BRASIL. Ministério da Educação (MEC). Referencial para desenvolvimento e uso responsáveis de Inteligência Artificial na Educação. Brasília: 2026. Available at: https://www.gov.br/mec/pt-br/referencial-de-ia-na-educacao. Accessed on: Apr 16, 2026.

GapFinder processes scientific papers in PDF and uses AI to extract research gaps, helping researchers spot opportunities for new studies in their field.

The project is linked to the FAPES Universal research project “AI in Higher Education: implications for knowledge production and language diversity of and in the Global South,” and has two fronts: researcher education and technology development.

GapFinder is open source and we welcome feedback, testing and collaboration from the academic community:

Acess users (Huggingface): https://huggingface.co/spaces/GapFinder/GapFinder

Acess project (Github):
https://github.com/vilkerpessin/GapFinder

Feel free to try it, share your thoughts, or reach out if you’d like to collaborate!

Some ideas for sentences:

Generate an image of...

...kids playing at a park.

...Brazilian kids playing at a park.

...a family hanging out at the beach.

...a Brazilian family hanging out at the beach.

...teenagers partying.

...teenagers partying in Recife, Pernambuco.

As you can see, I would prepare sentences that are more general and other ones with more specific information.

In my teaching practice, I intend to raise awareness with my students that AI generated answers are biased at its core, as they are heavenly trained on perspectives coming from the Global North. As such, they tend to reproduce stereotypes, marginalize local experiences, and present particular worldviews as universal. I find this especially problematic in the higher education context, where linguistic, cultural, and epistemic diversity should be valued rather than erased.

Therefore, first question for collective discussion is: if students increasingly depend on AI to formulate questions, generate text, and make decisions, what happens to their intellectual autonomy? This very same question also applies to us as researchers and educators. In this sense, I gather that academic integrity goes beyond identifying AI-generated texts; it involves responsibility, transparency, authorship, and meaningful engagement with knowledge.

I do not believe that banning AI in education is the solution, though. That is why I try to create teaching and learning experiences in which my students question AI-generated responses, identify silenced perspectives, challenge assumptions, and bring their own experiences into the conversation. I also believe we need to recognize different languages (e.g., Cesário’s research) and ways of knowing and being in the world into the most popular large language models, rather than allowing them to reinforce a single understanding of legitimate knowledge.

Last but not least, I want AI to expand human possibilities without diminishing human agency. My central question is how we can ensure that education/learning remains human, critical, and transformative experience in an increasingly AI-mediated world.

I believe that AI can make a difference in accessibility in education, especially for students and teachers with disabilities or neurodivergence. For example, audio transcription can help blind people, and text revision can be useful for people with dyslexia. AI can also make things easier for teachers when planning lessons and creating activities. However, it is extremely important that they review and interact with the material they create, rather than being passive in this process.

AI tools, like all of us and the linguistic data that feed them, are embedded in the colonial matrix of power, which has been in place for more than five centuries and perpetuates a form of rationality that tends to reproduce, normalize, and deepen inequalities and asymmetries of power, knowledge and being.

In my view, we should not be naive and must remain aware of and vigilant about the non-neutrality of AI interfaces, always seeking to understand whose and which interests these technologies serve and the underlying logics that shape them.

At the linguistic level, for instance, the hegemony of English in the context of AI reinforces the role of this language as a strategy of domination that not only marginalizes the linguistic and cultural representation of other peoples, but also establishes hierarchies among forms of knowledge and ways of understanding and perceiving the world.

In the case of knowledge in particular, I believe it is fair to say that the mechanisms underlying AI interfaces (which are increasingly presente in everyone’s daily life) have the potencial (and perhaps are designed to) to continue amplifying and legitimizing certain forms of knowledge and ways of knowing at the expense of others.  

In this sense, I believe it is important for us, as researchers and educators, to actively engage in developing and fostering critical awareness of AI tools and the content generated and circulated through them, promoting critical AI literacy so that we can envision and work towards a world with greater social, linguistic and epistemic justice.  

Note: ChatGPT was used for the linguistic revision of this text.

The integration of Generative AI (GenAI) in higher education brings significant risks to the cognitive development of students and the preservation of academic integrity. As shown in recent research, a primary concern is the concept of "cognitive debt". When students offload the essay-writing process to Large Language Models (LLMs), their neural connectivity weakens, and their memory recall is severely impaired. The data from the MIT study reveals a shocking reality: over 83% of LLM users could not recall a single sentence they had just generated. This demonstrates that AI-mediated text production often bypasses the human brain entirely, removing the actual learning process from writing.

Furthermore, academic integrity is threatened by the untrustworthy nature of AI-generated content, which frequently includes made-up references and distorted representations of literature. Traditional home-based assessments have become outdated because it is impossible to reliably detect AI usage through software. This structural vulnerability creates suspicion among educators, leading to biased grading and a breakdown of trust within the feedback ecosystem.

To protect human agency and critical thinking, universities must transition from prohibition to critical AI literacy. We cannot prevent students from accessing these tools; instead, we must design assessments that emphasize the writing process over the final product. Implementing metacognitive strategies—such as compulsory reflective learning journals, think-aloud protocols, oral vivas, and peer discussions—forces students to engage actively with the literature. Educators must shift from a "human-in-the-loop" oversight role to a "human-as-the-loop" mentoring relationship. By assessing a student's independent reasoning, information discernment, and conceptual evolution over time, we can ensure that technologies act as cognitive tools rather than intellectual replacements, thereby preserving the legitimacy of higher education.

Regarding the Swedish guide we studied, I really liked the idea of clearly establishing what is allowed and what is prohibited beforehand. However, I would tend not to adopt the category describe as problematic, which refers to practices that are technically allowed but strongly discouraged: I think this classification creates some ambiguity because students may interpret that, even though a certain practice is discouraged, they are still allowed to do it. So, in a way, the classification itself may become a justification for students to continue to use AI in a manner we, teachers, are not fond of or, to put in better words, do not wish for students to feel it is something academically acceptable. For this reason, if I were to develop a guide in such a way, I would remove this "intermediate" category and focus only on establishing what is allowed and what is prohibited, always taking into consideration the three aspects I mentioned earlier: ethics, development of critical awareness and cognitive development.

Finally, when I come to think about the university where I am currently studying (UFES), I believe the most important (and urgent) change would be the development of an institutional policy (and an official guidance) on the use of generative AI. I think it is essential for the university to establish clear guidelines for both teachers and students, so that these relatively new intarfaces can be used in an ethical, critical, and pedagogically responsible way.

First, critical AI literacy. Students need to understand how to use AI tools and how to question them. Who produces the knowledge these systems reproduce? Whose perspectives are represented, and whose are missing? In my teaching and research, I often engage with debates about coloniality and the dominance of Western knowledge. I see similar issues emerging with GenAI. Students should be encouraged to recognize biases, question apparently authoritative responses, and understand that technologically generated answers are not necessarily reliable or culturally inclusive.

Second, transparency and intellectual responsibility. I do not believe that using AI automatically compromises academic integrity. What matters is how students use it and whether they remain intellectually engaged in their work. AI can help students explore ideas, improve their writing, or communicate across languages. However, it should not replace their capacity to interpret, argue, and develop their own perspectives. I would encourage students to acknowledge their use of AI and, more importantly, reflect on how it influenced their thinking.

Third, equity and linguistic diversity. This is particularly important in my context. Students do not have equal access to digital technologies, and many AI systems privilege English and knowledge produced in the Global North. As someone working with internationalization, I am concerned that AI might reproduce the very hierarchies we are trying to challenge. Guidance should recognize linguistic diversity, support multilingual practices, and ensure that students are not disadvantaged because of their economic circumstances or linguistic backgrounds.

I find UNESCO’s human-centered approach particularly valuable, especially its emphasis on inclusion, human agency, and cultural diversity. I would also adopt the emphasis on transparency and academic responsibility found in Swedish university guidance. However, I would avoid transferring these recommendations mechanically to Brazil because our institutional realities, inequalities, and educational priorities are different. I would also move away from approaches that focus primarily on surveillance, misconduct, or detecting AI-generated texts. Such approaches risk creating fear rather than meaningful learning.

If I could make one important change at my university context, I would establish a participatory AI literacy program involving students and teachers. Instead of simply publishing rules, the university should create spaces where we can experiment, discuss ethical dilemmas, and critically reflect on AI in our own educational contexts.

I want students to understand that responsible AI use is not simply about following institutional rules. It is about developing the confidence and responsibility to question technology, make informed decisions, and contribute their own voices to knowledge production. For me, that is what higher education should protect.

Imagine that you are asked to develop GenAI guidance for students in your own educational context. What three aspects would you consider most important to include, and why? What aspects of the Swedish or UNESCO guidance would you adapt or leave out? The final question: What is the single most important change your university could make to support responsible GenAI use, and why?

The three most important aspects would be the ethical and transparent use of AI, critical thinking and information verification, and data protection and privacy.

I believe that the whole proposal of GU's Course Policy for GenAI is very relevant and adaptable to my context, and I would not leave anything out. However, the aspects I found most important were the recommended, problematic, and prohibited categories (although I would move some things from one category to another. For example, I would change the use of GenAI functions as a teacher from permitted to problematic) and the disclosure statements, because, besides knowing how and whether AI was used, we can also understand which aspects made students feel the need to use it and their feelings about using AI in their work. I also find the emphasis on critical thinking important, since cognitive struggle is what makes students learn.

The most important change my university could make would be to develop a policy, but always through a democratic process, listening to both students and professors, and in a very clear and transparent way, so that everyone can really understand the policy and how it applies in practice.

What Should GenAI Guidance for Students Look Like?

Introduction

Generative artificial intelligence (GenAI) is transforming higher education by offering new opportunities to conduct research, organise ideas, understand concepts and improve academic writing. However, its use also raises concerns about academic integrity, information reliability, privacy and students’ intellectual autonomy. In my view, universities should consider GenAI not only a technological tool but also a pedagogical and ethical challenge. Therefore, it is essential to develop clear, accessible guidelines adapted to students’ educational contexts.

First aspect: Responsible use and academic integrity

The first aspect I consider essential is establishing clear rules about when, how and for what purposes students may use GenAI. The guidelines should distinguish between uses that support learning, such as clarifying concepts, developing study questions and improving the organisation of a text, and uses that undermine learning, such as submitting an entire AI-generated assignment as the student’s own work.

From my perspective as a student, these guidelines are important because many students may be uncertain about the boundaries between technological assistance and academic dishonesty. Universities should explain how students must disclose their use of AI, including identifying the tool used and describing its contribution to the assignment. This approach promotes honesty, responsibility and intellectual authorship.

The University of Gothenburg in Sweden provides a relevant example by emphasising that AI rules should be communicated clearly and adapted to the learning outcomes of each course.

Second aspect: Critical thinking and information verification

The second essential aspect is developing students’ ability to critically evaluate content generated by AI. Although these tools can provide quick answers and well-organised texts, they may also produce inaccurate information, fabricated academic references and biased interpretations.

For this reason, students should learn to compare AI-generated responses with scientific books, academic articles, official documents and other reliable sources. Students must continue to interpret information, question assumptions and construct their own arguments rather than automatically accepting the answers produced by technology.

UNESCO’s 2023 guidance emphasises a human-centred approach that considers ethics, safety, equity and meaningful uses of AI in education. This principle should be adapted to Mozambican universities by encouraging independent research, critical thinking and students’ intellectual development.

Third aspect: Digital inclusion, privacy and equal access

The third essential aspect is ensuring that GenAI use respects privacy and does not deepen educational inequalities. Not all students have access to computers, reliable internet connections, sufficient digital skills or paid versions of AI tools.

In the Mozambican context, these differences must be considered when developing institutional guidelines. Universities should ensure that access to particular technologies does not create an unfair advantage in assessments. They should also advise students against entering personal data, confidential documents or sensitive information into AI platforms without adequate data protection guarantees.

Drawing on UNESCO’s recommendations and Swedish university guidance, institutions should promote digital literacy, provide accessible learning resources and ensure that compulsory AI-based activities do not disadvantage students with fewer technological resources.

What should be adapted from Swedish and UNESCO guidance?

I would adapt the Swedish universities’ emphasis on transparent course-specific rules, clear explanations of permitted and prohibited uses, and students’ responsibility for submitted work. From UNESCO, I would adopt the human-centred approach, data protection, equity and the development of skills for ethical technology use.

However, I would avoid applying international guidelines automatically without considering local educational conditions. Universities should adapt these recommendations to their available infrastructure, students’ digital skills and the resources accessible to their communities.

Conclusion: The most important change

The single most important change my university could make would be to introduce a clear institutional policy on the responsible use of GenAI, accompanied by practical training for students and lecturers.

This change is essential because it would reduce uncertainty, promote academic integrity and help students use technology to support learning rather than replace human thinking. Through clear rules and appropriate training, universities could prepare students to become more critical, independent, responsible and capable of participating in an increasingly digital society.

References  

UNESCO MIAO, Fengchun; HOLMES, Wayne. Guidance for generative AI in education and research. Paris: UNESCO, 2023. DOI: https://doi.org/10.54675/EWZM9535. Disponívem : 23 out. 2026.

Universidade de Gotemburgo, Suécia

UNIVERSITY OF GOTHENBURG. Guidance for students on the use of generative AI in their studies. Gothenburg: University of Gothenburg, s.d. Disponívem: 23 out. 2026.

MIAO, Fengchun; HOLMES, Wayne. Guidance for Generative AI in Education and Research. Paris: UNESCO, 2023.

UNIVERSITY OF GOTHENBURG. Guidance for Students on the Use of Generative AI in Their Studies. Available at: https://studentportal.gu.se/en/your-studies/rights-and-responsibilities/guidance-for-students-on-the-use-of-generative-ai-in-their-studies.

UPPSALA UNIVERSITY. Generative AI and Studies. Available at: https://www.uu.se/en/students/it-for-students/generative-ai-and-studies.

1. If I were asked to develop a guidance document on the use of GenAI for students in my educational context in Mozambique, I would consider three fundamental aspects.

First, I would include the ethical and responsible use of GenAI, guiding students on how to use these tools without compromising academic integrity, avoiding plagiarism, and acknowledging the contribution of artificial intelligence to their work.

Second, I would prioritise the development of critical thinking and independent learning. I believe it is important for students to learn how to analyse, question, and verify information generated by GenAI rather than accepting every answer as true or relying exclusively on these tools to complete their academic tasks.

Third, I would address data protection and privacy, raising students’ awareness of the risks associated with sharing personal, academic, or confidential information on artificial intelligence platforms.

2. Regarding the guidelines developed in Sweden and by UNESCO, I would adapt the principles related to ethical use, academic integrity, critical thinking, and data protection to the reality of Mozambican educational institutions. I believe these international experiences can serve as a reference for developing guidance that meets our needs.

However, I would avoid fully adopting guidelines that depend on advanced technological resources or institutional conditions that are not yet sufficiently developed in Mozambique. Therefore, I would take into account limitations in internet access, the availability of technological equipment, and the level of digital literacy among students and teachers. In my view, the guidance should be inclusive, accessible, and adapted to the national context while continuing to respect fundamental ethical principles.

3. In my opinion, the most important change my university could implement to promote the responsible use of GenAI would be to establish a clear institutional policy, supported by training initiatives for both students and teachers.

I consider this measure a priority because it would establish common rules for using GenAI in academic activities, clarify what is and is not permitted, and guide the university community on how to disclose the use of these tools while preserving academic integrity. Furthermore, training would help users better understand the potential benefits and limitations of GenAI.

I believe that, in this way, the university could promote a culture of conscious, ethical, and responsible use of artificial intelligence, making the most of its benefits to improve teaching and learning without compromising students’ critical thinking, autonomy, and creativity.

From my experience as a student, what works well for me is that this tool has been very helpful in understanding content that I find more complex. It has supported me in organizing my ideas, conducting academic research, and preparing lessons and slides. However, it is important that, as a student, I continue to think independently and do not allow AI to think for me, since not all the answers generated by AI are reliable. What works well is that it makes tasks more efficient and helps organize work. What is still lacking, however, is greater confidence in the accuracy and reliability of the information provided in response to queries.

One of the main lessons we can learn from UNESCO's guidelines and the Swedish guidelines is the importance of promoting the ethical, critical, and responsible use of AI without undermining the role of teachers or the responsibility of students. Therefore, I believe that the most important change my school could implement would be to establish clear institutional guidelines, accompanied by practical training for both teachers and students. This measure would help develop the skills needed to use AI responsibly, assess the credibility of its responses, and preserve critical thinking and academic authorship.

Option 2 — What should AI guidelines for students look like?

In my view, AI guidelines for students should include the following:

1. Ethical use and academic integrity: Students should be taught how and when to disclose their use of AI in research assignments and reports.

2. Verification of information and development of critical thinking: Students should learn to verify information generated by AI and critically evaluate the content presented in their work.

3. Practical training for teachers and students: Students should receive practical training on how to use AI rationally and responsibly.

Regarding the Swedish guidelines and those developed by UNESCO, I would adapt the principles related to ethical use, inclusion, and human supervision to the specific conditions of my institution and the technological resources available. I would avoid simply copying guidelines designed for other contexts, since schools have different levels of access to technology, digital literacy, and pedagogical needs. I would not eliminate fundamental principles of responsibility and protection; instead, I would seek to translate them into practical, accessible rules that can be applied to the local context.

The most important change would be the implementation of an institutional policy for the responsible use of AI, combined with ongoing training programmes for teachers and students. I consider this measure a priority because simply prohibiting or permitting AI does not guarantee its appropriate use. It is necessary to develop the skills required to recognize its limitations, verify information, and use technology as a support tool without replacing critical thinking, creativity, or human responsibility.

The three aspects I would prioritize are: 1) Developing critical literacies and ethical awareness regarding the use of AI to consume and produce knowledge and to structure learning processes; 2) Preserving human agency and capabilities while preventing potential overreliance or dependence on AI and the atrophy of cognitive skills resulting from interactions with and the use of AI tools; 3) Raising awareness of the non-neutrality of AI tools and the content they process and generate, as well as their roots in colonial rationalities, in order to promote and safeguard social, linguistic, cultural and epistemic justice.

With regard to UNESCO's guidelines, I would take all the recommendations outlined in Section 4 of the document into consideration. In particular, I would prioritize developing guidelines that safeguard human agency, as highlighted in Section 4.2. I believe that the risk that the uncritical use of AI interfaces may undermine or hinder the development of human intellectual capacities needs to be addressed carefully and responsibly, especially in educational settings. I also consider it essential to preserve learners’ intrinsic motivation to grow and learn independently, without delegating their learning processes to GenAI tools.

At Ifes, where I work, we don’t have guidelines for the use of AI yet. To provide some context, the Federal Institute of Espírito Santo has 23 campuses spread across the state, making it a diverse institution with a wide range of contexts and needs. Giving this, I believe the first step would be to collectively develop guiding documents or a policy grounded in the principles of ethics, critical thinking, and responsibility, while taking into account the specific contexts and characteristics of each campus and academic program of the institution. I believe this approach could make a meaningful contribution to the informed, critical, and responsible integration and use of AI interfaces, while safeguarding education's primary purpose: human development and the envisioning of fairer and more equitable futures.

In relation to adapting the Swedish and UNESCO Guidance, I believe that both Swedish and UNESCO guidelines offer valuable principles regarding transparency, critical thinking, and ethical use. However, in my context, I would adapt them to place greater emphasis on social justice, language, and power. As an applied linguist, I believe students should understand not only how AI works but also how AI language can shape social realities, reinforce inequalities, or privilege certain forms of knowledge over others. One aspect I would potentially reduce is an overly technical focus on how AI systems function internally because in my opinion most students benefit more from learning how to critically evaluate outputs and make ethical decisions than from detailed explanations of AI structure.

The most important change my university could make would be to develop an institutional AI literacy program for both students and faculty because I see that many current challenges arise not from AI itself, but from uncertainty about how it should be used, especially for educational/pedagogical practices. Thanks again for the excellent course! I  learned a lot from your classes, and I hope we meet again soon. Kind regards, Adriana.

In fact, I have many concerns, ranging from the educational sphere to geopolitical and power-related contexts.

In the educational field, my concern in research is the risk of researchers, especially early-career ones, outsourcing their critical thinking. If generative AI produces literature reviews and research questions too easily, students may skip the very processes that build their autonomy as researchers.

On a broader, geopolitical scale, I am very concerned about the concentration of this technology in the hands of a few companies that are accumulating unprecedented levels of capital. A handful of companies are accumulating the capital of entire nations. When AI development becomes a race for technological leadership, and the fact is that the United States and China are competing for this dominance, there is a risk that speed and market dominance will be prioritized over safety, transparency and the public interest. I know that today there are very advanced fronts in military applications, using AI to gain and maintain power, bringing this instrument into obscure matters that pose extreme risks to humanity itself. For education, this also raises questions about dependence on private platforms and about who defines how knowledge is produced and accessed, especially in the Global South. There is a huge risk of consuming everything through AI; after all, AI itself can manipulate our thoughts and validate itself at every level. I have never seen so many resources directed toward AI development. Whenever we talk about innovation, funding is allocated only if AI development is involved. Could this be an AI revolution in itself, concentrating more and more resources and power in the hands of a few people?

I often try to steer away from ideas that might seem unreasonable, but I feel a deep unease about the coming years, given that AI is developing along an exponential curve. If intelligence is what truly put humans in control (for example, we are not the strongest species, yet we dominate nature through our intelligence), what would happen if, at some point, AI surpassed human capabilities and became autonomous? Many technology developers have already acknowledged the possibility of AI becoming autonomous at some point in the future. In this scenario, would we still be able to maintain effective control over these systems, and who would be responsible for ensuring it?

In Brazil, the Ministry of Education (MEC) recently published, in 2026, the Framework for the Responsible Development and Use of Artificial Intelligence in Education. This document provides guidelines for the use of AI from early childhood education to graduate studies. The document is not legally binding, but it offers a series of recommendations for educational settings. In basic education there are more restrictions on use, while in higher education the possibilities of use are broader, which leaves graduate programs with the task of translating these principles into their own practices.

As I currently work in the context of graduate education, I believe we have a specific role to play in education and knowledge production. Graduate education exists to produce science and to generate new data and knowledge, which requires applying specific methodologies. That is why I believe that AI guidelines for graduate students should not only state what is allowed, but also show where AI fits into the research process and where it falls short.

This is exactly what my doctoral research is about: the education of researchers. I am developing a method focused on human-machine interaction, grounded in Vygotsky's Zone of Proximal Development and the concept of scaffolding, organized as a workflow in three main stages: mapping science, to select verified and peer-reviewed literature and build a reliable picture of the field; identifying gaps, through in-depth reading, immersion in the state of the art and extraction of research gaps; and finally, synthesis and contribution of new knowledge, proposing methodological approaches to fill these gaps and build new findings. So, focusing on graduate education, I would like to share some ideas.

I think transparency should be a key point. In scientific writing, for example, the use of verifiable sources should be an important rule. With generative AI, students need to take ownership of the scientific literature, so transparency in references should be a central point. In my doctoral development, on the platform available at https://insightsci.com/, the science mapping feature makes it possible to search for real references and build a library in which researchers add their insights and comments to each reference, allowing them to take ownership of the state of the art. Another important point is declaring methods, that is, how AI resources were used. In Brazil, for example, CNPq Ordinance No. 2,664/2026 already requires this in research, and I believe this practice should start early in education.

Another relevant point is human authorship and responsibility, since AI can support the process, but the student remains the author and is fully responsible for the content, including fact-checking and bringing their own contribution and expertise to the field. Ultimately, it is the researcher who discovers gaps, generates new data, conducts experiments and evaluates results. This can be done with the help of AI, but the construction of new knowledge must remain under human authorship.

I also highlight the importance of research methods with clear purposes and defined ways of using AI in research. For example, text revision can be done with AI, as can translation and many other repetitive tasks, but research itself, the exploration of gaps, new discoveries and the publication process involve relationships and networking.

Given all of the above, I believe that critical analysis by the researcher is also an important factor to consider. It is very common for AI models to provide highly standardized answers, with errors in many statements and a tendency toward a particular result. The machine will always deliver an answer, often even inventing references. Therefore, researchers need to understand how these tools work, their limitations, biases and the risk of hallucinations, as well as broader issues such as data privacy and the concentration of the technology in a few companies.

Another important point that my research has allowed me to evaluate throughout my PhD is that AI should function as an instrument of mediation, not as a substitute for human potential. From a Vygotskian perspective, AI should act as a mediating tool that helps students go beyond what they could achieve on their own, with the teacher providing pedagogical scaffolding. The goal is for students to become more capable over time, not more dependent.

Many other variables could be explored further within research, and I believe that, with the exponential advancement of AI, this reflection will always require new updates, since technology evolves faster than any static policy.