Abstract:

The use of Artificial Intelligence (AI) in education is growing as AI-based technologies become more and more developed and there are free versions available online for any user. Large Language Models (LLMs)-statistical models that predict the next token from the previous ones using a large amount of data-have already demonstrated their usefulness as conversational agents in multiple domains, including education. The question-and-answer dialogue ChatGPT is one of the most popular LLMs that has been released. It has access to a massive amount of data that enables a wide range of linguistic relationships providing understandable contextualized natural language. Therefore, it shows apparently admirable capacities for educational purposes (in fact, it was noticed worldwide that it could pass the United States Medical Licensing Exam and scored highly on the Psychology Today Verbal-Linguistic Intelligence IQ Test). Facing this new reality, how should academia act? Is ChatGPT a threat or an opportunity? Several newspaper articles and scientific papers were devoted to this topic, concluding that, in general, ChatGPT can be beneficial if integrated into the teaching-learning process, as long as the student is aware of its limitations. In fact, it has been reported that ChatGPT often provides biased and even, in some cases, incorrect information, so the students must have this awareness and should take the results with a critical mind. This is particularly relevant in Mathematics, where ChatGPT has shown several weaknesses and failures.

Our purpose is to find how ChatGPT performs simple mathematical problems of the 1st year course of Calculus of both mathematics and engineering students, such as symbolic integration and computation of areas. These students, by their early age and lack of experience in higher studies are, in general, less critical and therefore, have more tendency to accept ChatGPT answers as completely correct as it gives convincing answers. We used ChatGPT3 because, nowadays, it is the more developed version of ChatGPT that is free of charge and, therefore, more susceptible to be used by students.

Our results found several inaccurate answers and problematic issues in arithmetical and symbolic computations. Despite the apparent limitations, with proper guidance, ChatGPT could find the exact solution to some of those problems, showing how important teacher surveillance is when this LLM is used by a student. For some other problems, a guidance approach was not fruitful. However, the introduction of ChatGPT in classes could be an asset as it can improve students' critical thinking within an easy and friendly environment that encourages beneficial discussion among peers or between students and teacher. Moreover, with prudent use, it can provide a personalized approach to learning because the dialog is friendly, dynamic and adaptable, being a quick way of acquiring theoretical knowledge. These aspects are very important because the generation that is currently attending institutions of higher education, was born and raised in the digital age and is used to very quickly absorbing new software of all kinds, interacting with it, and customizing it. For this generation, this is almost as natural as any other aspect of their lives, and they expect to find it in their academic training as well. The biggest challenge for today's teachers will be to integrate these tools into their teaching practices.

Keywords:

ChatGPT, calculus, mathematics, education.