ChatGPT in Medical Librarianship: Update

(Michelle’s words) I know I said I would write a post once a month, but I wanted to do an update since several comments generated some interesting questions and ideas. Specifically, someone asked if the ChatGPT response could provide links to the examples in its post. I was intrigued and I wanted to see if the links suffered from AI hallucination like ChatGPT’s reference.

Here are the results:

  1. First time ever (screen shot below) ChatGPT3.5 gave me two side by side answers and asked which one I liked better. My guess is to help it to learn.
  2. Second, the answer differs some from the answer given in my original post. This is not unusual. You can ask it the same thing with the exact same wording and it will provide a different result each time.
  3. HOWEVER, my prompt to ChatGPT was different than the one I posed for my original post because I asked it to provide links to examples. So naturally it will be different. ChatGPT does not take its original answer and tweak it, like adding links. It often creates a new answer
  4. The links (copy paste snippet from CHATGPT3.5 below image) go nowhere. I clicked on them within ChatGPT and I got a 404 error. So the AI provides links but the links do not go to any real sites.

So I asked the question to ChatGPT as second time to see what changes would happen.

  1. I didn’t get the side by side 2 answers…weird.
  2. Content was different (as expected)
  3. Links changed and work…sort of. They go to actual websites and I don’t get a 404 error but as you read they really are sucky examples. I want the URLs for the 1st try Response 2 to work, not links to regular websites that are involved in AI.

So I tried for a final 3rd time to get the results I wanted. It is Halloween, I have candy to handout and beers to drink….no 4th try.

***PLEASE NOTE*** getting the results you want is as much a part of what you write in the prompt as it is ChatGPT interpreting what you wrote. ***I can’t stress this enought***

So here was my 3rd try prompt:

write a short blog post on the use of ChatGPT in medical libraries and provide examples of ChatGPT use in medicine with URLs to those examples

My Results:

  1. It did better on the linking. All but MedGPT’s link worked and went to real pages.
  2. Examples were meh and possibly not relevant. For example, it listed UpToDate and how it could benefit from ChatGPT. That was not what I wanted, I wanted specific examples of it use not what can benefit from its use. However…. ChatGPT interpreted my prompt differently.
  3. Please note, it did not provide linked examples for medical libraries, because I messed up in my prompt and wrote, “and provide examples of ChatGPT use in medicine with URLs,” so it gave me exactly that.

(Screen shot of ChatGPT answer with 2 responses side by side)

    (1st try Snippet of ChatGPT Response 2 words copy pasted: note links do not go anywhere)

    Examples of ChatGPT in Action

    To witness the transformative potential of ChatGPT in healthcare, here are some examples:

    1. Clinical Decision Support: Link to Example
    2. Patient Education: Link to Example
    3. Medical Research: Link to Example
    4. Language Accessibility: Link to Example
    5. Telemedicine Integration: Link to Example

    These examples illustrate the tangible benefits of combining ChatGPT with medical libraries, ultimately advancing patient care, research, and healthcare accessibility.

    Conclusion

    The partnership between ChatGPT and medical libraries is poised to reshape the healthcare landscape. By harnessing the power of AI-driven language models, healthcare professionals, researchers, and patients can access, interpret, and utilize medical knowledge more efficiently and effectively than ever before. As this collaboration continues to evolve, we can anticipate even more innovative applications that will elevate the quality and inclusivity of healthcare services globally.

    (2nd try snippet of ChatGPT… links work but don’t add value IMHO)

    Conclusion

    The integration of ChatGPT with medical libraries holds immense promise in revolutionizing the healthcare information landscape. By simplifying information retrieval, offering contextual recommendations, and summarizing complex research, ChatGPT empowers healthcare professionals, researchers, and students to make informed decisions and advance medical knowledge. The combination of AI and medical libraries represents a significant step forward in the quest for better healthcare and medical breakthroughs.

    Links to Examples:

    1. Demo of ChatGPT: Explore the capabilities of ChatGPT in natural language understanding and generation.
    2. PubMed Central: A well-known medical library where ChatGPT integration could enhance information access and retrieval.
    3. ArXiv: A repository for academic papers, including medical research, which could benefit from AI-powered search and summarization features.
    4. Google Scholar: A popular tool for academic research, which could potentially integrate ChatGPT for improved search and recommendations.

    (Snippet from my 3rd and final try with ChatGPT and linking)

    Examples of ChatGPT in Medicine

    1. Medical Literature Summarization:
      • URL: MedGPT
      • ChatGPT can be employed to summarize lengthy research papers or clinical studies, helping healthcare professionals and researchers quickly extract the key findings and implications from complex texts.
    2. Diagnostic Support and Medical Queries:
      • URL: AIDoc
      • AI-powered systems like AIDoc use ChatGPT to assist radiologists and doctors in interpreting medical images and answering specific questions related to patient cases, potentially leading to faster and more accurate diagnoses.
    3. Clinical Decision Support:
      • URL: IBM Watson for Oncology
      • IBM Watson for Oncology employs ChatGPT-like models to provide personalized treatment recommendations by analyzing a patient’s medical history and comparing it with an extensive database of medical literature.
    4. Medical Knowledge Retrieval:
      • URL: UpToDate
      • Platforms like UpToDate can benefit from ChatGPT’s natural language search capabilities, allowing healthcare professionals to find relevant clinical information quickly and efficiently.

    Benefits of ChatGPT in Medical Libraries

    The integration of ChatGPT into medical libraries offers several significant advantages:

    • Simplified Search and Retrieval: ChatGPT enables users to search for medical information using natural language queries, reducing the need for complex search syntax.
    • Contextual Insights: ChatGPT can provide context-specific recommendations and insights by understanding the user’s inquiries.
    • Multilingual Support: ChatGPT can assist users in multiple languages, making medical literature more accessible to a global audience.
    • Quick Summarization: ChatGPT can summarize extensive medical documents, making it easier for healthcare professionals to extract pertinent information rapidly.

    (Michelle’s words)

    Keep the comments coming. I promise due to work life balance I am really only going to try and publish once a month. I swear…I think…I will try really really hard.

    5 thoughts on “ChatGPT in Medical Librarianship: Update”

    1. Coming from background of decades in both public & medical libraries (several more of the latter than the former; *&* being a “professional skeptic” in general & of new tech in particular!), I’m not surprised that the jury is definitely still out on implementation of AI in our biz.

      What amuses me most is the fact that it significantly increases the already critical importance of asking the right questions in the “reference interview” of ChatGPT in order to make sure we elicit the most useful information in order to be able to provide the actual human patron with the answer they actually need! Thanks bookoodles for all your work, Michelle!

    2. Yes. IMHO I think it depends on how much time you save. Believe it or not both the first post and the second post I wrote took me less than an hour. My posts used to take me a lot longer than that to do. It definitely helped save me time when I was writing a professional and technical email about library stuff to an important non-librarian. When writing those types of emails I spend a lot of time de-jargoning the library terminology so as to get my point accros in a professional but understandable way. I used ChatGPT to write the email. The email was entirely too long but I cut and pasted parts of it when I crafted my own and it made the process pretty quick and easy.
      Right now in its current iteration, I wouldn’t spend the time on ChatGPT for medical searching. BUT be wary when we talk about amount of time to get results because there are ton of MEDLINE searches that I have spent hours on within PubMed or Ovid MEDLINE trying to get good results. I think time spent vs output received is important but not the only marker.
      Some have emailed me to look at Elicit, Lori, Perplexity as AI’s to think about for search PubMed. One person said they used http://www.perplexity.ai on medical topics and it reference some articles from PubMed. It still has paywall problems too but those are ones to try and see if input time vs output results is worth it.

    3. Thanks, Michelle. So what I’m reading/getting from this is that *shocker* you still have to spend time creating a well-formed and precise question in order to get the best results from AI. (Similarly to when using dumb data ases and search engines…)

    4. This may seem like a very cynical comment, but if it takes so many tries and so much tweaking of the question, it seems like it would be almost easier to just write the post yourself! Haha!

    Comments are closed.