Skip to Main Content
University of Texas University of Texas Libraries

Medicine

Artificial Intelligence

Artificial Intelligence

Generative AI tools may be used for many tasks.

Simpler tasks include brainstorming ideas, creating an outline, generating learning objectives, identifying keywords and synonyms for searching in library databases, editing/improving one's writing, generating communication pieces, creating images, conducting data analysis and presentation of the data, organizing notes, creating a project management plan, creating slide decks and presentations, etc.

More complex tasks include searching for evidence-based information and summarizing this information in order to explore and learn about a concept or topic.

This guide will focus on guidance and tools for the complex tasks that could be especially useful for your self-directed learning in PILLARS, for example.

Generative AI for Learning

There are many generative AI tools, and they fall into two buckets: not-grounded and grounded.

  • Not-grounded - the tool relies on a finite set of training data.
  • Grounded - the tool is more reflective of the real world through continuous learning and updating of data. This is accomplished through integration with the internet, ensuring more up to date and diverse data sources. 

Here is a suggested list of grounded tools. These more assuredly provide accurate citations and links to credible scholarly literature and quality webpages; thus there is a marked reduction in hallucinations (i.e. made-up citations). 

It is important to read and understand each tool's policy regarding privacy, data collection and data protection.  Links to the policies can be found under each tool.

Elicit's Privacy Policy - https://elicit.com/operations/privacy

Review of tool: Kung J. Elicit (product review). Journal of the Canadian Health Libraries Association. 2023;44(1). https://doi.org/10.29173/jchla29657

Consensus' Privacy Policy: https://consensus.app/home/blog/privacy-policy/

Review of Tool: Uppalapati V, Nag D. A Comparative Analysis of AI Models in Complex Medical Decision-Making Scenarios: Evaluating ChatGPT, Claude AI, Bard, and Perplexity. Cureus. 2024;16(1):e52485. https://doi.org/10.7759/cureus.52485

Perplexity's Privacy Policy: https://www.perplexity.ai/hub/legal/privacy-policy

Scite's Privacy Policy: https://scite.ai/policy

General Guidance

As part of your self-directed learning, you draw on multiple sources of information including textbooks, journal literature, library databases, question banks, lecture notes, etc.  AI is another assistive tool for your self-directed learning. 

As with all your learning, critical appraisal skills are necessary when gathering information.  However or wherever you find your information, you must evaluate it, verify it with additional sources, put the information into context, and understand its relevance in the growing body of acquired knowledge.

With that in mind, browse the various guidance found below:

Be cognizant of the following challenges when using AI for learning:

Verify that the information and sources are accurate

  • Check that citations are accurate
  • Verify the content output with other credible sources
  • If there are no citations provided with the output, consider using a different AI tool

Additionally, to evaluate the output and the source citations provided by the generative AI tool, use the criteria below.  As you do, remember:

  • Each criterion should be considered in the context of your topic or information need. For example, currency changes if you are working on a current event vs. a historical topic.
  • Weigh all five criteria when making your decision. For example, the information may appear accurate, but if the authority is suspect you may want to find a more authoritative site for your information.
  • When in doubt about a source, talk about it with your professor or a librarian.

Criteria to consider:

  1. Currency: When was the information published or last updated? Is it current enough for your topic?
  2. Relevance: Is this the type of information you need (ex. a research study or scholarly article)? Is it related to your topic? Is it detailed enough to help you answer questions on your topic?
  3. Authority: Who is the author or creator of the information (can be an individual or an organization)? Are they an expert on your topic? Has the source been peer reviewed? Who is the publisher? Are they reputable?
  4. Accuracy: Is the information true? What information does the author cite or refer to?  Is this a research study with methods you can follow? Can you find this information anywhere else? Can you find evidence to back it up from another resource? Are studies mentioned but not cited (this would be something to check on)? Can you locate those studies?
  5. Purpose/perspective: What is the purpose of the information? Was it written to sell something or to convince you of something? Is this fact or opinion based? Is it unfairly biased?

Source: University of Texas Libraries. (2023, July 10). Evaluate Sources: Evaluation Criteria. Retrieved June 19, 2024 from https://guides.lib.utexas.edu/c.php?g=539372&p=6876271

A prompt is an input to an AI tool. For research AI tools, the prompt is usually a question, request, or topic.  Writing good prompts will facilitate good outputs from the AI tool.

Here is some guidance:

CLEAR: Framework for Prompting - this framework provides a standard method for composing effective queries in academic settings:

  1. Concise: brevity and clarity in prompts - be specific
  2. Logical: structured and coherent prompts - use a logical flow and order
  3. Explicit: clear output specifications - give precise instructions on your desired output format, content, and/or scope
  4. Adaptive: flexibility and customization in prompts - experiment with various prompt formulations and phrasing in an iterative process
  5. Reflective: continuous evaluation and improvement of prompts - adjust and improve your prompt by evaluating the previous responses

For more detailed explanations and examples, read the full article: Lo LS. The CLEAR path: a framework for enhancing information literacy through prompt engineering.” Journal of academic librarianship. 2023; 49(4). http://ezproxy.lib.utexas.edu/login?url=http://dx.doi.org/10.1016/j.acalib.2023.102720

Go to your UT LinkedIn account and search in the Learning section for "How to Research and Write Using Generative AI Tools: Meet your AI Creative Collaborator"[Video].

OpenAI Developer Forum

LearnPrompting. Prompt Engineering Guide. 2024. https://learnprompting.org/docs/intro

Prompt Examples
Bad prompt Good prompt Explanation for good prompt
What are the most effective treatments for diabetes? Create a concise summary of the key findings from studies published within the last 5 years on effective treatments for diabetes for middle age women.  Gives parameters on how the old the studies should be and the age of the patient group.
Tell me about the symptoms of Covid. List the key symptoms of Covid in bullet point format and include brief information about the nature and relevance of each symptom. Gives information about the format and length of output.
Describe the healing process after knee surgery. What can a patient typically expect during the first six weeks of healing in terms of pain, swelling, and mobility after knee surgery and what are some potential complications that may arise during that six week period. A specific period of time is given and a more detailed request regarding healing and complications is provided. (This may need to be divided into two prompts if the AI tool is overwhelmed by multiple facets in one prompt)
How does one conduct a neurological exam? Provide a step-by-step guide to performing a neurological examination on an adult patient. Gives the AI tool a task rather than asking an open-ended question.

More examples may be found in the following:

When incorporating information you receive through use of a generative AI tool, you need to indicate that you have done so and you should cite that tool.

People in the academic, research, and publishing worlds have begun to develop guidance for citing AI tools and AI-generated content. Below are some guidelines from academic and medical libraries:

  • cite the tool and its version (if known)
  • provide the actual prompt you used somewhere, such as in the text, in your citation, or in an appendix
  • provide the date that you used the AI tool 
  • indicate the sources of the information from which the AI tool based its response (good AI tools should provide their sources!) 

Here is some guidance from publishers:

Here are some citation examples:

Example 1

In text:

When prompted with "What are the common side effects of Erlotinib to treat lung cancer in an adult patient?", Microsoft CoPilot generated the following six conditions: rash, diarrhea, lost of appetite, weakness, cough, and shortness of breath. 1 

Reference:

1. Azure OpenAI. Microsoft CoPilot [Large language model].  Accessed June 13, 2024 from https://copilot.microsoft.com/

Example 2

In text:

According to RxList, Becker et al., Kiyohara et al., OncoLink, and MedlinePlus, the common side effects of Erlotinib are: rash, skin changes, nail changes, diarrhea, fatigue, loss of appetite, nausea, vomiting, cough, trouble breathing, and weakness. 1 

Reference:

1. "What are the common side effects of Erlotinib to treat lung cancer in an adult patient?" prompt. Perplexity (Perplexity AI Inc.). Accessed June 13, 2024 from https://www.perplexity.ai/

Example 3

Or, you can just use the AI tool as the conduit for finding the information, then go to the sources that are cited in the output, read those sources, and cite them. Describe in your work how you used the AI tool in your learning or research process.  

In text:

When researching adult lung cancer, the AI tool Elicit (https://elicit.com/) was prompted with the question "What is the typical life expectancy of an adult lung cancer patient who is being treated with Erlotinib?" Studies provided in the AI-generated output reported a median survival time of 6.7 months, 8.4 months, and 10.9 months. 1,2,3  These studies' participants were a median age, respectively, of . . . . [and so on].  Additionally, a review article analyzed the results of studies that looked at Erlotinib's effect on the primary outcome of survival for advanced, recurrent, and relapsed non-small cell lung cancer. 4  The authors concluded that . . . 

References:

1. Shepherd FA, Rodrigues Pereira J, Diuleanu T, et al. Erlotinib in previously treated non-small-cell lung cancer. N Engl J Med. 2005;353(2):123-132. doi:10.1056/NEJMoa050753

2. Pérez-Soler R, Chachoua A, Hammond LA, et al. Determinants of tumor response and survival with erlotinib in patients with non--small-cell lung cancer. J Clin Oncol. 2004;22(16):3238-47. doi: 10.1200/JCO.2004.11.057

3. Jackman DM, Yeap BY, Lindeman NI, et al.. Phase II clinical trial of chemotherapy-naive patients > or = 70 years of age treated with erlotinib for advanced non-small-cell lung cancer. J Clin Oncol. 2007;25(7):760-6. doi:10.1200/JCO.2006.07.5754

4. Feld R, Sridhar SS, Shepherd FA, Mackay JA, Evans WK; Lung Cancer Disease Site Group of Cancer Care Ontario's Program in Evidence-based Care. Use of the epidermal growth factor receptor inhibitors gefitinib and erlotinib in the treatment of non-small cell lung cancer: a systematic review. J Thorac Oncol. 2006 M;1(4):367-76.

 

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 Generic License.