Using Character Chatbots in English Class

In October, students in my freshman English class finished reading The Catcher in the Rye and were full of questions. What happened? What was the point? Did Holden get better? Why does he act like that? I decided to let them interview an AI-generated version of Holden Caulfield on Character.AI to answer their questions.

Many teachers are looking for methods to integrate artificial intelligence into education in a way that generates learning experiences and still require that students demonstrate understanding and critical thinking. Platforms like Character.AI offer a wide range of possibilities for teachers to implement engaging assessment techniques that can encompass a wide range of learning standards—and can show students some of the shortcomings of AI. Here is a three-step process for using Character.AI as a fun assessment tool in English.

The assignment has five goals:

1. Assess understanding of character and plot.
2. Provide an opportunity for students to ask the unanswered questions of the book.
3. Develop AI literacy through student-led evaluation of the AI chatbot effectiveness and utility.
4. Develop critical thinking and perspective through the generation of broad, open-ended and empathetic interview questions.
5. Develop active listening skills via the safe space of a conversation with an AI chatbot.
   

Step One: Preparing the class for the assignment

First, interview your own “HoldenAI” on Character.AI’s platform. As you might expect, these bots are buggy. They can produce inaccuracies, misunderstand questions, and mix up characters and events from their own story. Furthermore, there are many versions of the same character (since they are user-generated). Before assigning this project, interview a handful of bots related to your book, subject, or task before settling on one that most closely resembles the historical or fictional character you have chosen. My favorite Holden bot was created by a user named @Karakter.

Next, you want to present and hype up the project. Have the class generate a list of “burning questions” for your given character with the class. Model curiosity by naming some of the questions you have always wanted to ask. Let students know they will be interviewing their own version of the character and evaluating its effectiveness. 

For homework, ask students to register for Character.AI, and have them interview their favorite celebrity or historical figure for fun. This step allows students to familiarize themselves with the platform in an engaging way. Character.AI provides a blanket statement regarding NSFW content that guarantees the blocking of pornographic content. However some characters—such as villains or movie characters—may use profanity or be aggressive to match their personality.

Step Two: Engaging with the Platform

After students have developed an understanding of the program, ask them to generate broad, open-ended questions that also exhibit an understanding of the character’s motivations and personality. 

One of my favorite questions for Holden was “Why didn’t you ever call Jane Gallagher?” The question works well because Jane was Holden’s crush and was very meaningful to him, but he never built up the courage to call her in the book. It gets to a core aspect of Holden’s character—an inability to face his feelings—while also being broad, open-ended, and a good conversation starter.

This part of the rubric or assignment should encourage students to engage in active listening. Students should respond to what the bot says rather than simply jumping to the next question. Student recognition of the bot responses could be as simple as “Thank you, that was helpful,” or even “I’m sorry to hear that.” 

The goal of requiring active listening is to give students enough material to evaluate the bot’s effectiveness and to develop AI literacy. Requiring this type of conversation will help many students understand what works and what doesn’t work with AI. The truth is more complicated than many realize, and our students need to be given at-bats to develop an understanding on their own.

After students are finished with the conversation, have them copy and paste the text into a separate document and submit it electronically. I found the student conversations to be fascinating. In one conversation, a student opened up about his own struggles with his family and found the chatbot to be a helpful talking partner for his problems. In another, a student had a long debate with the bot about the rationality of Holden’s actions that saw the two of them break down complex emotions into logical and understandable components.

In perhaps the most interesting conversation, a student confronted Holden so directly that the bot shrank in shame and slowly morphed into J.D. Salinger. The bot itself changed personalities from Holden to Salinger in the middle of the conversation.

Step Three: Evaluating the Bot’s Effectiveness

In this portion of the assignment, students must show understanding of the character by comparing and contrasting the comments from the chatbot with the actions, words, and personality of the character from the book. Students can pair evidence from the chat with evidence from the book to make their case. In a fun twist, the conclusions will vary by student, since the bot has a different conversation with every user.

Here is the essay prompt I used: “Evaluate the effectiveness of the AI chatbot in furthering your understanding of Holden Caulfield. To what degree did your interview with HoldenAI increase your understanding of his character? In what ways did HoldenAI match up with or differ from J.D. Salinger’s Holden Caulfield?”

As an example, the Holden chatbot that we used mischaracterized his brother Allie in conversation. In the book, Allie is Holden’s younger brother, who died of cancer three years earlier. The chatbot repeatedly thought Allie was a girl. Sometimes, the bot thought Allie was a girl whom he had dated before. In others, he thought Allie was a superficial sister. In these ways, students can see that the bots are flawed and often operating on incomplete information.

Unpacking the assignment

After they submitted their chats, we had a “postgame” class discussion. Each student shared the highlights or lowlights of their conversation. Students were surprised to realize that the bot had very different conversations with each person. For some, Holden was combative and angry. For others, he was sad and sullen. For still others, he was talkative and open.

When students realize that every one of their classmates had a completely different conversation with the same bot, they begin to understand one of the project’s more nuanced objectives: Questions matter, and so does listening. The way that the students prompt, listen, respond, and then converse will dramatically affect the conversation itself.