Perceptions and Intentions to Use Generative AI Among First-Year Medical Students in Japan: Cross-Sectional Survey Study

Introduction

Generative artificial intelligence (GenAI), particularly large language models like ChatGPT (OpenAI), has emerged as a tool for brainstorming, information gathering, proofreading, translating, and other academic tasks [1]. The rapid evolution of ChatGPT has made it essential for medical students to understand and use these tools [2]. As digital technologies become more embedded in health care education, understanding how future physicians will engage GenAI is vital. Many students demonstrate familiarity with digital tools and are exposed to them during childhood. One study showed that medical students had positive attitudes toward GenAI, albeit with low use rates [3]. In Japan, although some educators have adopted GenAI, information on how incoming medical students educated during the GenAI boom perceive these tools remains limited [4]. Thus, we explored GenAI use, learning behaviors, and perceptions of first-year medical students in Japan.

Methods

Overview

An anonymous online survey targeting first-year students immediately after high school graduation was conducted at the Chiba University School of Medicine in April 2025; 120 students were invited to participate and 118 valid responses were received (response rate 98.3%).

The questionnaire comprised four sections: (1) demographics; (2) prior GenAI exposure; (3) willingness to learn and perceptions (16 items rated on a 5-point Likert scale [1=strongly disagree to 5=strongly agree], based on prior studies [5,6]); and (4) an open-ended item on academic use. The items were pilot-tested for clarity and reliability.

Group differences were examined using independent 2-tailed t tests and chi-square tests. Spearman correlations were used to identify factors associated with learning motivation. Internal consistency (Cronbach α) was 0.81 for risk awareness (6 items), 0.81 for benefit recognition (6 items), and 0.53 for concern items (4 items; reverse-coded). Statistical significance was set at P<.05. Two independent coders performed content analysis of open-ended responses, with substantial agreement (κ=0.67). Analyses were conducted using JMP Pro (version 18; JMP Statistical Discovery LLC).

Ethical Considerations

This study was approved by the ethics committee of Chiba University Graduate School of Medicine (approval 3425). Electronic informed consent was obtained prior to participation. The study database was anonymized. Study data are stored on an offline, password-protected computer with full-disk encryption. Participants received no compensation for this study. All methods were in accordance with the relevant guidelines and regulations.

Results

Among 118 responses (n=79, 66.9% male; mean age 18.5, SD 0.7 years), 84.7% (n=100) reported GenAI use, mainly for language learning (n=53, 44.9%) and information searches (n=46, 39.0%). Only 49.2% (n=58) of the participants had learned about GenAI, primarily through browsing (n=34, 28.8%) or through peers (n=24, 20.3%).

Students demonstrated high willingness to learn (mean score 4.3, SD 0.9), with good understanding (mean scores 3.9‐4.6) and positive attitudes, especially regarding the relevance of GenAI (mean score 4.3, SD 0.8) (Table 1). Perceived individualization was lower (mean score 3.5, SD 1.0), with concerns over educational impact (mean score 3.6, SD 0.9).

Willingness to learn correlated positively with expectations of GenAI’s benefits, such as enhancing digital literacy and career preparedness (both P<.001).

Students with learning experience were slightly younger (mean age 18.4, SD 0.7 vs 18.7, SD 0.8 years; P=.02) and demonstrated greater awareness of potential bias in GenAI outputs (P=.045).

The attitude toward using GenAI for assignments was neutral (mean score 3.0, SD 1.0). Open-ended responses were categorized into 3 attitudinal groups, “positive,” “cautious,” and “negative,” reflecting enthusiasm, balanced concerns, or skepticism (Table 2).

Discussion

Most medical students had used GenAI before university; however, structured learning was limited. Minor differences in perceptions between those with and without learning experience suggest that casual exposure alone may be insufficient to develop a critical understanding of GenAI, highlighting the need for foundational GenAI education in universities [7].

Although the students were interested in learning about GenAI, they remained cautious about using it for assignments, reflecting varied levels of familiarity and trust. While most viewed GenAI positively, some expressed concern that overreliance could hinder creativity and critical thinking.

GenAI dependence may hinder originality and decision-making [8], whereas trust in GenAI may foster motivation and proactive learning [9]. Students’ neutral views suggest uncertainty or low confidence regarding appropriate use, underscoring the need for a balanced education that addresses concerns and fosters critical evaluation skills. This aligns with previous studies highlighting the importance of reforming curricula to integrate GenAI-related competencies, including ethical reasoning, clinical relevance, and communication skills [10].

A limitation of this study was that only first-year students at a single institution were included. Further, the “concern” subscale had relatively low internal consistency, and qualitative analysis relied on a single open-ended item. Finally, multiple comparisons were performed without correction, increasing the risk of type I errors. Nevertheless, this exploratory study aimed to identify trends rather than test specific hypotheses.

In conclusion, our findings support the inclusion of structured GenAI curricula in higher education, suggesting that such programs go beyond technical training to address students’ expectations, values, and concerns.