Abstract:

With the rapid advancements in technology and science, new paradigms have emerged in Education, with Information and Communication Technologies (ICT) and artificial intelligence (AI) as pedagogical resources.

The way our students acquire information nowadays differs significantly from consulting the bibliography recommended by the professor. Internet and YouTube are often the primary sources due to the way they interact with technology, and, of course, the use of ChatGPT. It has to be noted that the most widely used AI implementations to provide contextual responses, such as those expected for a tutoring tool, like ChatGPT, still suffer from a lack of accuracy or even mistakes in the responses provided, which demands its users to be proficient in the topics for which information in sought. Therefore, it is advisable to develop scaled versions of those tools, specific for certain topics, with the control of the instructors, both in the selection of sources to extract information for the answers, assessing the sources and references our students use to give greater value to the generated responses as well as in the supervision of the answers provided. In addition, this allows to add tools to carry out a predictive analysis of the learning process of the students so that the information given can be tailored to his/her needs, in an open-access and free way, leveraging the power of the tool adding value to this interaction, such as personalizing the learning experience. Integrating artificial intelligence into learning ecosystems is one of the biggest challenges for higher education.

The identification of concepts that students find difficult to understand is a key element in constructive learning and also allows for the development of specific intervention actions. Academic tutoring serves as a tool for this purpose and should be to individualize and personalize teaching. An update of this tool is the Intelligent Tutoring Systems (ITS), which are designed to enhance learning both inside and outside the classroom, replicating the effectiveness of human tutoring in digital tools.

In this work, an AI-based ITS has been designed with accurate bibliographic references using reliable sources that cater to specific needs of the students. The control of the possible responses is the first stage of a process that elaborates a predictive analysis of each learning study. For this, a first diagnostic phase was carried out that determined the selection of initial material that will be used by the neural network or classifier algorithm where we use LlamaIndex (a Python framework specifically designed to implement ChatGPT-like apps and interact with multiple LLM related components) so that the teacher has control of the answers offered, limiting the sources of information used. Subsequently, this selection of documents is part of the input, together with the question in conversational network large language model (LLM) like ChatGPT. Since training the network is not the goal of the work, a pre-trained network from an open-access repository is utilized. All developments have been done using open access resources. A preliminary test has been carried out by analyzing the responses to 40 different questions related to the documents fed to the application. These responses have been classified in different categories according to their accuracy and the results show that they could have a great potential.

Keywords:

Artificial intelligence, autonomous learning, intelligent tutoring.