What are the perceptions of physics teachers in Brazil about ChatGPT in school activities?

In this work, we propose to discuss the perceptions of physics teachers in Brazil about AI generative like ChatGPT. Data were collected by an online Focus Group (FG) held during three meetings of one and a half hours each, with six Brazilian physics teachers with varied experience and backgrounds. Participants’ discourse was analysed according to three different questions: (a) the players involved in using ChatGPT in physics classes, (b) the attitudes towards the introduction of ChatGPT in physics classes, and (c) the main functionalities of ChatGPT in physics classes. Our results indicate that physics teachers’ perceptions of GPT, in general, involves more the role of students than the role of the teacher, correspond to more positive than negative perception, and allows identifying four main functionalities defined as a co-pilot of lessons, as an educational bureaucracy manager, as a simple problem-solving tool, and as a literal information providing tool.


Rationale
The wonders and doubts associated with the effects of digital technologies in school have come to the forefront since the emergence of ChatGPT.It's not for less!It is expected that Artificial Intelligence (AI) will change our way of interacting with the world, with others, and with knowledge, and consequently, also within the school.ChatGPT, an acronym for Generative Pretrained Transformer, is a mechanism that works through a chat interface that allows to establish conversations from the targeted processing of a large volume of data (humanly unattainable in the time that is established).It was launched and opened to Internet users in 2022 and reached the mark of more than 100 million users in only two months.Access to this AI tool is simple through its official site (chat.openai.com),and users can currently choose between the free version (GPT-3.5)and the more actual and paid version (GPT-4), although those versions may change shortly.
However, those wonders and doubts are not completely new.In education, we already experienced previous examples of "insecurities" caused by other "new technologies".The use of calculators in the 1970s slowly began to enter the classroom, with many fears by teachers and researchers, like the students wouldn´t learn more from their errors, or that they would become too reliant upon machines.More recently, a few years ago, when part of this discussion took place with the spread of Google's digital search engine that allowed anyone to search for information, an initial frenzy was generated, but now 2 this type of search is an everyday-life action and, schools have been adapted without major issues.Some procedures had to be adjusted, such as plagiarism detection, but overall, school life continued to move forward.Furthermore, chat systems themselves are not novel tools, neither for students nor for teachers.Since the remote ICQ and mIRC, from the 1990s to the current chats of virtual social networks like Messenger, WhatsApp, or Telegram, we have been using chats to establish conversations between people or groups.Perhaps now, the "fears" are because the questions that arise are linked to the generating capacity of this type of chat since the form of a chat was no longer big news, but the generative capacity, which makes it possible to create something that did not exist before, is perhaps one of the great points of discussion.On the one hand, now the interlocutor is a digital robot that uses a language model capable of understanding the meaning of sentences to return dynamic responses "generated" from the context of the question.On the other, this robotic interlocutor accesses and debugs the information contextualizes it, and learns from it, differentiating itself from traditional chatbots that are based on previously established rules.Thus, the replacement of a human interlocutor by a robotic one based on an artificial language model and with quick access to a huge volume of data and information trained not only to "provide" answers but to "generate" different answers to each question on several areas, it rekindles points of deep debate in the educational field that cover, among other aspects: the possibilities and limitations of its pedagogical use; the ethical issues associated with this use; the responsibility and the need for critical reading of the answers; issues of teacher training, evaluation etc.In this sense, it is important to remember that this generative AI is trained with the support of internet texts, and it is still not efficient enough to combat inappropriate biases, it does not give you results, but predictions, which may be correct, partially correct, or incorrect.In short, ChatGPT is being developed to generate responses with credible results, but not necessarily correct.We should not forget that in its origin, ChatGPT is a product of the purest behavioral psychology of reinforcement learning and that the model incorporates a whole series of rewards and penalties that fulfill several functions: reinforcing the coherence of the generated text, avoiding blatant falsehoods, but also moderating possible toxic drifts in advance [1].
For all of these reasons, the current debate surrounding GPT's role in education is significant for the entire educational ecosystem.This debate includes, for example, issues related to ethical concerns (including data privacy, biases in AI algorithms, and the impact on students' autonomy…), teacherstudent dynamics (the potential shift in the teacher-student relationship when AI is introduced as an instructional tool, and how it affects traditional teaching roles), equity and access (concerns about the digital divide and whether AI in education exacerbates educational inequalities or helps bridge the gap), teaching insecurities (the need for teacher training in using AI tools effectively and integrating them into the classroom), risks related with science denial and disbelief in a world where science denial is a growing concern, etc.To address all of these issues, many initiatives have begun to consider when and how to use GPT in the classroom.These initiatives are driven by the recognition that integrating AI technologies like GPT requires careful planning and consideration of both the benefits and potential challenges.Educators, policymakers, and researchers are actively exploring strategies for incorporating GPT into the curriculum in ways that enhance learning, empower teachers, and ensure equitable access for all students.For example, UNESCO [2] has recently proposed a flowchart to determine when it is safe or unsafe to use GPT, depicted in Figure 1.And what about physics teaching and learning?As occurs with many other tools, its application in physics classes (and the associated debates) is influenced by the specificities of this discipline, such as the prevalence of mathematical procedures in many physics' classes, the role of theories and experiments, the presence of abstract idiosyncratic physics' entities (forces, fields, flows, energy, states, particles, etc.).On this basis, many questions arise: can ChatGPT provide real-time support, explanations, and personalized guidance to students?Can it enhance their understanding and problemsolving skills?Recent investigations in physics education research have indicated the need for a better and more detailed understanding of the potentialities and limitations of ChatGPT in different contexts [3], [4], [5].In comparative studies such as [6], it was shown that ChatGPT version 3.5 can match or exceed the median performance of a student who has completed one semester of college physics, or in the case of [7], which delved into ChatGPT's performance through representative assessment content of a calculus-based physics course, and when evaluated the responses using the same criteria applied to human responses, it was concluded that large language model chatbots like ChatGPT, would narrowly pass this course while exhibiting many of the preconceptions and errors of a beginning learner.
In this context, in this work, we propose one exploratory study, to discuss the potentialities and limitations of AI generative like ChatGPT in the view of different physics teachers in school activities.To this end, we aim to address the following research question: What are the perceptions of physics teachers about generative artificial intelligence in school activities?

Methods
This is exploratory research, and we opted for the Focus Group technique (FG), both for production and for collecting information.As highlighted by [8] "Focus Groups are discussion groups that discuss a topic when they receive appropriate stimuli for the debate.This technique is distinguished by its characteristics, mainly by the process of group interaction, which is a result of the search for data" (p.780).In general, the main objective of FGs is to identify the subjects' perceptions and ideas about a given topic, as the information collected is more detailed and diversified.Furthermore, FG makes it possible to intensify access to information about a given phenomenon.
For the organization of this FG, we were guided by [9], accepting the indications on care about the selection of members that will make up the group in terms of homogeneity, as well as the recommendation that an FG not have more than eight participants, since if the purpose of the group is to maximize the depth of expressions of each participant, a smaller group works better [10].

Participants
Six people composed this FG, including four experienced teachers who teach Physics in a Brazilian High School, and two physics students in internship activities, who work in public and private Brazilian elementary schools and who were "struck" by doubts and questions about ChatGPT in their classes.This situation is also reflected in the organization of the FG itself, since, as [9] points out, participants must share characteristics that are of interest to the investigator.
Synthesized in Table 1, we present some information from the participants of this FG, which was obtained through authorization granted through a Term of Free and Informed Consent.

Implementation of the focus group
This FG took place during three consecutive meetings of approximately 1 hour and 30 minutes each, implemented during May and June 2023.The moderator of this FG was one of the researchers himself, who played a non-directive role and with a discrete approach, proposing open questions and interfering the least in the group's dialogue in a semi-structured perspective.Due to the condition of the moderator to be outside Brazil, this took place in an online format through the Google Meet platform, with all its participants.Online FG is a very common procedure in qualitative research [10,11], and its use has increased enormously in the last few years due to the need for physical distancing imposed by the COVID-19 pandemic.
The moderator prepared the Topics Guide, organized from the general to the specific.In this investigation, three Topic Guides were created, one for each meeting, with a similar structure, that started discussing the objectives of the meeting, some briefings were provided, followed by a warm-up, a central discussion, and a closing.Those central discussions were different in the three meetings but interconnected about the central theme, which was GPT in physics teaching.
• In the first meeting, the central discussion focused on the view of teachers about the AI role, especially the ChatGPT, in helping with school tasks.• In the second meeting, the central discussion aimed to identify whether teachers envision AI as promoting changes in formal educational processes.• In the last meeting, the central discussion had a more specific focus, aimed at the teachers' understanding of the ChatGPT performance in the resolution of different types of school physics questions.

Definition of Units of Analysis (UA)
The recording of the three meetings gave us a total of 260 minutes of audio-recorded documents.It included about 400 utterances, which is a different uninterrupted chain of participants' discourse (speaking turns).To identify the relevant utterances for this study, many utterances were excluded from the analysis, especially those related to contextual and personal comments, jokes, welcome, farewell, etc.This led us to a total of 119 utterances for the analysis.The average time for each utterance was about half a minute but with a wide variety of length.Following, in the second step, those 119 utterances were divided into two groups.On the one hand, 65 utterances were directly related to ChatGPT, such as: "GPT perform some tasks well, and others very badly.Yet! [...] I did some tests with some questions, and it performed very well with the easier and simpler [physics] questions." (P6).
On the other hand, 54 utterances were excluded from the analysis, since they were not directly related to ChatGPT, but to other ICT (Google, Arduino, Digital Social Network, etc.), such as: "Today students think they have all kinds of information on social media, but it is biased information, often false, and they [...] they don't know how to differentiate, you know?" (P2).

Data Analysis
After selecting the 65 utterances directly related to ChatGPT, a detailed reading of them was carried out, and they were grouped according to the similarity and contrast technique.Through this analysis, a system of categories emerged, and it was refined until reaching the final version.This system of category included three main dimensions of analysis: • Which are the identified players in the discussion about the ChatGPT in physics classes?
• What are the identified attitudes towards the introduction of ChatGPT in physics classes?
• What are the identified functionalities of ChatGPT in physics classes?
For each of those dimensions, different categories were identified.To represent how those 65 utterances were distributed among participants and over time, a graphic such as Figure 2

Which are the identified players in the discussion about the ChatGPT in physics classes?
The references about the players involved in the use of ChatGPT according to participants included both teachers and students.Overall, we observe more utterances about the possibilities of using ChatGPT for/by students than about themselves.Numerically, considering the 65 utterances, 40 mentioned students, while 25 of them are about the possibilities and potentialities for using ChatGPT by teachers.Specifically, in the utterances mentioning teachers, we identify utilitarian mentions for and against the ChatGPT, expressing concerns about how this tool may change the role of educators, and also the importance of adapting to technology changes, such as: "I think that at one point it [IA] will arrive at the school with a foot in the door, and we will have to choose which side we want.[...].Better to have them helping us [teachers] than against it because I think it will pass on top of everything" (P6).
According to utterances referring to students, it appears the ideas about the difficulties students will have, the changes and opportunities for learning processes, and the students' lack of criticism about the texts generated by ChatGPT, for example:

"This chat, I believe, will mess up the students' heads even more, not for him, but for the students, who don't know how to differentiate in physics what is right or what is wrong" (P2).
In Figure 3 we present the distribution of analyzed utterances, presented according to participants and temporal arrangement, and classified whether they mention students or teachers (themselves).According to time, although there are more utterances mentioning students than teachers themselves, it can be seen that both players are permeated by the discussion about ChatGPT in physics teaching, but with some differences according to the topic prevalent of the meeting.For example, P3 and P4 were those participants that maintained an above-average proportion between mentions of teachers and students (more references about teachers), while for P2 and P5 this proportion was below average (more references to students).

What are the identified attitudes towards the introduction of ChatGPT in physics classes?
According to the attitudes, our understanding refers to the expectation of the result provided by ChatGPT.We identified some positive attitudes, that is, those in which ChatGPT could generate useful and effective solutions for various school activities, for example, when they mention that ChatGPT can somehow help in teachers' tasks.
"I think that today, this [ChatGPT] is still not very good [...] but it will be.And I think it will soon be able to help us with things [...] not the nonsense on the internet, to help with diet and such, but with school things, in bureaucracy " (P3).
Other utterances emphasized negative aspects when, in the view of participants, the answers generated by ChatGPT bring more harm than gain to the teaching and learning processes.It included, for example, the limitations of ChatGPT in solving physics problems by different typologies or the blunders it can make in simple exercises, like for example:

"And suddenly the boy tells me that GPT wasn't very intelligent, because he gave to it [GPT] some simple physics exercises, and when I corrected [the exercises] in the classroom, the GPT didn't get a six [over ten]." (P2).
Finally, other utterances were classified as neutral or ambivalent, since they did not present positive or negative characteristics, or when the type of mention could give both a positive and negative indication, for example: "I don't know…, I don't know.Now I can't [...] I need to test [...] but I think that it can answer the same question right or wrong, whether it's easy or difficult.[...] When I played, it made more mistakes than it got right, but he also got a lot right, you know?" (P6).
Overall, we identified that there are more positive (34) than negative (23) mentions of expectations with the help of ChatGPT in various school tasks.Regarding positive aspects, there are 21 mentions of teachers and 13 of students, indicating comparatively that teachers have a more positive expectation of feedback from ChatGPT for their activities at school than for their students.As for the negative aspects, there are more mentions of students (16) than of teachers, which is, an indication in the teachers' view that the role of ChatGPT in student activities will be more negatively affected compared to teachers' activities.There are also eight neutral or ambivalent utterances in this classification, four related to teachers, and four related to students, presented in Figure 4 According to time, there is a prevalence of initial utterances with an ambivalent attitude, and meeting 1 is mainly compounded by negative attitude utterances, indicating an initial fear of participants positively mentioning ChatGPT in school activities, despite knowing about it.However, meetings 2 and 3 are compounded by more positive than negative utterances, showing a change of attitude during the 260 minutes of focus group.Focusing on differences between participants, P1 and P2 are those with a higher ratio of negative utterances.On the other hand, P4 stands out for its neutral utterances, and P3, P5, and P6 show a higher ratio of positive utterances.

What are the identified functionalities of ChatGPT in physics classes?
According to the functionalities of ChatGPT in physics classes, some utterances referred to ChatGPT as a teacher's assistant in didactic intervention, for example, to support teachers in designing and implementing an experimental class, or to replace the teacher in the development of students' material.In some cases, this functionality was defined as if GPT were a "co-pilot" for the teacher, as can be observed in the following dialogue between participants: "There are some little things that might help, who knows [...] Like us asking during class for a compilation of rainfall data from here at BPS over the last 5 years, have you thought about that?Not like J.A.R.V.I.S. from Tony Stark [referring to Iron Man film], but there, asking, just like we do with Google (P3).
[laughter of participants] problem-solving tool functionalities combine both positive, neutral, and negative attitudes, but with a prevalence of positive ones.Finally, the information-providing tool functionality implies more negative than positive utterances.

Conclusions
In this research, we conducted an online FG with six different physics teachers from Brazil from different backgrounds and experiences, to identify which are the perceptions of physics teachers about ChatGPT in school activities.Because of the methodological approach, it is important to emphasize the role of the mediator in the process, and also the individual differences among participants.For example, there was a greater frequency of intervention from P3, who was the teacher who anticipated all the themes and had the most interventions, and it may have influenced the progress of the focus group.In addition, the presented findings may not necessarily be representative of the broader population of physics teachers in Brazil or in another context.Nevertheless, we assume that such limitations are inherent to this type of methodology.
According to the obtained results, we can conclude that the perception of participants of ChatGPT implies a positive trend that this tool can help due to educational bureaucracies outside the classroom and the didactic interventions, although they recognize little use of ChatGPT.This perception is grounded on the idea that ChatGPT can help their students positively with simple physics questions and negatively with general information that is usually copied and pasted automatically, without promoting reasoning skills, critical thinking, and concept understanding.The characterized perception of physics teachers also implies a disparity between what teachers perceive about students and themselves about ChatGPT.The more negative perceptions are based on the lack of confidence in the responses generated by the chat, as discussed by [13], and the fear of students' lack of criticality when dealing with the information provided by ChatGPT.
We indicate, as one of the limitations of our conclusions, the fact that we compared the ratio of the utterances of the teachers participating in the FG without taking into individual utterances variations.We justify that such an analysis would require a more specific look at the individual instead of an investigation that sought to identify more collective movements of this individual with the group.
Finally, even in Brazil, with great social inequalities and difficulties in accessing technologies in schools, in-service physics teachers recognize the arrival of ChatGPT in schools, and given its impact, more specific and in-depth studies will be required, as we need to critically understand how these technologies work and what it means to use them (or not) at different tasks and levels of education.

Figure 1 .
Figure 1.When is it safe to use ChatGPT?
was used.

Figure 2 :
Figure 2: The n = 65 analysed utterances according to their participants and temporal arrangement 3. Results

Figure 3 :
Figure 3: Distribution of utterances according to the player involved.

Figure 4 :
Figure 4: Distribution of utterances according to positive/negative attitudes shown.

Table 1 .
Profile of teachers participating in the Focus Group.

Table 2 .
Relationship between functionalities given to ChatGPT and attitudes to it.