How to Build a Simple Chatbot for Interactive Classroom Experiences
A chatbot building experiment in the tone of Al Capone.
Can teachers and instructional designers (with minimal coding experience) build their own chatbots?
I set out to find out if I could use simple AI tools to build a chatbot that would interact with students allowing them to carry out a meaningful task in Spanish, such as making an appointment or placing a food order. Before you start to think that building a chatbot is outside of your skillset, you should know that my total lifetime coding experience happened over the holidays when I coded a free tarot card reading on a website using ChatGPT as a late Secret Santa gift. That’s it. Still, my hope was that I could build a chatbot that would be a time saver for teachers and provide an engaging interactive experience for students.
I decided to use Hubble, an AI tool for building custom user interfaces and ChatGPT 3.5 Turbo chatbots. I chose Hubble because it promised that it required no coding experience and because the interface appeared to be user friendly. To test out the possibilities, I began building a bot that would act as a receptionist in a medical clinic, so that students could potentially have the experience of making an appointment in Spanish.
Designing a simple English language chatbot
I started out in English to see how well the chatbot could carry out the task in general without adding in the second language component. I found that the bot was able to do this rather well, with only a few glitch moments where it was repeating questions. I also prompted the chatbot to speak in the tone of Al Capone to test out the degree to which the bot could take on a distinct personality and, in all honesty, for my own amusement.
A few minutes later, I sent my twelfth grader, Ernesto, a text link to the chatbot and asked him to make an appointment. I found that the bot’s responses were sometimes crude, but to its credit, the bot was just taking on the tone as I had prompted, and it was actually pretty engaging and funny. I was surprised that “Al the bot” seemed able to pick up on conversational nuances such as the urgency behind some of Ernesto’s word choices.
Designing a Spanish speaking chatbot
After a little back and forth with the chatbot, Ernesto was able to make an appointment. So, after testing the English version, I moved on to the Spanish version of the chatbot. I decided to leave out the Al Capone tone this time and tested it myself. I gave the types of responses that I imagined a novice mid-range learner might give including grammatical and word choice errors and incomplete sentences in response to questions. This was not unlike the informal language that my son had used with the English chatbot. Unfortunately, the Spanish version of the bot repeated questions that I had already answered multiple times, and I was unable to complete the task of making an appointment, eventually giving up. I found the Spanish chatbot pretty unusable and frustrating.
I tried to prompt the Spanish chatbot to only ask simple questions, one question at a time, and to make it easy for the user to make an appointment. No luck.
I also tried responding to the chatbot in complete sentences only, instead of in fragments, and even provided a data source to the chatbot made of pre-scripted question and answer sets to train the chatbot to ask and answer questions. Unfortunately, using complete sentences and using the data source did not help, and I found that the chatbot was still repeating questions and failing to carry out the task of guiding the user to make an appointment.
The chatbot experiment results and future potential
In the end, my experience building a chatbot was not as successful as I had hoped, but I did see a glimpse of the potential that chatbots have to be valuable tools in education. Chatbots can potentially provide students with an interactive and engaging way to practice real-world tasks in a safe and controlled environment. They can also help students improve their language skills by providing opportunities for them to practice speaking and listening in a low-pressure setting. Through my chatbot experiment, I found that this tool already has a lot of possibilities in English, but it does not seem to work as well in Spanish at this point. I also learned that building a chatbot for use in the classroom requires a significant investment of time and resources, and maybe some coding experience on my part wouldn’t hurt.
