Throughout history, medical education has undergone several revolutions. One of the most significant began with the Flexner Report⁽¹⁾, published in 1910. Abraham Flexner, an American educator, evaluated medical schools in the United States and Canada, criticizing the poor quality of medical education at the time and recommending a model based on science, well-structured laboratories, and supervised clinical learning. This report influenced medical education in Brazil, leading to the adoption of a rigid curriculum^(2-4).

In Brazil, the Flexnerian model prevailed until 2001, when the National Curriculum Guideline (NCG) proposed changes, discouraging fragmented hospital-centered teaching and introducing a minimum curriculum based on objectives—indicating the essential content for medical training. In 2014, a new NCG⁽⁵⁾ brought another transformation by recommending competency-based education and active teaching/learning methodologies, encouraging students to develop critical thinking and clinical problem-solving skills.

Objective-based education prioritizes content delivery structured by the teacher or curriculum, focusing on knowledge of the human body and diseases. Competency-based education, in turn, aims to develop practical skills applicable to real-world settings, preparing physicians to meet society's needs.

In addition, medical education has incorporated key concepts such as:

– Humanization of medicine, valuing empathetic contact with patients and addressing ethical, psychological, and social aspects of care;

– Evidence-based medicine, ensuring that physicians know how to critically interpret and apply scientific research, avoiding decisions based solely on tradition or personal experience.

Currently, the most recent revolution in medical education is driven by Artificial Intelligence (AI). Tools such as ChatGPT are making learning more accessible, interactive, and efficient⁽⁶⁾, supporting question-and-answer-based learning, enabling students to review medical concepts, and providing detailed explanations about diseases, treatments, and clinical procedures, without the need to search through multiple sources.

On the other hand, AI is also transforming the role of educators, making teaching more dynamic and efficient, with benefits such as:

– Creation of personalized teaching materials, speeding up the production of summaries, exam questions, and presentations;

– Simulation of clinical cases, allowing students to practice diagnosis and decision-making in simulated environments;

– Automation of administrative tasks, where processes such as grading objective exams, organizing schedules, and providing automated feedback can be streamlined, freeing up teachers' time to focus on practical teaching.

A major criticism and possible limitation of using these tools to optimize student learning and critical thinking is the risk of generative AIs spreading inaccurate information. However, in a supervised academic environment, teachers can assess the reliability of the provided content.

Nevertheless, it is essential to adapt medical training to include best practices in the use of artificial intelligence. Students must understand the fundamentals of online information retrieval, the potential biases in the process, the tool's limitations, and the risk of "hallucinations," as well as develop strategies to use AI critically^(7,8).

In the context of generative artificial intelligence, the term "hallucinations" refers to the production of false, inaccurate, or nonexistent information presented convincingly by the tool. This phenomenon is inherent to the functioning of language models, which, by predicting the most likely sequence of words, can generate responses that are coherent in form but incorrect in content. These "hallucinations" may take different forms, from fabricated data to nonexistent bibliographic references, and are especially problematic when the user trusts the response without verifying its accuracy.

In particular, the trap of AI-generated "hallucinations" can mislead both experienced professionals and those in training. As generative models become capable of performing increasingly complex tasks, the risk of "hallucinations" also increases, reaching up to 20% in some studies⁽⁹⁾. Using the tool as a complementary resource—especially in contexts where the user already has consolidated knowledge—combined with careful prompt design and restriction of consulted databases, represents an effective strategy to mitigate this risk.

In a preliminary version of a research, Kosmyna⁽¹⁰⁾ showed, by dividing volunteers tasked with writing texts under electroencephalographic monitoring, that the use of ChatGPT generates neural connectivity patterns markedly different from those observed in the absence of the tool. Participants who did not use AI exhibited stronger and broader neural networks, while those who used it showed weaker neural, linguistic, and evaluative performance. This finding may support the hypothesis of reduced cognitive autonomy associated with systematic AI use⁽¹¹⁾, where excessive user reliance leads to overlooking even basic errors. This risk becomes even more concerning in the context of medical training.

However, the most relevant finding of Kosmyna's study emerged in the final stage of the experiment. After three consecutive writing sessions with AI, participants were asked to write without the tool. On the other hand, those who had written three sessions without AI were invited to use it in the final stage. In this context, the group that began using AI after three sessions without prior exposure performed better: they demonstrated greater memory of the produced text and used the tool more tactically, reflected in more cohesive neural signatures. These results reinforce the idea that, when well applied pedagogically, AI can serve as an ally in consolidating learning.

More than a century after the Flexner Report redefined the course of medical education by emphasizing science and rationality as pillars of training, we are experiencing a new turning point. The incorporation of artificial intelligence, while it may threaten to reduce cognitive autonomy, also reveals—when properly applied—pedagogical potential capable of strengthening memory, fostering more sophisticated learning strategies, and consolidating essential competencies. It is therefore the responsibility of medical schools to ensure that AI is integrated as an ally in the educational process, without undermining critical thinking, intellectual autonomy, and the ethical commitment that sustain medical practice.

REFERENCES

1. Flexner A. Medical education in the United States and Canada: a report to the Carnegie Foundation for the Advancement of Teaching [Internet]. Bulletin No. 4. New York: Carnegie Foundation for the Advancement of Teaching; 1910 [cited 2025 Sep 4]. Available from: https://archive.org/details/medicaleducation00flexuoft

2. Flexner A. Medical education in the United States and Canada. From the Carnegie Foundation for the advancement of teaching. Bulletin Number for, 1910. Bull World Health Organ [Internet]. 2002 [cited 2025 Set 8];80(7):592-602. Comment in: Bull World Health Organ. 2002;80(7):592-3. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC2567554/

3. Boelen C. A new paradigm for medical schools a century after Flexner1s report. Bull World Heath Organ [Internet]. 2002 [cited 2025 Sep 4];80(7):592-3. Available from: https://pmc.ncbi.nlm.nih.gov/articles/PMC2567557/

4. Pagliosa FL, da Ros MA. O relatório Flexner: para o bem e para o mal. Rev Bras Educ Med. 2008;32(4):492-9.

5. Brasil. Ministério da Saúde. Resolução CNE/CES nº 3, de 20 de junho de 2014. Institui Diretrizes Curriculares Nacionais do Curso de Graduação em Medicina e dá outras providências [Internet]. Brasília (DF): MS; 2014 [citado 2025 Set 4]. Disponível em: https://www.gov.br/saude/pt-br/acesso-a-informacao/acoes-e-programas/pnsp/legislacao/resolucoes/rces003_14.pdf/view

6. Carvalho MR, Greco ME, de Souza DM. Uso do ChatGPT como ferramenta complementar de estudo e ensino no curso de medicina. Educ Pesqui. 2025;51(00):e288875.

7. Kara-Junior N. Value of critical literature analysis for knowledge development. Arq Bras Oftalmol. 2024;87(2):e2024-1008.

8. Kara-Junior N. The open access and dissemination of predatory journals. Arq Bras Oftalmol. 2024; 87(3):e2024-1009.

9. Metz C. A.I. is getting more powerful, but its hallucinations are getting worse. The New York Times [Internet]. 2025 May 5 [cited 2025 Sep 4]. Available from: https://www.nytimes.com/2025/05/05/technology/ai-hallucinations-chatgpt-google.html

10. Kosmyna N, Hauptmann E, Yuan YT, Situ J, Liao XH, Beresnitzky AV, et al. Your brain on ChatGPT: accumulation of cognitive debt when using an AI assistant for essay writing task [Internet]. Cambridge (MA): MIT Media Lab; 2024 [cited 2025 Sep 4]. Available from: https://www.brainonllm.com/

11. Gerlich M. AI Tools in society: impacts on cognitive offloading and the future of critical thinking. Societies, MDPI. 2025;15(1):1-28.

Submitted for publication: September 1, 2025.
Accepted for publication: September 8, 2025.

Funding: This study received no specific financial support.

Disclosure of potential conflicts of interest: The authors declare no potential conflicts of interest.