Fact-Checking and Biases
This module summarizes ways that generative AI tools can spread incorrect information, create fake citations and links, and generate biased output. It also provides strategies for fact-checking the output of AI tools. The audience is all students, including both AI novices and those who use tools like ChatGPT frequently. You can access the text and key pieces of the module below. You can also use a Canvas module that can be imported into course sites.
• Understand the strengths and weaknesses of using generative AI tools in the research process
• Understand how biases and misinformation are incorporated into AI tools
• Identify examples of generative AI tools creating incorrect output or “hallucinations”
• Identify strategies for fact-checking the output of generative AI
• Misinformation – False or inaccurate information that is spread without an intent to mislead, though its effects can still be harmful
• Implicit Bias – automatic, unconscious associations and stereotypes that individuals make about people or groups
• Scraping – Using software to automatically collect large amounts of text and data from websites, which can raise ethical concerns if done improperly. Scraped data is then used to train AI models.
• Disinformation – False or inaccurate information information that is purposely spread with the intent to mislead
• AI Hallucinations – When AI models produce completely false or imaginary output; these hallucinations are usually presented with confidence, even when obviously false
We will begin with a reading and a quiz. As you read, think about the concerns that are raised and how they relate to the intellectual work and value of universities. The module also contains a video with guidance on identifying misinformation and biases. There is an exercise asking you to practice and reflect on engaging with AI-generated information. And there is an opportunity for you to use AI to look for sources and develop a bibliography. Select the Next arrow below to get started.
Reading and Quiz
In this activity, we will learn why experts are concerned about the potential spread of disinformation via tools like ChatGPT.
Begin by reading the brief New York Times article, “Disinformation Researchers Raise Alarms About A.I. Chatbots”: https://www.nytimes.com/2023/02/08/technology/ai-chatbots-disinformation.html
After you’ve read the article, complete the quiz below.
Why are researchers more concerned about ChatGPT’s ability to spread disinformation as compared to previous AI chatbots like Microsoft’s Tay?
It’s slower but more accurate
It’s free for anyone to use
It’s more sophisticated, convincing, and powerful (correct answer)
It uses incorrect grammar to sound more informal
In the example where NewsGuard asked ChatGPT to write conspiracy theories and false claims about vaccines, what were the results?
Confident but incorrect responses
Fake citations to made-up studies
Additional false information not included in the original prompt
All of the above (correct answer)
True or false: OpenAI, the company that developed ChatGPT, uses both computer-based and human monitors to moderate the tool and try to prevent harmful output.
True (correct answer)
For this activity, you will watch a video, and then complete an exercise. Begin by watching the video below:
After you have watched the video, compete this exercise. After you complete the exercise, you can submit your work below.
- Using a text-based generative AI tool like ChatGPTLinks to an external site., ClaudeLinks to an external site., or BingLinks to an external site., type in the following prompt: “Summarize recent research on [health topic of your choice], using peer-reviewed research articles.”
- Try to open the sources the AI tool cites by copying and pasting the titles into Google ScholarLinks to an external site. or Articles+,Links to an external site. databases provided by UNC Libraries.
- If the AI tool doesn’t provide citations to sources, prompt it with, “Can you link to where you found this information?”
- Capture the results of your chat session. You can either copy the link to the session or download the transcript.
- Reflect on your experience:
- What types of sources did the AI tool provide (for example, peer-reviewed research articles, government websites, news articles, etc)?
- Did you discover any false information, fake citations, or other hallucinations?
- Did the AI tool include any warnings about its limitations or “guardrails” about how it should or should not be used?
Submit the assignment using the text box below to complete this step of the module.
Reflect and Share
This activity will prompt you to reflect and share your experiences using AI. Complete the following steps:
- Pick a research topic that you know very well and could be considered an expert in. This could be an academic or personal topic of interest.
- Go to an AI Chatbot (e.g., ChatGPTLinks to an external site., BingLinks to an external site., ClaudeLinks to an external site., etc.) and prompt the AI to create a bibliography of 5 important texts or sources in your area of expertise.
- Verify that the sources exist. Using your own knowledge, search tools like Google,Links to an external site. or by searching the Library websiteLinks to an external site., double-check that these texts are real.
- Use AI to create citations. Prompt the AI tool to create citations for the 5 bibliography entries in several different citation styles, such as MLA, APA, Chicago, or Council of Science Editors. Verify that the citations are formatted correctly using the UNC Libraries’ Citation GuideLinks to an external site. or tools like the Purdue OWL.Links to an external site.
- Create an annotated bibliography. Ask the AI to format your sources into an annotated bibliography in the citation style of your choice.
- Reflect on your experience:
- What parts of the process did the AI tool do well? Where did it stumble?
- Did it miss any key texts that should have been included, or did it hallucinate any texts that don’t exist?
- Were the citations it created correctly formatted and in the correct order as required by your style guide (for example, alphabetical by author’s last name)?
- Did the annotations relate to or build upon each other in any way, or were they straightforward summaries of the texts?
To complete this module, submit your annotated bibliography and responses to the reflection questions using the options set up by your instructor.