From clowns to clarity: How NotebookLM elevates student research with accurate AI assistance
From clowns to clarity: How NotebookLM elevates student research with accurate AI assistance
An AI course bot that knows everything that you want it to, and nothing extra!
One of the (many) problems with generative AI is how it can make up just about anything to keep you happy, a bit like a kid's party clown juggling flaming bowling pins while blindfolded—hoping you'll be impressed by anything that doesn't end in a disaster.
When our students have used AI to help with research project scopes or write a report section, they often skip essential information that has been intentionally clear in course readings (e.g., just using a dictionary.com reference to define their methods). On top of this, it is also very obvious where they have 'inserted' information from generative AI that isn't accurate; it just sticks out like a sore thumb - perhaps burnt on the hot end of a flaming bowling pin. While consistently encouraging generative AI use, we also want to ensure our students were using it critically with accuracy and clarity. With that in mind I made an interactive course helper bot using BotPress that aimed to collate all of the readings and resources for the methodology course I'm currently teaching. The idea was that the bot could be asked any question based on those resources - without using open access to the internet. Without these distractions and further opportunities for hallucinations, the students could deepen their knowledge of the specific readings and use that to support the research project scopes or project reports. However, after running the bot for over six months, the analytics tell me that there have been about ten uses of my friendly AI bot helper, and I'm confident that a few were mine!
So, when I saw a demo of NotebookLM a couple of weeks ago, I was excited to try a different approach. NotebookLM is an AI-powered research tool from Google that helps users generate insights and summaries from their own notes and documents. It seemed like more of a student-driven version of what I was trying to achieve (AI-supported accuracy and clarity in interactions with course content). Instead of me choosing the resources that the AI has access to as its knowledge base, NotebookLM allows the student agency by incorporating an easy interface that either links to a Google Drive or supports uploaded documents (such as their course notes and resources).
To test it out, I uploaded the OERs that we use for course content, as well as the course lecture slides and notes, to NotebookLM and then created some key notes that would be useful for students. From that point, I could also choose to share the notebook, and students would be able to interact and chat with the resources in their copy of the notebook, much like how you'd copy a Google Doc, not just comment on the same one.
When I originally proposed using the BotPress bot, I was aiming to create a 'course assistant,' as highlighted in Boston College's useful role descriptors.
Now, after using NotebookLM for a while, I can see how it can better align with Boston College’s initial descriptors. Depending on how students choose to engage with the tool and the sources they upload, it can offer a combination of roles supported by transparency, scaffolding, cognitive load, and metacognition. *Update: The new podcast feature will support students (and anyone else using the tool with cognitive load - being able to hear a ‘humanised’ audio version of the content in an AI-generated podcast makes the content accessible in an output that is usually under 10 minutes.
The tool can primarily act as an equitable and transparent course assistant, helping students understand how a course works by enabling open chat access to course guidelines, notes, and assessments and supporting the processing of multiple texts and readings. Once students develop their notes, NotebookLM provides them with specific prompts to create study guides from curated resources.
If NotebookLM can access the relevant readings and the assessment rubric or marking schedule, it could also act as an assignment tutor by scaffolding submissions (see below).
Once the student has an idea about the assessment structure, NotebookLM can also support them as a process coach by providing step-by-step guidelines on how to begin the submission and subsequent steps. The new ‘podcast’ feature here is going to make a big difference in summarising large amounts of information before the student begin digging deeper.
If the student uses the custom note function, NotebookLM can also serve as a reflective guide to prompt questioning and reinforce learning. One thing I like about this is that you can pin notes to the top of the interface, unlike on ChatGPT where you chats, and the important parts within them just disappear into the mountains of text that it creates.
My favourite thing about the tool is that it will say it doesn't know the answer based on the sources provided, therefore not setting the dreams of a concise submission up in smoke, unlike our other fire-juggling generative AI tools. However, with the new audio feature, the audio content may have inaccuracies - so perhaps steer clear if you’d like to keep a level of accuracy.
You can try the tool out for yourself now, as it seems like most people have access to it, but I made this short video before it was available just to show the capabilities. It would be great to hear if anyone else has any other ideas for how it could be used. If you are looking to try out the new podcast feature then you will need to start a new notebook as I couldn’t manage to make it work on any of the existing notebooks that I had.
Using Notebook LM to Enhance Teaching and Learning - Watch Video
Overall, I can see that more organisations will move to a more customised walled garden to support the safe and ethical use of generative AI. However, this is a great first step to experiment with how it may look, and to my main point, I think this is an excellent study tool for students.
This post was written by me, a human - however I asked ChatGPT 4-o to come up with a title and I thought it was effective so used it!