Abstract:

The “boom” of AI, specifically with ChatGPT, brought various challenges to education. The use by the students to “cheat,” the difficulty from the teachers to validate what was created by ChatGPT or not, the misinformation the ChatGPT allows, and the low literacy in the use of the tool itself, among others, are now concerns for the majority of educational institutions.

With that in mind, we built an experiment that will allow the development of the user skillset to interact with ChatGPT and to use it in a “non-standard” way, therefore opening new uses for the tool as an education aid instead of a threat for the sector, or as a tool used only for cheating.

This paper will focus on the ongoing experiment, exploring its proposed goals and setup.

The more common use of ChatGPT, along with almost all chatbots and even search engines, is prompting a question and reading and exploring the answer given by the bot/app. These kinds of use cases lead to a very straightforward use of these tools, incentivizing cheating and reducing some of the most critical skills in such huge demand nowadays. With this kind of use (copy&paste of a straightforward answer), the user will not promote “critical thinking”, cross-reference sources, debate, and argument about the subject, among other skills that we need to promote among our students.

With that in mind, and considering this tool’s potential, we aimed at using ChatGPT not as an “answering machine” but instead as a virtual student to whom we teach a subject, and that will question our explanations and, in the end, do a summary of the topic we explained, give a rating on our explanation and some improvement feedback.

This way, it will allow the “tutor” to think about the subject to explain it to the ChatGPT, and with the follow-up questions, it will promote critical thinking and the need to really grasp the knowledge on the topic to be allowed to make a better explanation. With this experiment, we envision assessing if ChatGPT is a valuable tool to verify the Protegée Effect, and the Illusion of Explanatory Depth.

Moreover, using this table as a reference, the interaction proposed in this experiment will allow a Level 5 interaction, “Mutually explaining to each other, " the highest level and the most effective learning process.

To have different types of explanations, we set up two routes for explaining the subjects, one more technical (more detailed and accurate as possible) and another more based on analogies (examples and metaphors) and therefore less technical but more accessible.

Based on the summary results of the ChatGPT, we will have some data on each path and see if one brought more effective learning than the other.
By this interaction, we will also assess if this interaction improved the skill in the use of ChatGPT itself.

Keywords:

ChatGPT, AI, learning by teaching, protégé effect, Illusion of explanatory depth, metaverse, blockchain, extended reality.