fbpx

AI-Assisted Learning and Teaching

Artificial Intelligence in Medicine: AI-Assisted Learning and Teaching

One use of artificial intelligence in medicine that many people overlook is AI-assisted learning and teaching! There are several great ways to do this and we’re going to spill the details on precisely how to go about it for the following 4 use cases:

  • Using AI as a speed reading assistant, to help you can find information fast
  • Asking the AI to create flashcards for Anki, to help you learn new material
  • Enlisting the AI as an interpreter to help you quickly simplify complex topics, for yourself or your patients
  • Making the AI your teaching assistant when creating cases for your residents

Maximising potential of artificial intelligence in teaching and learning within medicine, while maintaining accuracy

An impressive example of artificial intelligence in medicine was when ChatGPT passed the USMLE exam! But just like humans, it sometimes gets things wrong. So, you need guardrails in place when you use it to assist you with teaching and learning (the number one most important one being fact-checking). 

But that turns out to be a beneficial feature, not a bug. 

Here’s why: 

Fact-checking turns your teaching prep or study session into an active learning process that can lead to better retention and recall of information. When you engage with the material by cross-referencing and verifying, it requires more cognitive effort than simply reading or listening. This type of critical thinking encourages a deeper understanding of the material. So, don’t view this as wasted time… it’s actually an opportunity to further enhance your understanding of the topic.

On that note, let’s now take a look at four ways to use artificial intelligence in teaching and learning:

1. Use AI as a speed reading assistant for long research papers (so you can find information fast!)

AI’s like Claude and ChatGPT can read PDFs, and this is very handy when it comes to long research papers. You can ask these AIs to do the following:

  1. Read the paper and answer your questions about it. 
  2. Summarise key points for you (handy if you’re preparing a presentation). 
  3. Ask it to quiz you on key takeaways in the study.

As always, this type of use case for artificial intelligence in medicine requires fact checking. So, the best way to go about this is to use an AI that will tell you which page number each bit of information is from. That makes it fast and easy to verify its accuracy.

ChatGPT

With ChatGPT, the most reliable way to accomplish this is by using a plugin for this. (You’ll need the ChatGPT Pro Plan to access their plug-in store.)

How to install a plugin for ChatGPT

AI PDF is a free plugin that we like. Search for it in the plugin store, then click ‘install’.

Screenshot of the Plugin store for ChatGPT, displaying plugins with names containing "PDF".

Next, exit the plugin store. Click the dropdown arrow next to ‘ChatGPT Plugins’, then select ‘Plugins’.

Screenshot of the ChatGPT desktop interface where you enable plugins from a dropdown menu.

After you do that, click the dropdown arrow next to ‘no plugins enabled’, and select ‘AI PDF’.

Screenshot of the ChatGPT desktop interface where you enable the "AI PDF" plugin.

Pro tip: If you’re unsure how to use a plugin, just ask! Here are the results of doing so with the AI PDF plugin:

Screenshot of the ChatGPT desktop interface with the AI PDF plugin enabled. The user asks it, "How do I use this plugin?" and ChatGPT generates instructions for use.
How to use the AI PDF plugin with ChatGPT

After you have AI PDF enabled, share a link to the PDF you want to discuss with it, then ask it whatever questions you’d like.

Some of the things we liked about using AI PDF include…

  • It’s fast.
  • It can review the entire PDF at once.
  • In our tests, it didn’t make any mistakes. (Nonetheless, we still recommend that you fact-check its output.)
  • It provides you with page numbers for any information it provides to you (thus making fact-checking easier).

For best results, give the AI context to help it better hone in on the answer. But for the purposes of this article, we wanted to see how the plug-in would perform without our help. So, we used a very basic question about this paper, with no context:

A screenshot of a user using artificial intelligence in medicine, asking the AI PDF plugin within ChatGPT questions about a PDF.

It mistook the title of Table 3 for the title of the entire document, but the rest of the information it provided was correct. 

ChatGPT’s Scholar AI plugin: another way to use artificial intelligence in medicine

As an experiment, we also tried the Scholar AI plugin. Its description says that in addition to being able to search over two-hundred million peer-reviewed papers, it can also explore text from scientific PDFs.

A screenshot of a user enabling the ScholarAI plugin within ChatGPT.

During our testing, this plugin was slower than AI PDF, and could only read smaller sections of the PDF at one time.

A use case of artificial intelligence in medicine: A screenshot of a user using artificial intelligence in medicine: they're using the ScholarAI plugin within ChatGPT and asking it questions about a research paper.

It took six requests to continue searching before it finally found the answer on pages eight and nine of the PDF. Having to go through this process might be okay for a short PDF, but would be frustratingly time-consuming with a longer one.

A use case of artificial intelligence in medicine: A screenshot of the ScholarAI plugin for ChatGPT and the results of it searching a research paper for specific information requested by the user.

Granted, when we asked ScholarAI how to use it, one of the options it mentioned was that we could use Zotero—maybe this process would have gone faster had we done so.

Screenshot of instructions regarding the option to use Zotero with the ScholarAI plugin within ChatGPT.

But given that AI PDF can handle this task without any 3rd party tools, we feel that it’s the winner for this type of AI-assisted learning.

What to do if the plugin you chose can’t read your PDF

When we tried to use AI PDF to read this paper, here’s what it said:

Screenshot of a failed attempt to get the AI PDF plugin within ChatGPT to read a pdf document.

So, we tried again, this time using the link that the plugin recommended, and this was the result:

A screenshot showing AI PDF unable to read a PDF, and instructing the user to open a support ticket at a link it provided.

We wondered if that meant that none of the AIs available to us would be able to read it. So, we tried Anthropic’s Claude. Here’s what happened:

A use case of artificial intelligence in medicine: A screenshot Anthropic's Claude AI successfully reading a PDF that a ChatGPT plugin couldn't.

You’ll notice that we even tried to mislead it with a leading question that implied the study was already done. But the AI caught that and made a point of telling us that the study wasn’t completed yet.

So, the moral of the story is that even if one AI tells you a PDF can’t be read, it could be worthwhile to try a different AI.

Another ChatGPT alternative: SciSpace

SciSpace allows you to interact with PDFs, using functionality that’s very similar to ChatGPT. It can answer questions about PDFs, plus, summarise and paraphrase PDFs.

But in terms of accuracy, it was hit and miss. Once again, we used the corneal pain paper, asking it the same question we asked the AI PDF Plugin via ChatGPT. Here’s SciSpace’s response:

A screenshot fo the SciSpace response to a user question about a research paper.

As you can see, it told us carbamazepine was a TCA (tricyclic antidepressant), which isn’t true. 

Secondly, SciSpace wrongly claimed the paper doesn’t specify which TCA is best (whereas what it actually says is that nortriptyline is preferred).

When you ask SciSpace a question about a PDF, in theory it can also tell you exactly where it found the information. However, on the day we tested it, even when it gave us accurate answers to our questions, it was unable to find the correct location of the answer within the PDF. Here’s an example using that BMJ paper about cognition after stroke.

A screenshot showing SciSpace correctly answering a question about a research paper.

It’s true that the main blood biomarker predictor in the study is NFL. But SciSpace was completely wrong about the portion of the paper where that information was found. Here’s where it claimed the info was from:

A screenshot showing the portion of the paper that SciSpace thought its answer came from.

As you can see, there’s no specific mention of NFL there. 

Despite the fact that SciSpace highlights the section of the PDF where it “thinks” it found the info within its answer, during our testing, none of the highlighted locations were correct. So, we had to manually find the info in the PDF using a basic text search (Control-F on a PC, or Command-F on a Mac), and do our fact-checking that way.

We used the free version of SciSpace for these experiments. That’s important context because the free version generates answers using their “Standard” AI model. Whereas if you upgrade to a paid plan, you’ll get answers from their “High Quality” AI model (That costs 20 USD per month on the monthly plan, or 144 USD per year on the annual plan.) In other words, compared to what we did, you might get better results with this use case of artificial intelligence in medicine if you use the upgraded “High Quality” model.

2. Use AI to create flashcards to study

Say you wanted to learn the info in this document from the American Association of Clinical Endocrinology and their Consensus Statement: Comprehensive Type 2 Diabetes Management Algorithm. You can use ChatGPT to create flashcards for you, which you can import into an app like Anki for further study.

Step 1: Get the info into ChatGPT.

Download the guidelines as a PDF, upload them into ChatGPT (use the paperclip icon), and double check if ChatGPT is able to read the PDF.

A use case of artificial intelligence in medicine: A screenshot of the ChatGPT4 interface. The user uploaded a PDFand asked ChatGPT if it would read it. ChatGPT replied with a summary of what it read.

Step 2: Ask ChatGPT to generate quiz questions for you in the format of your choosing.

A screenshot of a user asking ChatGPT to create flashcards to help them study the material contained in a PDF document.

Step 3: Ask ChatGPT to rewrite the questions for your favourite flashcard app. We chose Anki.

A screenshot showing the flashcard questions that ChatGPT4 created based on a PDF document provided by the user.

*After ChatGPT finishes generating the questions and answers, fact-check them against the original PDF document to ensure it didn’t make any mistakes.

Step 4: Ask ChatGPT to put the questions in a .csv format, so you can import them to your flashcard app.

A use case of artificial intelligence in medicine: A screenshot showing ChatGPT creating a CSV file full of flashcard questions that the user can download. ChatGPT provides the user with instructions on how to import the CSV file into Anki.

Step 5: Download the .csv file and import the flashcards into Anki or whichever app you like best.

Would you prefer a video tutorial? If so, Medmastery’s 14-lesson ChatGPT Essentials Course is exactly what you need, since one of its lessons goes over this topic in detail. 

3. Use AI to simplify complex topics

Say you come across something in a journal article that you aren’t quite sure you understand correctly. Try asking an AI like ChatGPT, Claude, or Bing Chat to explain it to you. 

For example, it’s common to come across statistical concepts within research papers. For most of us, the last time we set foot in a statistics class was years ago. So, we’re understandably rusty. Or maybe you never took a statistics class at all. Either way, whenever you come across something like that, it’s worth asking an AI for help.

And it’s not just your own learning that can benefit from this strategy.

If you’ve ever struggled to find the right words to explain a complicated concept to a patient, and you find that topic comes up repeatedly, here’s something to try: 

Ask an AI for ideas on how you could simplify the concept for a layperson. 

As the expert, you’ll instantly know whether or not the AI’s explanation is accurate, and if you’re satisfied with its wording, you can use it the next time the topic comes up with your patients.

4. Use AI to create cases for teaching students and residents

One use of artificial intelligence in medical education that’s sure to come in handy is using AIs like ChatGPT to help you create cases for medical students and residents. ChatGPT can be used to…

  • Organise your thoughts into structured materials
  • Develop key learning points tailored to clinical cases and learner levels
  • Assist in designing scenarios with specific objectives, introducing complexities, and suggesting debriefing points.

Through the ChatGPT Plus interface, you can also create custom GPTs that gather information for cases only from documents that you provide. To access this feature, select “Explore” from the left sidebar, then select “Create a GPT”.

A use case of artificial intelligence in medicine: A screenshot showing the screen where you can create a custom GPT within ChatGPT.

For a detailed real-world look into how to use AIs like ChatGPT to create teaching cases, be sure to take a look at our interview with ICU and ER physician, Dr. Sameer Shaikh

Embracing artificial intelligence in medicine, using a balanced approach to AI-assisted learning

As we’ve explored, AI offers a suite of tools that can enhance both learning and teaching experiences. Whether it’s speeding through research papers, creating flashcards, simplifying complex topics, or designing educational cases, AI can be a versatile ally. Just remember that the key to effectively using AI for these types of tasks lies in a balanced approach—embrace its strengths, but stay vigilant about its limitations!


References

Educational Resources

AI in HEALTHCARE

Want to become a pro at prompting, and consistently get usable results? Be sure to check out Medmastery’s AI prompting course. Learn techniques to apply to the plethora of AI resources in constant development.

Sheralyn Guilleminot LITFL Author

BSc.Pharm (University of Manitoba), Pharmacist and Medical Writer

BA MA (Oxon) MBChB (Edin) FACEM FFSEM. Emergency physician, Sir Charles Gairdner Hospital.  Passion for rugby; medical history; medical education; and asynchronous learning #FOAMed evangelist. Co-founder and CTO of Life in the Fast lane | Eponyms | Books | Twitter |

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.