Helder Nakaya, a senior researcher at Einstein Hospital Israelita, discusses a study involving more than a thousand participants that provides evidence on how the use of artificial intelligence (AI) during learning can affect users’ behavior and performance. 

According to the paper “AI assistance reduces persistence and hurts independent performance”, published on the arXiv preprint platform at Cornell University, participants completed a series of math and reading-comprehension tasks of increasing difficulty.

Divided into two groups, one was allowed to use AI tools while learning and solving exercises, while the other was not.

During the assisted phase, the group using AI demonstrated better immediate performance—they answered more questions correctly with the tool’s help.

However, when scientists evaluated participants on subsequent tasks without access to AI, the picture changed: those who had used AI showed less persistence, leaving more questions unanswered and performing worse than the group that had never used the tool.

The Einstein researcher emphasizes that how AI was used made a difference. Within the group that had access to the tool, some participants requested ready-made answers, others used AI to ask for hints and explanations, and some chose not to use it at all.

Those who used AI only as an explanatory aid or for hints showed greater effort in attempting to solve problems on their own and achieved better results than those who requested ready-made answers.

Using it as a shortcut to ready-made answers, however, tends to reduce persistence and may compromise long-term learning.

Watch the video to see the data, the charts comparing the groups, and the researcher’s detailed explanation of how to integrate AI into study without sacrificing essential skills.

Opinion articles and videos do not necessarily reflect the views of Science Arena or Einstein.

O post Can AI use affect learning? apareceu primeiro em Science Arena.

The findings, published on February 3 in the journal Einstein, reveal consistent performance gains between earlier and newer versions of these platforms.

In total, the study analyzed 464 multiple-choice questions submitted to four large language models: ChatGPT 3.5, ChatGPT 4.0, Gemini (Bard), and Gemini 2.5 Flash.

The questions were drawn from entrance exams administered in 2024 by medical residency programs at UNICAMP; the University of São Paulo (USP) campuses in São Paulo and Ribeirão Preto; the Federal University of São Paulo (UNIFESP); the Institute of Medical Assistance to São Paulo State Public Servants (IAMSPE-SP); and the São Paulo State Unified Health System (SUS-SP).

Practical experiment with the models

Initially, the researchers selected 580 questions but excluded 116 that required interpretation of images or radiological exams. Because the models evaluated operate primarily in text, including image-based questions could have introduced bias.

The final dataset consisted of 464 questions, each with a single correct answer.

Each question was copied in full and entered individually into the models’ interfaces. The page was refreshed after each question to minimize the potential effect of contextual memory.

The researchers applied statistical tests (chi-square and t-test) to compare performance across versions and to contrast the results with averages reported in international studies.

The questions were also categorized by institution of origin and cognitive typology—the type of skill required to answer them: conceptual, diagnostic, behavioral/management, or mixed.

High Performance on the Exams

On average, ChatGPT 3.5 correctly answered 55.4% of the questions analyzed (257 out of 464). Its performance was slightly higher than that of Gemini (Bard), which achieved 51.1% accuracy (237 correct responses).

By contrast, the most recent versions demonstrated a marked improvement. ChatGPT 4.0 achieved an average accuracy rate of 77.6% (360 correct answers), while Gemini 2.5 Flash reached 81% (376 correct answers).

Statistically, there was no significant difference in performance across question categories. Nevertheless, qualitative trends were observed.

In addition, agreement between models increased in the newer versions. The rate of correct responses shared by ChatGPT 4.0 and Gemini 2.5 Flash was higher than that observed between their earlier iterations, indicating a possible convergence in performance on standardized tasks.

Tool to Support Medical Training

The authors emphasize that, although the results demonstrate robust performance on standardized examinations, these models do not replace traditional clinical training.

Even so, their educational potential is clear. In a context where medical residency exams rely heavily on multiple-choice questions, systems capable of explaining answer choices, synthesizing content, and simulating clinical reasoning may serve as valuable tools for training and review.

Reference

Figueiredo MC, Diniz VH, Granado AC, Paulino GC, Oliveira GR. Performance of the Artificial Intelligence large language models ChatGPT 3.5, Gemini (Google Bard), ChatGPT 4.0, and Gemini 2.5 Flash in surgical subspecialty questions of Brazilian medical residency exams.einstein (São Paulo). 2026;24:eAO1436. https://dx.doi.org/10.31744/einstein_journal/2026AO1436

O post ChatGPT 4.0 scores 77% and Gemini 2.5 scores 81% on medical residency exams apareceu primeiro em Science Arena.

The study, involving 52 trainee teachers (with an average age of 20) and 21 teacher training professors (with an average age of 54 and average of two decades of experience), was conducted at a university in the northwestern USA in January and February 2024. The data reveal minimal use of generative AI in teacher training:

• Among trainee teachers: 48 do not use the technology in their courses; 49 have never received training on how to use it; and only 3 reported having received any type of guidance.
• Among teaching course professors: 18 do not use AI in their classes; 18 have not received training on how to do so; and only 9 allow students to use the tool, albeit with limitations, such as for brainstorming or drafting lesson plans.

The low level of adoption is reflected in self-assessments: on a scale of 0 to 10, trainee teachers rated their AI literacy at an average of 4.2, while teachers scored themselves at 3.9.

AI: a mixture of enthusiasm and distrust

Despite having limited hands-on experience, many trainee teachers see advantages in using AI, such as saving time, generating ideas, and learning new content. Educators also cited benefits, including greater technological awareness and assistance in creating teaching materials.

Both groups also expressed concerns, however. 

Students worry that overuse could lead to superficial learning and dependence, while teachers highlight issues of academic integrity, plagiarism, and the learning curve required to master the technology.

Focus groups signaled a certain tension between optimism and resistance. 

One future English teacher said she was interested in using AI to create starting points for learning but warned about the risks of irresponsible use. A trainee mathematics teacher said he was opposed to adopting it in the classroom, although he acknowledged that his students “could be left behind” if he ignored AI entirely.

Barriers and pathways to AI adoption

The study was structured around Everett Rogers’s Diffusion of Innovation theory, which evaluates attributes that facilitate or hinder the adoption of new technologies. 

With regard to AI in education, the respondents recognized clear advantages but reported low compatibility with current practices and a lack of formal training, limiting the perceived benefits.

To overcome these barriers, Panday-Shukla developed a pilot workshop at the WSU Global Campus, inspired by a matrix created by the state education agency (OSPI). 

The workshop proposes four levels of use, from total prohibition to mandatory use in certain activities, remaining transparent at all times about what is or is not permitted.

Implications for the workplace

The AI debate is not confined to the classroom. The study cites McKinsey Global Institute projections that up to 30% of US working hours could be automated by 2030, forcing millions of workers to seek new career paths. In this context, resisting AI and failing to properly prepare could harm not only future teachers, but also the students they will go on to educate.

The key recommendation is to invest in digital literacy and AI for trainee teachers and their educators through training programs, clear institutional guidelines, and spaces for ethics debates. 

For Panday-Shukla, preparing future educators to teach in a world already being shaped by artificial intelligence is a matter of educational justice.

Diffusion of innovations applied to AI in education

O post Educators and trainee teachers want to use AI but lack the necessary training apareceu primeiro em Science Arena.

But a score is needed to organize the ensemble, to give meaning and direction to the harmony we want to produce.

This requires a paradigm shift in medical education that invites us to look beyond the mere sum of the content and to clearly define what educators and society expect from the doctors we train. 

This is the paradigm of competency-based medical education, which seeks to integrate knowledge, skills, and attitudes to promote professional and reliable practice, resulting in safe, responsible, and socially committed healthcare.

The training was fragmented, varying enormously between institutions and with no established quality standards. The Flexner report brought the scientific revolution that the twentieth century demanded: curricular standardization, laboratories, research, and methodological rigor. It was a decisive turning point that established medicine as a scientific profession.

However, the Flexner model also solidified medical education as a content-heavy, discipline-based structure. Students learned a lot of information, but they were not always prepared to apply it in an integrated manner in complex care situations.

Medical schools thus began to reproduce an approach of excellence in science, but limited in its connection with professional practice and social demands.

Paradigm shift

The transition now underway is based on changes in the perception of medical education and the needs of society. One of the first signs of this transition was given by the World Health Organization (WHO) in 1978, in a document that presented a competency-based curriculum.

Another important milestone was the “To Err Is Human” report, published by the US Institute of Medicine in 1999, which exposed the extent of preventable errors in healthcare. The report highlighted the need for training to go beyond transmitting knowledge, and the need to educate students on patient safety, social responsibility, and complex health systems.

This movement spurred the shift in medical education towards competency-based teaching, in which competencies are individual attributes that professionals need to demonstrate as a whole, including not only technical knowledge, but also attitudes and interpersonal skills, such as professionalism, clinical communication, and collaboration as a team.

These competencies, however, were abstract concepts.

Then, in 2005, a framework known as Entrustable Professional Activities (EPA) was proposed, describing concrete medical practices that are observable and time-limited, allowing supervisors to decide whether a trainee is capable of performing them independently.

These competencies represent how each individual masters their own instrument, while the EPAs make the music audible, meaning it can be observed, trusted, and evaluated in real practice.

This integration makes it possible to assess competencies within the real world of healthcare—something that previously seemed impossible.

Impact on patient care

Migrating to competency-based learning and EPAs is about more than simply changing the curriculum. It is a cultural transformation in how medical care is provided.

When it is grounded in practical outcomes, education increases safety in four key dimensions:

• For patients, it reduces avoidable errors and increases confidence in care.
• For trainees, it ensures supervised progression, reducing the risks of premature and unsafe independence.
• For professionals, it improves confidence and preparedness in increasingly complex health systems.
• And for society, it helps train doctors aligned with the real needs of the health system.

In Brazil, an example of alignment with this transition can be seen at Einstein Hospital Israelita, where EPAs are adopted as evaluative criteria for medical students and to monitor the progress of medical residents. In continuing education, training programs are designed around competencies.

In some areas of medical practice at Einstein Hospital Israelita, such as intensive care and primary healthcare, EPAs have been proposed as a way of supporting the creation of personalized development plans for professionals working at the institution.

The aim is to provide structured, career-long support—an innovative approach both in Brazil and internationally.

Family and Community Medicine

In 2025, the Brazilian Society of Family and Community Medicine (SBMFC) published the National EPAs in Family and Community Medicine, a landmark that clearly outlines the expected activities of an expert in the field.

To support their implementation in residency programs, the SBMFC released a Theoretical and Practical Guide that explains the evaluation methods and offers guidelines for applying them during on-the-job training.

According to the Ottawa criteria (an internationally accepted expert consensus), a good assessment is one that presents validity, reliability, feasibility, educational impact, catalytic effect, and acceptability.

The assessment is thus about more than simply measuring performance: it is about taking care of the learning process, guiding progress, and responsibly determining the autonomy of students and residents at each stage of their training.

Cultural change

The transition from content-focused teaching to competency-based medical education must align with the needs and culture of each country—simply copying external models is not enough.

Brazil has developed its own solutions, such as initiatives by specialist medical societies and successful local experiences, which illustrate that it is possible to transform conceptual proposals into real tools and actions.

The challenge now is to expand these practices nationwide, to establish a medical education system aligned with safe and legitimate decisions about the professional autonomy of doctors in training.

This change is not just pedagogical, but cultural. It involves guiding medical training to respond to the demands of health systems and promoting safer healthcare that is committed to the needs of society.

More than teaching single notes of knowledge, training doctors is about building a symphony in which professional practice, knowledge, attitudes, and clinical and relational skills are integrated in harmony to generate high-quality care.

A transition to competency- and outcomes-based learning can help fine-tune our instruments to produce a sound that resonates both with patient safety and with the responsibility and commitment of medical schools and institutions to society.

Leonardo de Andrade Rodrigues Brito is a family and community doctor. He earned his master’s degree and PhD in clinical medicine, with a focus on health education, at the University of Campinas (UNICAMP). He currently works as an academic mentor for Einstein’s Family and Community Medicine Residency and as coordinator of the hospital’s postgraduate course in medical teaching and mentoring. He is also a teaching and research assistant at the University of Campinas (UNICAMP).

Opinion articles do not necessarily reflect the views of Science Arena or Einstein Hospital Israelita.

O post What should we expect from the doctors we train? apareceu primeiro em Science Arena.

More information is available at Science Arena Opportunities.

The agenda includes workshops and talks by experts such as Galina Gheihman, professor of neurology at Harvard Medical School; Priscila Cruz, president of the Todos Pela Educação (“All for Education”) movement; and Welbert Pereira, head of the Postgraduate Program in Bioengineering at Einstein Hospital.

Click here to see the full list of speakers.

Teacher development

The aim of the symposium is to stimulate reflections on pedagogical practices capable of meeting the demands of a changing world.

Some of the topics set to be addressed include social and emotional development, interprofessional teaching, and the implications of artificial intelligence (AI) in education.

“Some of the key skills for development in teaching, especially in higher education, seek to encourage more innovative and student-centered teaching,” explains Dr. Elda Maria Stafuzza Gonçalves Pires, academic coordinator of Einstein’s undergraduate medical degree and a member of the event’s organizing committee.

Below, Stafuzza lists the gaps and demands in teacher training, which will be discussed in more depth during the symposium:

1. Student-centered teaching strategies:

– Understand and adopt approaches that place the student at the center of the process, surpassing the expository model in which the teacher is the “owner” of knowledge.

– Teachers can act as facilitators in the classroom, challenging students, proposing problems and questions that stimulate critical thinking—which is applicable to any field of knowledge.

– Specific pedagogical strategies include team-based learning, gamification, and the use of simulations.

– Training on the use of AI as a form of support in the classroom or for preparing lessons.

– In medicine, key strategies include teaching and assessment in a multidisciplinary clinical environment (such as bedside teaching in outpatient clinics or wards) and microteaching as a means of analyzing speaking skills and receiving feedback.

2. Development of mindset and interpersonal skills:

– Teachers can learn to create a welcoming environment, where students feel comfortable sharing thoughts and asking questions.

– It is possible to improve skills related to classroom control and dynamics, which are important to leading collaborative classes in which students actively participate and make choices and decisions, as well as justifying them in discussions.

3. Assessment and feedback skills:

– Student assessment needs to evolve beyond memorization, instead focusing on evaluating their reasoning and ability to articulate information and make inferences.

– For courses that require the development of techniques, skills, and competencies (such as medicine), teachers need to know how to “diagnose” a student’s learning and give feedback.

– In the context of medicine, it is essential to teach clinical reasoning, which involves putting the pieces of the puzzle together to reach a diagnosis. This includes using open-ended questions to understand the difficulties faced by students (collecting information, interpreting data, or conducting exams) and recognizing biases they may hold, such as anchoring on their first hypothesis.

4. Encouraging research:

– Teacher development, with a focus on student autonomy, can spark interest in scientific research among students, stimulating their curiosity and creativity.

– Contextual teaching, which shows students the practical relevance of what they are learning, avoids memorization and stimulates reasoning.

– It is important for students to develop self-regulated learning and metacognition, enabling them to look inward, perceive what they know and what they do not, identify gaps, and seek improvement.

O post Four teacher development strategies apareceu primeiro em Science Arena.