Directions for teachers: Ask students to read the online Science News Explores article “Think twice before using ChatGPT for help with homework” and have them answer the following questions. A version of the article, “Homework Help? ChatGPT and other AI tools are poised to disrupt education,” appears in the April 8, 2023 issue of Science News.

A special thanks goes out to the Science News Learning teachers who sought student responses to the questions used in the ChatGPT quiz. Some of the students’ answers were used in the quiz and are highlighted in the Science News articles about ChatGPT. Although there wasn’t room to publish every student response, each response was reviewed and supported the project. To stay informed about future opportunities like this one, check out the weekly Science News Learning Friday newsletter. You can receive the newsletter by signing up for our resources.

1. What is ChatGPT, and what can it do?

ChatGPT is an artificial intelligence (AI) chatbot that will respond to prompts with answers. This chatbot can write essays, summaries and stories that sound natural. It also can write code.

2. What challenges does ChatGPT cause in education right now? What actions have schools taken and why?

Educators and schools haven’t had time to develop protocols for using ChatGPT to best support student learning. Educators are concerned that students might use ChatGPT to cheat on assignments and tests or might submit ChatGPT writing as their own, without citation. The educators also believe students might not learn how to solve problems or think critically if they depend on ChatGPT. Because the chatbot makes mistakes, educators worry that students will produce inaccurate work. These concerns have led some schools and teachers to ban ChatGPT.

3. According to the article, what are some ways ChatGPT can benefit students?

One teacher quoted in the article suggested that ChatGPT can be used to coach students as they learn to write. For example, students could pick and choose from ideas generated by ChatGPT to develop their writing projects. ChatGPT could be helpful to students working in a second language, and the chatbot could also tutor in difficult subjects.

4. What are some limitations of ChatGPT? What can humans do that ChatGPT can’t?

Depending on the training data, what ChatGPT knows might not be current. ChatGPT also makes up information and makes up references; it cannot be trusted to present facts. Unlike humans, ChatGPT can’t reason or have thoughts and ideas of its own.

 5. Often ChatGPT generates the wrong answer to a prompt. What causes this AI tool to make mistakes?

ChatGPT makes mistakes because it does not pull facts from a database, and it combines language without understanding what the words mean.

6. How did OpenAI train ChatGPT?

OpenAI used a large language model to train ChatGPT. The model relied on an artificial neural network that looks for patterns in data and learns to predict what order words are likely to appear in a sentence. Books, journals, dialog, essays and exams were used in ChatGPT’s training. More recently, OpenAI used a different kind of machine learning called reinforcement learning with human feedback to fine-tune ChatGPT.

7. Find an example in the article where people were fooled by ChatGPT.

Catherine Gao, a medical researcher at Northwestern University, asked reviewers to distinguish between real scientific abstracts and ChatGPT-generated abstracts. She found that reviewers thought one in three abstracts generated by ChatGPT had been created by a human.