AI Tools for Research

AI is changing how students conduct research, even if they're not using these specific tools to automate parts of the research process. Any Google search now is preceded by an AI answer with links to various sources. 

Any use of AI in the classroom should come with direct instruction on how to evaluate that tool and to think critically about strengths and weaknesses. As we consider why we teach research, it is also important to find ways to use these tools to help with research, without replacing the actual thinking that students do when they're developing research.

AI tools can be helpful in the research process when students understand the strengths and weaknesses, and how to use the tool without replacing their own creative and iterative processes when conducting research.

Teaching Ideas for Finding Sources: 

1. Identify Hallucinations: Use ChatGPT with a prompt including language to provide 5 scholarly articles to answer the question. Have students get full-text access to the articles. Discuss any hallucinations or inaccuracies in the citations. Then discuss the response. Is there anything missing? Are these the best sources to answer this research question? Why or why not? How would you fact-check the information? 
    
2. Identify Search Result Bias: Using Google's Gemini, conduct two side-by-side searches using opposite search language (i.e. "Is advanced recycling good?" and "Is advanced recycling bad?"). Students evaluate the differences between the two AI summaries and top results returned. Reframe their searches using neutral language, or move into a source-evaluation lesson using the results the AI summary points to. Is the AI summary accurate to what the sources actually say? Are the sources themselves authoritative? 
    
3. Suggest Keywords and Research Strategies: Use ChatGPT with a prompt identifying a developed topic. Request that the tool provide suggested keywords to use in library databases to return scholarly literature on that topic. A follow-up might be requesting GPT to suggest specific search strategies or database names for research on that topic.
    
4. Narrow a Research Topic: Ask ChatGPT to provide multiple ideas for narrowing down a broad research topic. What are some areas of inquiry that could help shape a topic. Avoid asking ChatGPT to start the research topic idea itself. 

A major consideration when students are conducting research is whether they can evaluate sources appropriately. AI changes this conversation, and students will use the same information literacy tools to evaluate AI output. Below is a list of ideas for class activities to address issues with AI in research and AI-generated content as a source.

Teaching Ideas for Evaluating AI Output:

1. Identify Errors: Have ChatGPT generate text on a topic the students know well, or where inaccuracies would be easily identifiable an example might be "What are women's reproductive rights in the US" or another contemporary topic. Have students identify inaccuracies or generalities in the text, or omissions. In the example above, Gemini provided a response that indicated women's reproductive rights include "access to contraception, prenatal care, safe childbirth, and abortion."
    
2. Identify Authoritative Voice: Specifically looking at the authority of the text, have students evaluate, rate, or rank output based on their perceived authoritativeness. Discuss why and how they ranked the results this way, and discuss what about AI seemed authoritative based on other sources they were considering. 
    
3. Identify Bias: Ask ChatGPT to define terrorism and give five examples. Evaluate output, which will include almost exclusively islamic terrorists. Choose an AI image generator to show Egyptian students going to school. It will put them in pyramids and sands, but modern Egypt looks like New York. You can prompt it to see those kinds of biases. Or use Poe.com to generate an image of a diverse classroom. Discuss issues with the interpretation of diversity in the classroom and differences between different students' outputs. What is missing? [Ideas drawn from Dr. Maha Bali in the Teaching in Higher Ed Podcast]