of the well-known AI platform, Chat GPT. It has become a valuable tool for learning designers, assisting with researching content topics and helping write and refine your learning content as you work.
When we now talk about AI, we are referring to what is called ‘Generative AI' - artificial intelligence that can generate content, i.e., words and images. The other well-known AI type, ‘General AI’ - think the classic Sci-Fi/Hollywood learning machines trope - is just a concept for now, not a reality.
Generative AI – what it can and cannot do
Generative AI is built to recognise and generate new patterns. It is exceptionally good at replicating human language because speech and text are largely made up of syntax and rules. Generative AI uses our own patterns of grammar and meaning to replicate information that it has been trained to draw from.
Generative AI cannot learn in real-time. Instead, it is trained once on one set of static data, largely from online sources. For example, ChatGPT 4.0 has a knowledge cutoff date of December 2023. That means the information that was put into that data set was all captured in December 2023. New ideas, concepts and knowledge from beyond the cutoff will not be included in any of its responses. Some Gen AI tools can find links, search the web and read your documents. They can generate content based on what they have been fed, but they cannot learn or make new connections about what you have told them.
When you use a tool such as ChatGPT, it can seem as if it has learnt from your interactions and provided a nuanced response, something with some depth to it. However, this is not the case. Generative AI can mimic nuance and intelligence in its responses to you but in reality, this is just pattern recognition, allowing it to make seemingly sophisticated connections and links between what you are asking and the concepts and ideas contained in the data used to train it.
Will AI take our jobs?
We don't believe generative AI will take our jobs, even in the future. Despite the hype, generative AI may have reached its peak form for now. ChatGPT’s current version is 4.0. Version 5.0, which has not been launched, will require five times the data that 4.0 required to train, and version 6.0 will require five times the data of 5.0. This is an exponential problem. We are already running out of fresh, non-AI-generated data to train on.
This brings us to an important realisation - generative AI, for all its sophistication, still needs human input to function correctly. It needs people to provide new ideas, concepts and data. As learning designers, we can seek help from generative AI, but we still need to apply good adult learning principles to its output, such as consultation with subject matter experts and supporting research from reputable sources.
Is Generative AI reliable?
Generative AI is unreliable. Sometimes it makes up things that seem true. It is trained on human data, so misinformation can be found in its responses.
It is notoriously bad at generating images of words and can be bad at mathematics. However, it is driven to always generate something and to do it in a way that appears authoritative, even if that something is not real or true. When AI makes up things that seem true, this is called hallucination.
Tips for AI accuracy
Tip 1: You can check your Gen AI tool’s accuracy by telling it not to lie. Tell it over and over again. This is surprisingly effective.
Tip 2: Use direct, unambiguous language. Avoid jargon and avoid words that could have multiple meanings.
Tip 3: Cross-check your answers, for example by using other AI tools and Google search. You can run the same prompt on search and on your Gen AI tools and cross-check the answers. Remember that occasional ‘spot’ checking is not enough, and you must always keep your eye on it. Even if something looks true and legitimate, always check it.
Tip 4: Change its perspective. Tell the AI tool in your prompt to provide the response from the perspective of a historian, for example. You could also ask it to provide differing perspectives or opinions based on the thinking of leading subject matter experts. Be aware that if there is not a good differing perspective or opinion available, it will sometimes make one up in an effort to satisfy your request.
Tip 5: Another way to get the information you want is to create a persona, providing the tool what it needs to know about this person and how it should respond. If you are writing about plumbing, for example, define your persona as a plumber, tell it about the kind of work they do, and to respond using clear, technical responses.
Tip 6: You might give the program sources that you trust and tell it to cite primary sources. You can tell it to spell out internet addresses in full so you can see at a glance where the information is from.
Engaging with AI – it's your choice
One of the best aspects of bringing AI into your workflow is the fact that you can choose the level that you wish to engage with it. While it is a powerful tool, it is not a one-click solution. Humans are still essential to the process, especially in learning design, as our guidance and experience are necessary to producing high-quality, accurate, and engaging learning content. AI can help – and it wants to – but we need to remember that we’re in the driver’s seat.
This blog is based on a July 2024 eWorks webinar hosted by eWorks Corporate Courseware Lead Jesse Harrison and Instructional Designer Rachel Thorne.
In the webinar, they demonstrated how AI could be used and you can view this in the webinar recording.
View the webinar here.
Looking to elevate your eLearning training with the eLearning experts?
Our dedicated eWorks team are here to help. Get in touch today at: sales@eworks.edu.au
