2.1 Artificial intelligence in education

Although AI technology has been the subject of study and scientific interest for more than six decades, its practical applications for education have been developed only recently (^{Russell; Norvig, 2010}). Researchers such as ^{Bates (2015)}, ^{Renz and Hilbig (2020)}, and ^{Tavares, Meira and Amaral (2020)} highlight that the first attempts to replicate the teaching process using AI emerged in the 1980s. Since then, Artificial Intelligence in Education (AIED) has established itself as an academic research field, introducing new tools into the educational environment with the potential to significantly transform traditional teaching and learning methods.

One of the most common definitions of AI in education (AIED) was proposed by ^{Popenici and Kerr (2017)}. According to these authors, AIED refers to “computational systems capable of engaging in human-like processes such as learning, adaptation, synthesis, self-correction, and the use of data for complex tasks” (^{Popenici; Kerr, 2017}).

Based on this concept, AI technologies have had a significant impact on the educational process, especially with the emergence of new functionalities, such as recommendation systems that provide personalized instruction; intelligent tutoring systems that offer instant feedback; and automated assessment systems (^{Adiguzel; Kaya; Cansu, 2023}; ^{Ouyang et al., 2023}; ^{Tavares; Meira; Amaral, 2020}).

As pointed out by ^{Malmström, Stöhr and Ou (2023)}, higher education students are increasingly immersed in an environment influenced by AI, both in their academic and daily lives. This means that today’s university students can experience an interactive and personalized learning environment thanks to AI applications. At the same time, instructors can explore new teaching methods, identify different learning styles of students, and provide personalized guidance, adapting their methods according to the individual needs of each student (^{Kuleto et al., 2021}). However, ^{Silveira and Vieira Junior (2019)}, when analyzing the structure of federal and state learning environments, show that there is a technological gap that depends on the public sector for expanding data communication and infrastructure. Therefore, exploring the possibilities of AI is conditioned by structural improvements that allow for its successful implementation (^{Silveira; Vieira Junior, 2019}). In this sense, ^{Renz and Hilbig (2020)} observe that successful and sustainable cases of AIED implementation in higher education are still very scarce.

Furthermore, ^{Adiguzel, Kaya and Cansu (2023)}, argue that although AI has the potential to revolutionize education, achieving positive educational outcomes requires more than just the use of advanced technologies. The application of AI in education must be aligned with educational and learning theories, guiding instructional design and technological progress.

A recent and controversial advancement in AIED is using chatbot systems to assist in teaching and learning activities. These intelligent agents can interact with users, answering questions and providing appropriate responses (^{Adiguzel; Kaya; Cansu, 2023}). One of the most well-known examples of a chatbot, nowadays, is ChatGPT.

2.2 Generative artificial intelligence

^{Lecun, Bengio and Hinton (2015)} indicate that AI is based on deep learning, which allows computational models with multiple processing layers to learn data representations at various levels of abstraction. These methods have significantly improved the state of the art in areas such as speech recognition, visual object recognition, object detection, and drug discovery. According to the authors, using the back-propagation algorithm, deep learning adjusts internal parameters to uncover complex structures in large datasets. Furthermore, deep convolutional networks are effective in processing images, videos, speech, and audio, while recurrent networks are suitable for sequential data, such as text and speech.

^{Banh and Strobel (2023)} explain that this deep learning serves as the foundation for what has come to be known as generative AI, a tool used by users through simple commands to produce new and realistic content across a wide spectrum (e.g., texts, images, or programming code). One example of a generative AI tool is GPT (Generative Pre-trained Transformer), a Natural Language Processing (NLP) model that uses deep learning to enhance its outputs (^{Adiguzel; Kaya; Cansu, 2023}; ^{Barbosa; Portes, 2023}; ^{Benevento; Meirelles, 2023}). Developed in 2018 by OpenAI, a research organization in AI, the GPT model has undergone several updates, resulting in its most well-known version, ChatGPT. This version was specifically designed for human interaction and has since been widely used in chat applications and services to provide coherent and fast information to users (^{Velásquez, 2023}).

The functioning of ChatGPT is based on a pre-training process in which the model is exposed to large quantities of textual data from various sources, such as articles, websites, books, and written conversations. During pre-training, GPT learns to predict the next word in a sentence or complete a sentence meaningfully, thereby improving its ability to generate cohesive and semantically accurate texts (^{Benevento; Meirelles, 2023}).

According to ^{Barbosa and Portes (2023)}, the incorporation of thousands of human language examples has allowed the technology to deeply understand the context of user requests, resulting in more accurate responses to demands.

ChatGPT has a wide range of functionalities, including language translation, text summarization, answering questions, creative writing, generating high-quality content, and correcting errors in existing code or generating new code (^{Eke, 2023}). According to ^{Adiguzel, Kaya, and Cansu (2023)}, these functionalities make ChatGPT a valuable tool in various fields, becoming a relevant resource for students and professionals.

However, like any system, ChatGPT is not without flaws. Some of the main challenges raised by ^{Adiguzel, Kaya and Cansu (2023)} and ^{Velásquez (2023)} regarding the development and programming of chatbots based on generative AI include the fact that the accuracy of the responses provided by the chatbot is intrinsically linked to the quality of the input data. In other words, to obtain accurate responses, users need to provide precise input data. Additionally, despite being highly accurate most of the time, ChatGPT can still make mistakes or provide imprecise and incorrect answers in some situations. It is important to mention that there are problems with the databases of these systems, which may be biased and not adequately meet users’ real needs. Finally, there is the risk of the chatbot providing harmful instructions to users, such as sharing malware code, as well as potentially reproducing biases and inequalities present in the training data obtained from the internet.

In the educational field, the use of ChatGPT and other AI-powered text tools raises significant concerns about academic integrity (^{Eke, 2023}). Stakeholders in higher education and scholars have raised questions about the advancement and influence of AI on teaching and learning, especially in the context of evaluation/exams and academic ethics (^{Malmström; Stöhr and Ou, 2023}).

In this context, it becomes essential and urgent to analyze the main ethical and social challenges related to the use of this technology (^{Barbosa, 2023}) to ensure sustainable and ethically sound academic practices.

a) The Use of AI in teaching practices

AI technologies have had a significant impact on the field of education, as evidenced by resources such as recommendation systems, intelligent tutoring, and automated assessment (^{Adiguzel; Kaya; Cansu, 2023}; ^{Ouyang et al., 2023}; ^{Tavares; Meira; Amaral, 2020}). In this context, the EBIA document emphasizes that AI will impact the Research, Development, and Innovation environment as a facilitator of new discoveries and scientific advancements through its ability to process large volumes of data, identify patterns, explore hypotheses, and make inferences that are beyond the capabilities of humans (^{Brasil, 2021}).

In this regard, Interviewee 1 highlights that the use of AI can bring significant advances in teaching practice, particularly in the development of courses, the creation of case studies, the development of virtual scenarios, and a search tool. According to him, “AI should be applied to assist in the development of new challenges, as a tool for searching specific topics, in the creation of virtual realities, and mapping market demands for inclusion in course content.”

On the other hand, Interviewee 2 presents the perspective that AI should be used, as long as it always prioritizes the pedagogical approach and serves as a facilitator to enhance the university’s educational mission, without specifying how this usage could be materialized. For him, “[...] AI should be used if it frees students and professors to fulfil their ultimate missions, which are to learn and to teach. I don’t have answers about the conditions under which this can happen.”

The interviewees also emphasize that AI can assist students in the context of research development, as an intelligent tutoring system or a recommendation system. “When you’re starting from scratch, and you want to relate two things that you don’t even know where to begin” Interviewee 2. In this perspective, “I believe these tools can be used to support students, assist in research, and summarize large amounts of content” Interviewee 1.

However, Interviewee 1 draws attention to contexts where AI should not be used in research due to ethical concerns that touch upon critical development and dependence on tools. For him, “[AI] should not be applied in the activity of reflection and creation itself (...) we would then have an ethical problem because the professional or student might become dependent on the insights and ‘starts’ that AI can promote.”

On the ethical front, the EBIA states that AI should “promote mechanisms that encourage the development of AI systems that adopt ethical principles and values” (^{Brasil, 2021, p. 36}).

Table 2 presents the triangulation of interview data, the theoretical framework, and the EBIA to enrich and synthesize our understanding of the use of AI in the educational context.

The opportunities prescribed by the interviewees confirm many positive aspects and potentials of AI in education, as highlighted in the theoretical framework and the EBIA. However, no actual AI practices are presented, with concerns primarily focused on ethics, and there are reservations regarding its implementation. Thus, the scarcity of successful and sustainable cases of AIED in higher education, emphasized by ^{Renz and Hilbig (2020)}, is confirmed by the prescriptive approach of the interviewees, who do not specify instances of AI in the teaching practice of the Production Engineering Program at COPPE itself. Therefore, these analyses also support the ideas of ^{Silveira and Vieira Junior (2019)}, who highlight that technological lag in state and federal learning environments is an issue that hinders the successful implementation of AI practices in this context.

b) Training and normative knowledge on AI

^{Tavares, Meira and Amaral (2020)}, assert that the implementation of AI in the educational field must be planned, gradually, and emphasize supporting teaching, avoiding the automation of the educational process. This includes the training of professionals working in this area, through the development of the technical skills necessary to interact with and take advantage of the possibilities that AI can offer them (^{Brasil, 2021}). However, the interviewees stated that they had not participated in any specific training on the topic and were unaware of any regulations or guidelines that directly or indirectly impacted them. It is important to note, however, that the interviewees acknowledge some activities related to open debates and experimentation on the subject within the university, especially regarding ChatGPT.

According to Interviewee 1, “there were some debates [on the topic], but they did not result in any regulations or guidelines. Regarding COPPE, we don’t have anything consolidated.” Interviewee 2 states, “I haven’t received any communication about it yet. I think we are still in the initial stages, with lectures and beginning a reflection on the topic.”

Table 3 highlights how information from various sources converges to provide a more complete and accurate view of AI training and related regulatory guidelines.

The triangulation reveals that the interviewees confirm the need for training and the development of technical skills for the integration of AI in education, as advocated by the theoretical framework. However, the lack of specific training and limited knowledge of regulatory guidelines indicates a disconnect between theoretical guidelines and institutional practice at the university. Informal activities, such as debates and experimentation, complement but do not replace the structured and planned implementation suggested by the EBIA and researchers like ^{Tavares, Meira and Amaral (2020)}.

c) Challenges and opportunities of AI in teaching practices

The research findings reveal a series of challenges and opportunities that arise with the application of AI in teaching practice. Regarding the challenges, it is important to highlight that the use of technologies such as ChatGPT and other AI-based text generators poses risks to academic integrity and generates issues related to the originality and authenticity of academic work (^{Eke, 2023}). In this sense, Interviewee 1 emphasizes that the challenges are precisely related to plagiarism: “[...] you can identify that the work is the initial one, and the person took the idea from somewhere as inspiration, but sometimes they take everything, and that’s an ethical issue. Was it really done, did I use it, or did I just copy it? We still don’t know how to map this yet.”

^{Silveira and Barros (2021)}, discuss the challenge of motivating students and aligning teaching and research practices with the high availability of knowledge from these tools. In this context, Interviewee 2 speaks about the changing role of the teacher and how they bring themes into the classroom. According to him, “In the classroom, in teaching, I really think the role of the teacher has changed. Since I entered university and became aware of this, I said: chalk doesn’t work anymore... I’ll start talking about an author, and the person will open Wikipedia and read about the author, and they’ll know more about them faster than I can talk. Or from an authorized source, like Stanford, for instance.”

It is worth noting that despite the EBIA advocating for the idea of “expanding research, development, innovation, and the application of AI by enabling the provision of specific resources for this topic and coordination between existing initiatives” (Brazil, 2021, p. 35), Interviewee 1 still views the lack of university infrastructure, the absence of full-time students, and limited funding for projects addressing the theme as challenges. According to him, “We’ve had a latent infrastructure problem for some years, especially if we want to start talking about artificial intelligence. We don’t have infrastructure preparedness, given that the internet itself is not robust, even when thinking about something more interactive with the students. [Additionally], another challenge would be the lack of an office at Coppe to capture, maintain, and manage projects so we could offer the student a scholarship that would allow them to do the program full-time (studying and developing).”

Regarding opportunities, ^{Malmström, Stöhr and Ou (2023)} argue that the contemporary university context has the potential to experience an interactive and personalized learning environment, thanks to AI applications. In this context, both interviewees highlight how AI could assist in interactivity and the facilitation of managerial, administrative, and routine processes within the university, saving teachers’ time and increasing the university’s managerial efficiency. Interviewee 2 emphasizes, “If I had a scenario where AI helped me with administrative tasks, allowing me more time to dedicate to the core activities, it would be wonderful. [...] Freeing me from administrative tasks [since] my role is to guide the students!”

Similarly, Interviewee 1 states that AI could provide more efficient systems for managing routines and academic follow-ups with students. “Today, for example, discovering the deadlines and information of each student is a huge task. I don’t have all the advised students registered in SIGA, and that would be a tool that could already be developed” Interviewee 1.

In parallel, ^{Kuleto et al., (2021)} emphasize that AI provides new opportunities for teachers to explore different teaching methods, identifying students’ individual learning styles and offering personalized guidance. Additionally, the EBIA “proposes establishing connections and partnerships between the public sector, private sector, scientific institutions, and universities for advancing the development and use of AI in Brazil” (^{Brasil, 2021, p. 35}). In this sense, the interviewees discuss how the use of AI could foster learning dynamics. According to Interviewee 1, “The main opportunity would be bringing real-life scenarios into the classroom. For professional purposes, development, and even everyday management, in the classroom. I’ve used ChatGPT myself to generate some case studies for the undergraduate course, and it helps bring more up-to-date material.”

Interviewee 2, on the other hand, believes that AI can have educational functions, such as “‘opening the mind’ or presenting things that don’t require creativity, invention, or creation in general [...] it [AI] can integrate into educational processes, but in that way.”

Beyond opportunities, the ideas raised by the interviewees previously already indicate and relate to the perspective of a new teaching dynamic, based on new ways of constructing the educational process and, why not, the roles of each actor within the classroom. In line with this, the EBIA reinforces, regarding qualifications for a digital future, the need for “innovative pedagogical practices and the importance of re-signifying teacher training processes to address the challenges arising from the inclusion of technology and AI as a pedagogical tool in the classroom” (Brazil, 2021, p. 28).

In Tables 4 and 5, we highlight the triangulation of information to enrich the understanding of the challenges and potentialities (respectively) associated with the integration of AI in education.

Thus, from the triangulation of the information in Tables 4 and 5, it was possible to verify that the interviewees confirm and complement the ideas presented in the theoretical framework and the EBIA. They emphasize the challenges identified in the theoretical framework, especially regarding academic integrity (^{Eke, 2023}), the changing role of the teacher (^{Kuleto et al., 2021}), and infrastructure issues (^{Malmström; Stöhr; Ou, 2023}). At the same time, they complement the highlighted opportunities, reinforcing the potential of AI to create interactive learning environments and improve administrative and academic efficiency.