Classroom Questions

Questions for “Think twice before using ChatGPT for help with homework”

a side profile of a young man sitting in a classroom taking notes at a desk. He is looking at another young man next to him, who is also writing. Overlaid over the second student are animations suggesting computing, algorithms and artificial intelligence

ChatGPT is impressive and can be quite useful. It can help people write text, for instance, and code. However, “it’s not magic,” says Casey Fiesler. In fact, it often seems intelligent and confident while making mistakes — and sometimes parroting biases.

Glenn Harvey

By Science News Explores

To accompany feature “Think twice before using ChatGPT for help with homework”

SCIENCE

Before Reading:

1. Describe two ways you use the internet to help you with homework. Do these ways help you find answers, understand answers, or both? Do you think using the internet for homework helps you get better at doing homework or not? Explain your answer.

2. Once you find information on the internet, how do you verify whether it is accurate or not? Why is it this important?

During Reading:

1.  What is ChatGPT? What are three things it can do?

2.  How has Brett Vogelsinger had his students use ChatGPT? What does he see as two potential advantages to using the tool?

3.  What are two drawbacks other educators and schools have expressed about students using ChatGPT for schoolwork?

4.  What task did Xiaoming Zhai try with ChatGPT? What aspects worked well? What aspects did he end up doing himself?

5.  What does Casey Fiesler mean when she says ChatGPT is “confidently wrong”? How is this different from asking a person for information they may or may not know?

6.  Why is ChatGPT limited to information from 2021 and earlier?

7.  What is a language model? How does it work?

8.  Why does Emily Bender say ChatGPT users are “working for OpenAI for free”? What do users provide that AI tools need?

9.  In Catherine Gao’s study of AI-generated article abstracts, how often were reviewers fooled into thinking fake abstracts were human-written?

10. How do some AI-detection tools work?

After Reading:

1.  As described in the story, the GPT3.5 language model has been trained to not return biased, hurtful or inappropriate responses. Who decided for the model what to consider as biased or inappropriate? How might this affect the responses it gives? Write out one question that two people with different points of view on a topic might answer very differently. How do you think ChatGPT might answer the same question? Explain your answer.

2.  Think about how you might find a piece of information in a book, such as an encyclopedia, and how you might find the same information on the internet. In what ways do the processes differ? Are any aspects the same? Now think about how you would start a writing project, such as an essay. How would you do this on your own? How might you use an AI tool such as ChatGPT? In what ways do the processes differ? Are any aspects the same?

3.  High school sophomore Avani Rao says she does not think AI will replace human-written text because people will still want to write on their own. Do you agree or disagree with this point of view? Why? What does the experience of writing offer that prompting AI to generate text does not?