Assignment Description

Reading academic science articles can be challenging for non-experts. However, being able to access the information communicated in academic journals is important for staying informed, educating oneself on important and controversial topics, and conducting personal research projects. While many scientists work hard to make their research accessible, being able to read academic science articles for ourselves can be a valuable skill. While one response to the challenge posed by scientific jargon, lengthy articles, and unfamiliar concepts is to say that practice makes perfect, it’s worth considering another question: are there tools that can help non-expert readers translate such texts?

In this assignment, we will experiment with using Generative AI (GAI) as a tool for reading. The question we will aim to answer for ourselves with this experiment is whether AI can (and should) help us read an academic science article. 

During this activity, keep in mind our conversation in class about how Large Language Models (LLM) work to generate answers (and the similarities between Chat GPT and a Blurry JPEG!). For review, watch this 5-minute video: “What are Large Language Models?” 

Proceed through the following steps using this journal article as your test subject: “Evidence that Magnetic Navigation and Geomagnetic Imprinting Shape Spatial Genetic Variation in Sea Turtles.”

Instructions

Step 1) Access and familiarize yourself with Microsoft Co-Pilot.

Click here for instructions for accessing Microsoft Co-Pilot as a UNC student.

Familiarize yourself with Microsoft Copilot’s Data, Privacy, and Security Policies and Features

Note: Microsoft Copilot, formerly called Bing Chat, is powered by GPT-4. If you prefer to use ChatGPT for this assignment, feel free to do so, though keep in mind that ChatGPT currently uses GPT 3.5. While Microsoft AI may be more secure in some ways than ChatGPT, always avoid feeding AI programs information or data you’d be uncomfortable sharing.

Step 2) Feed the program the information it needs.

You will likely find that the AI program’s character limit prevents you from feeding it the entire article you are trying to read. While there are several ways to respond to this challenge, one good option is to feed it sections of an article at a time and ask questions focused on that section before repeating the process with another section. You are encouraged to be creative with how you obtain the information you need from the program to comprehend the article.

In a separate document, record the steps you take during this experiment. You will submit these notes as a homework assignment on Canvas.

Step 3) Obtain the information you need to comprehend the article by asking a series of questions.

For the sake of this experiment, please obtain the following pieces of information: 1) a summary of each section of the article in language you feel you are able to easily read 2) explanations of any words or concepts you do not understand 3) answers to at least two other questions you have about the article.

Copy these items to your notes document along with the questions you asked to obtain this information.

Step 4) Evaluate the accuracy of the information provided.

Compare the summary you obtained with your classmates’ versions. Work together to see if you can identify any information provided that doesn’t feel quite right in your summary or others’. 

Highlight in yellow (on your separate notes document) any information that you feel is less than satisfactory, leaving a note or comment explaining why. 

Step 5) Individually, gather your thoughts and submit your notes on Canvas.

Write a one-paragraph reflection on this experiment. Consider addressing some of the optional questions below:

• After conducting this experiment, do you think AI can help read/translate difficult texts? Why or why not?
• What do you think are some of the risks of using GAI as a tool for reading difficult texts?
• Would you consider using GAI to help you read difficult texts a “shortcut?” Why or why not?
• What advice would you give someone wanting to use AI to help read an academic science article.
• In what ways was Microsoft Copilot more or less helpful than the Google search feature?
• What future study might be necessary to better determine whether AI can be helpful for reading difficult texts? Are there other AI programs that might be a better fit for this experiment?