Proactive Push versus Reactive Response Artificial Intelligence for Thyroid Cancer Guideline Learning Among Resident Physicians: A Nationwide Online Randomized Controlled Trial
In a nationwide randomized controlled trial among resident physicians, proactive push AI tutoring did not significantly improve immediate total knowledge scores compared to reactive chatbots but demonstrated superior benefits in delayed retention, specific module performance, engagement, and self-efficacy.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the world of medical training, learning is often a race against time and memory. Doctors in training, known as residents, must absorb vast amounts of complex information about diseases, treatments, and guidelines, often while rotating through different hospital departments. The challenge is not just knowing the facts when they are fresh, but remembering them weeks or months later when they are needed in a real emergency. For decades, educators have relied on a simple principle: to make knowledge stick, you must practice retrieving it from your mind over and over again, spaced out over time, rather than cramming it all at once. This is the difference between reading a map once and navigating the same route repeatedly until the turns become second nature. As artificial intelligence enters the classroom, a new question arises: does it matter how the AI helps? Does it make a difference if the AI waits for the student to ask a question, or if the AI actively nudges the student to practice at regular intervals?
A team of researchers in China set out to answer this question by testing two different ways of using AI to teach thyroid cancer guidelines to resident physicians. Thyroid cancer is a common disease in China, and its treatment guidelines are updated frequently, requiring doctors to stay current with the latest evidence. The researchers recruited 120 residents from hospitals across the nation and split them into two groups. Both groups used the exact same artificial intelligence system, which was built on a powerful language model and contained the same medical knowledge and quizzes. The only difference was how the system interacted with the students. One group used a "reactive" system, which acted like a traditional chatbot: it sat quietly waiting for the student to type a question or request a quiz. The other group used a "proactive" system, which acted like a diligent tutor: it sent two short, personalized quizzes to the students every single day, regardless of whether the students had asked for them.
The study ran for three weeks, during which the proactive group received these daily nudges, while the reactive group had to initiate all their own learning sessions. The researchers wanted to see if the proactive approach led to better test scores immediately after the training and, more importantly, if it helped the students remember the information weeks later. When the training ended, the researchers tested the students on their knowledge of thyroid cancer guidelines and how to manage drug side effects. The results showed a nuanced picture. The group that received the daily proactive nudges did not score significantly higher on the immediate final test compared to the group that had to ask for help. The difference in their total scores was so small that it could have happened by chance. However, the story changed when the researchers looked at how the students performed a few weeks later.
When the students were tested again four weeks after the training ended, the proactive group remembered significantly more information than the reactive group. They also performed better on specific questions about drug side effects, a critical area where mistakes can be dangerous. Beyond the test scores, the way the students behaved told a clear story. The proactive group logged into the system on nearly twice as many days as the reactive group and completed three times as many quizzes. They asked more questions and reported feeling more confident in their ability to apply the guidelines in real clinical situations. The students in the proactive group also rated their learning experience more highly, feeling that the tool was more helpful and personalized.
The researchers concluded that while the proactive AI did not instantly boost the total test score at the end of the training, it successfully changed the rhythm of learning. By sending regular, low-effort prompts, the system turned fragmented moments of the day into structured learning opportunities. This constant, distributed practice helped the students retain the information longer and feel more capable of using it. The study suggests that for medical education, the value of an AI tutor may not lie in how smart it is or how much it knows, but in how it encourages the student to keep practicing. The proactive approach did not force the students to learn more in a single sitting; instead, it helped them learn more consistently over time, turning a passive tool into an active partner in their education.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.