The deployment of artificial intelligence (AI) in healthcare is accelerating, simultaneously promising enhanced efficiency and raising profound questions regarding the human dimensions of care. Central issues emerge around empathy, shared vulnerability, and care ethics, as AI’s capabilities challenge the irreplaceable quality of authentic human-to-human healthcare interactions.This review aims to critically examine how AI’s incorporation into clinical practice influences core values of empathy, mutual vulnerability, and the ethical obligations of caregivers. The goal is to identify both the risks and opportunities presented by AI in upholding humanistic care, and to propose frameworks for ensuring technology remains a servant—not a substitute—for relationship-centered care.A systematic literature review was conducted using the Consensus platform, Semantic Scholar, PubMed, and other databases. From roughly 980 initial publications, 50 highly relevant and recent peer-reviewed studies were selected. Thematic analysis focused on interdisciplinary perspectives—philosophy, ethics, sociology, and clinical evidence—exploring theoretical foundations and practical impacts of AI on empathy, care ethics, and vulnerability.Findings reveal a robust consensus: AI can streamline clinical workflows and free provider time but cannot genuinely replicate the depth, reciprocity, or transformative impact of human empathy. Simulated empathy from chatbots or care robots is usually perceived as superficial, risking decreased trust and relational authenticity. Conversely, AI—used judiciously—may reinforce relationship-based care by enabling professionals to concentrate on emotional and ethical engagement. Shared vulnerability, central to therapeutic alliance, remains largely unprogrammable and is best supported through direct human connection. Additional risks include potential exacerbation of inequities, clinical depersonalization, and ethical ambiguities in algorithmic decision-making.Empathy and vulnerability are not merely "soft skills" but pivotal determinants of patient outcomes, trust, and justice in healthcare. Care ethics, alongside alternative frameworks (Ubuntu, co-design), offers critical guidance. The literature strongly advocates for rigorous ethical oversight and sustained co-creation with clinicians, patients, and communities—to ensure technology complements rather than displaces care.AI can support the humanization of care if developed and deployed within ethically robust, relationship-centered models. Ultimately, authentic empathy, recognition of vulnerability, and ethically guided practice must remain at the heart of healthcare, with AI as a tool enhancing—never replacing—the essence of care.

The integration of artificial intelligence (AI) into medical education is profoundly reshaping pedagogical approaches in general practice. Advanced AI tools—from intelligent tutors to augmented reality systems—enable personalized learning, enhance performance analysis, and foster reflective practice. However, these innovations bring significant challenges related to training, ethics, and educational accountability. In an era where digital proficiency is essential, examining the role of AI in preparing future general practitioners is both urgent and foundational for the profession.This study aims to explore how AI enriches learning for medical students and practitioners in continuing education, identify the observed pedagogical benefits, and examine the ethical, curricular, and organizational challenges crucial to its sustainable implementation.A narrative literature review (2020–2025) was conducted using PubMed, Scopus, and Consensus.app, focusing on the keywords "artificial intelligence," "medical education," "general practice," "simulation," and "ethics." Publications were analyzed through three lenses: educational innovation, curricular integration, and ethical considerations. Additionally, a thematic mapping of research trends was provided to highlight emerging domains in this field.Findings reveal that AI facilitates personalized learning trajectories (adaptive feedback, gap identification), enhances clinical simulation (via AR/VR), and supports data-driven clinical decision-making. Nonetheless, significant obstacles include insufficient AI-specific training, disparities in technological access, and risks related to algorithmic bias. AI-integrated educational approaches remain heterogeneous and lack standardization, although several pilot programs show promise for broader adoption.AI-augmented learning opens new horizons for general practice education, strengthening active pedagogy and practitioner reflexivity. Success requires robust ethical governance, comprehensive digital training embedded in curricula, and ongoing interdisciplinary dialogue between educators, computer scientists, and clinicians.AI is a powerful driver for training the 21st-century general practitioner. Its successful integration must preserve humanistic aspects of care, promote equitable access, and support the development of critical and ethical competencies.

Objective Structured Clinical Examinations (OSCEs) are universally recognized as the gold standard for assessing clinical competence. However, they remain highly resource-intensive to organize, severely limiting the frequency and scalability of student training opportunities. While Generative Artificial Intelligence (AI) shows promise in medicine, randomized controlled trials evaluating its direct impact on high-stakes examination performance are lacking.This study evaluated the efficacy of an AI-Standardized Clinical Examination (ASCE) framework, powered by GPT-4 Turbo (DocSimulator™), on medical students' faculty OSCE performance. Secondary objectives assessed emotional readiness, exam-related stress, student satisfaction, and the perceived realism of the AI-generated patient encounters.We conducted a single-blind, two-site randomized controlled trial including 247 early-clerkship medical students from November 2023 to January 2024. Participants were stratified by site and randomized to ASCE training (n=125) or standard training (n=122). The intervention group had 24/7 access to 16 expert-authored scenarios, enabling text-based interaction with virtual patients and immediate, detailed automated feedback. A physical examination station (otoscopy) served as a control to ensure intervention specificity. The primary outcome was analyzed in the intention-to-treat population.The ASCE group achieved significantly higher OSCE scores compared to controls (median 11.4 vs. 10.7 out of 20; p=0.02; 95% CI 0 to 1.2). Notably, exam absenteeism was significantly lower in the intervention group (0.8% vs. 4.9%). Post-intervention data showed that ASCE improved emotional readiness (p<0.001) and reduced stress (p=0.02). In-session feedback indicated high acceptance: 80% of users felt they interacted with authentic patients, and 67% felt evaluated by teachers. Post-hoc analysis confirmed a positive dose-response relationship between platform engagement and exam scores (p=0.01).Although the score improvement was modest, suggesting subtle enhancements across multiple domains rather than a single skill, the marked improvement in emotional preparedness establishes ASCE as a powerful pedagogical adjunct. Unlike traditional methods constrained by standardized patient availability, ASCE offers scalable, standardized training. However, it cannot currently replace physical examination training.ASCE training significantly enhances clinical reasoning and student confidence while addressing critical logistical challenges. As a scalable solution for diverse scenarios, AI-driven simulation represents a transformative, resource-efficient standard for modern medical education curricula.

Ensuring that health professionals meet verified thresholds of clinical competence is fundamental to societal expectations for safe, high-quality care. Competency-based medical education therefore requires assessment methods that are frequent, scalable, and standardised across settings. However, Objective Structured Clinical Examinations (OSCEs)—the current reference standard—are resource-intensive and difficult to scale. AI-Standardized Clinical Examinations (ASCEs), using conversational virtual patients with automated scoring, have been proposed as a potential solution. Although a recent randomised trial demonstrated their effectiveness for training purposes, their validity as a scalable instrument for assessing clinical competence remains unknown.Can an AI-simulated, AI-scored clinical examination provide a valid and reliable assessment of clinical competence compared with a synchronised, human-graded national OSCE?We conducted a multicentre prospective cohort study across six French medical schools. Among 948 eligible students, 586 completed a synchronised 10-station ASCE using conversational virtual patients approximately six weeks before undertaking the national 10-station OSCE. To assess scoring reliability, 20 ASCE transcripts were independently rated by eight expert examiners, yielding 143 completed transcript evaluations.ASCE performance was significantly associated with national OSCE scores (Spearman ρ = 0·57; P < 0·001; range across centres 0·48–0·72). Between-centre variance in ASCE scoring was minimal in mixed linear models (ICC = 0·006; model R² = 31%). Expert raters demonstrated strong inter-rater agreement for clinical skills (Krippendorff α = 0·80) and moderate agreement for communication and professional attitude (α = 0·70). Using the majority expert rating as the reference standard, automated scoring with GPT-4.1 did not differ significantly from expert assessment. Internal consistency was acceptable for a 10-station ASCE (Cronbach α = 0·67) and was projected to exceed 0·80 with 20 stations.ASCE scores were strong predictors of national OSCE performance, with consistent accuracy and negligible center influence. Secondary analyses showed acceptable reliability and expert agreement, and automated scoring matched expert performance, supporting ASCEs as scalable complements to traditional clinical competence assessment.This work represents the first large-scale, multicentre validation study demonstrating that AI-based clinical examinations can provide standardised, reproducible, and scalable assessment of clinical competence, with potential implications for the development of digital infrastructures supporting large-scale clinical assessment, accreditation, and continuous professional evaluation.

Generative artificial intelligence (AI) systems are becoming increasingly widespread. These systems are being developed in the healthcare sector and have numerous applications in general practice. However, their use carries significant risks that must be managed. In order to promote quality and humanism in practice, the HAS has published its first guidelines for the proper use of generative AI in healthcare, intended for professionals, whether or not they are experts in the field.The method used is based on a critical analysis of scientific literature, interviews with stakeholders, and a review by targeted stakeholders within the ecosystem.Every use of a generative AI system must be conscious, supervised, and reasoned. 4 guidelines are recommended: Learn: professionals familiarise themselves with how the generative AI system works and how to use it. Verify: professionals pay attention to the relevance of its use, the quality of their queries and the control of the content generated. Assess: professionals analyse the quality and suitability of the generative AI system over time. Communicate: professionals engage with their ecosystem in a process of continuous improvement.These recommendations are advisory. Thus, their implementation must be ensured. In order for these recommendations to be understandable and implementable by inexperienced users, they do not cover all issues associated with the use of AI. This guide will therefore need to be supplemented by further work. To this end, the HAS is developing additional recommendations to help healthcare institutions implement AI systems.The HAS published the first guidelines for the use of generative AI in healthcare, with a rigorous scientific approach. This guide, the first of its kind in France, is designed to be pragmatic, grounded in reality and useful.

The COVID-19 health crisis has had a positive impact on the rise of digital health tools. The healthcare system is increasingly centered on predictive, data-driven governance models enabled by Artificial Intelligence (AI). Numerous conversational agents, commonly referred to as chatbots, now guide patients online or by telephone throughout their care pathway. However, few studies have examined these chatbots, which could, in the near future, interact with patients prior to a teleconsultation and provide general practitioners (GPs) with a preliminary pre-diagnosis.This study aimed to evaluate the acceptability of AI-driven pre-diagnosis in teleconsultations by examining GPs’ and patients’ intention to use a health chatbot.A quantitative study was conducted between July and December 2024 using two online self-administered questionnaires completed by GPs (recruited through professional networks) and patients (users of a teleconsultation platform). A research model was developed based on established measurement scales from the scientific literature, including the Technology Acceptance Model (Davis) and the Source Credibility framework (Ohanian).A total of 223 GPs and 249 patients participated in the study. Among physicians, most reported an intention to use the health chatbot. Attitude and social influence emerged as the two main factors positively influencing GPs’ intention to use AI, whereas perceived risks exerted a negative effect. Among patients, the results highlight the central role of the chatbot’s perceived humanity and perceived social presence. Social influence, self-efficacy, and perceived risks also significantly affected their intention to use the chatbot.A key strength of this study is its inclusion of the two main stakeholders in teleconsultation: patients and healthcare professionals. Both groups demonstrated an overall optimistic outlook regarding the future use of such chatbots, which aligns with findings from comparable studies outside the healthcare sector. GPs and patients alike value the human dimension of care, even in a teleconsultation context. Strengthening assurances around responsibility, integrity, security, and respect for the patient could help mitigate currently perceived risks.The acceptability of using AI for pre-diagnosis in teleconsultations is high among both patients and GPs. Nevertheless, several limitations were identified, which should be carefully addressed to ensure safe, effective, and responsible implementation of this technology.

Artificial intelligence (AI) is rapidly transforming healthcare, with applications ranging from imaging and diagnostics to administrative support. However, its use in general practice remains limited, despite the central role of general practitioners (GPs) in the healthcare system. Understanding their expectations, needs, and concerns is crucial to designing tools that are both ethically sound and operationally useful in primary care.To explore the perceptions, expectations, and reservations of general practitioners in the XXX region regarding the use of artificial intelligence in clinical practice, in order to identify the features of an ideal, acceptable, and ethically integrated AI tool.A qualitative study inspired by grounded theory was conducted between February and May 2024 among eleven community-based GPs (both established practitioners and locums). Semi-structured interviews were recorded, transcribed verbatim, anonymized, and analyzed through double coding and triangulation. Conceptual categories were developed to construct an explanatory model. The study followed COREQ criteria and complied with all ethical and data-protection standards (RGPD).Participants expressed interest in AI tools that are simple, intuitive, and seamlessly integrated into existing medical software. AI was perceived as a potential support for clinical decision-making and a way to reduce administrative workload. Nevertheless, significant concerns persisted: data reliability, loss of professional autonomy, alteration of the doctor–patient relationship, and the risk of dehumanizing care. GPs emphasized the need for transparency of algorithms, human oversight, and direct involvement of physicians in the design and validation of these tools. AI should assist, not replace, the clinician’s judgment.GPs view AI as a pragmatic opportunity to ease cognitive and administrative burdens, provided it respects the singularity of the clinical encounter. Their cautious optimism reflects both curiosity and ethical vigilance. Insufficient training and lack of exposure to AI technologies reinforce the need for participatory, co-design approaches involving frontline physicians.Artificial intelligence in general practice should be designed by and for physicians. By incorporating user needs—simplicity, reliability, human supervision, and ethical transparency—AI can become a genuine tool for enhancing care quality without compromising the human relationship at the core of medicine.