1.

AI technologies in healthcare should serve as supportive tools that aid, rather than replace, the clinical reasoning and decision-making of clinicians and other healthcare professionals.

Possible risks

Weaknesses

Potential solutions

Overreliance on AI may erode clinical judgment

AI may be misused for cost-saving, compromising care quality

Promote clinician-led decision-making and emphasize AI as a support tool

Reduced human interaction in care

AI may overlook rare or atypical conditions

Reinforce the clinician–patient relationship and uphold human-centered care

AI holds significant transformative potential for healthcare. By accelerating decision-making processes and streamlining data analysis, AI can help address many of the current limitations within healthcare systems, particularly in settings characterized by workforce shortages and constrained resources.

SIMI strongly supports the future integration of AI in healthcare. However, it is essential to ensure that healthcare remains fundamentally human-centered. Healthcare professionals have a vocation and responsibility to serve and protect human life. Therefore, the clinician–patient relationship must remain the foundation of medical practice. Empathy, holistic care, and attention to the person—rather than merely the disease—must continue to define healthcare systems. Upholding these values necessitates a clinician-led model of care.

Clinicians are uniquely equipped to address the complexity of the whole person, while AI can support them by rapidly analyzing specific data points [4, 8, 34,35,36]. This collaboration enables clinicians to allocate more time to direct patient care and contributes to a more efficient use of limited resources [8, 34, 37, 38].

SIMI also underscores the potential of AI in reinforcing existing medical knowledge. The COVID-19 pandemic highlighted the vital role of clinical intuition and judgment, particularly in the face of novel diseases. In this context, SIMI envisions AI as a tool that supports—but does not replace—clinical reasoning, within a clinician-centered framework. This approach fosters a collaborative—or even “symbiotic”—relationship, wherein artificial and human intelligence work together to enhance medical decision-making [8, 38].

AI’s ability to process large volumes of data can significantly reduce the time required for differential diagnosis [6, 9, 39, 40]. By identifying relevant patterns, AI can assist clinicians in making more accurate and personalized treatment decisions earlier in the care pathway [41,42,43]. In addition, AI may help bridge knowledge gaps within specific medical specialties [44,45,46]. Even in Internal Medicine, where broad expertise is crucial, rare or atypical conditions may be overlooked. AI can aid in detecting such cases, but the final clinical judgment must always remain with the physician.

Finally, SIMI cautions against the misuse of AI in healthcare. While AI can be useful for collecting preliminary information, direct patient interaction with AI should never replace the clinician. A healthcare system driven solely by cost-saving imperatives must not compromise the integrity of the clinician–patient relationship. Preserving this relationship is not only an ethical obligation but also essential to maintaining the quality of care.

As SIMI, we expect this recommendation will help preserve the central role of clinicians in patient care, ensuring that AI augments rather than replaces clinical expertise. This approach is expected to enhance diagnostic accuracy and efficiency, by the use AI as a supporting tool, while maintaining the humanistic values of empathy and holistic care that are fundamental to Internal Medicine. With this structure, healthcare systems can reduce the risk of overreliance on automation, ultimately improving patient safety and sustaining trust in medical decision-making.

2.

Clinicians should acquire the necessary knowledge and competencies to interact effectively with AI-enhanced healthcare systems. Therefore, there is the need for the integration of AI-related education and training across all levels of medical education and professional development.

Possible risks

Weaknesses

Potential solutions

Automation bias due to blind trust in AI outputs

Lack of clinician understanding of AI systems

Introduce AI literacy in medical curricula and continuing education

Misinterpretation of AI due to poor data quality

Limited attention to data quality in model-centric approaches

Promote data-centric AI education focusing on data integrity and bias awareness

It is essential to promote the integration of AI education and training at all levels of medical education and continuing professional development [47,48,49]. According to the World Health Organization, the full potential of AI to improve healthcare can only be realized if healthcare professionals are equipped to understand, apply, and critically evaluate these technologies within the context of everyday clinical practice [50].

Incorporating AI literacy into both undergraduate and postgraduate medical education is crucial to prepare the healthcare workforce for an evolving, digitally driven health system [51,52,53,54]. This requires a rethinking of academic curricula and continuing education programs to include dedicated modules on AI, emphasizing multidisciplinary approaches and scenario-based clinical learning. Consistent with a data-centric AI approach (Box 2), training should also equip clinicians with the skills to assess how data quality, bias, and completeness directly affect the reliability and safety of AI systems [55].

The training of healthcare professionals in the use of innovative AI technologies could also be supported by third-sector organizations already engaged in educational activities [1]. These actors should operate in accordance with national guidelines and contribute to the delivery of shared, multidisciplinary, standardized, and longitudinal continuing education programs.

Adequate training is essential to prevent—or at least mitigate—the risk of automation bias, defined as the uncritical or excessive reliance of clinicians on automated decisions and recommendations [56].

SIMI is strongly convinced that clinicians will critically evaluate and effectively utilize AI tools in daily practice only if AI education will be integrated in all levels of medical training. Therefore, we recommend a reduced automation bias and improve the safe adoption of AI technologies, leading to better patient outcomes aligning with real-world clinical needs.

3.

AI technologies should be developed with the explicit goal of reducing administrative and cognitive workload for physicians and healthcare staff, thereby enhancing the quality and efficiency of patient care.

Possible risks

Weaknesses

Potential solutions

Privacy violations from using general-purpose AI tools

Lack of certified platforms for clinical use

Use certified, GDPR-compliant healthcare-specific platforms

Overdependence on automation for documentation

Clinicians may lack training in AI tools

Provide continuous education and maintain clinician oversight

The considerable administrative and cognitive burden placed on healthcare professionals must not be underestimated. Current healthcare work structures require substantial time, psychological effort, and emotional energy for tasks that are not directly related to patient care [57, 58]. Over time, there has been a marked increase in bureaucratic duties, which has reduced time available for direct patient interaction, communication with families, clinical reasoning, and the optimization of therapeutic strategies—factors contributing to both clinician and patient-related burnout [59].

The widespread adoption of Electronic Health Records (EHR), Picture Archiving and Communication Systems (PACS), and laboratory information systems—often integrated—has already simplified certain aspects of clinical work, particularly information retrieval and documentation, and contributed to the standardization of clinical processes [60].

Artificial Intelligence has the potential to further enhance these systems by streamlining cognitively demanding workflows and automating repetitive tasks [61]. These include the retrieval of relevant anamnestic data (especially from connected EHRs), correction of redundancies or errors in clinical documentation, assisted completion of clinical-administrative forms (such as hospital discharge summaries [SDOs], electronic prescriptions, or multidimensional assessment tools), and the application of diagnostic and prognostic scoring algorithms based on patient data [62, 63]. Currently, many clinicians independently use general-purpose generative AI tools for such purposes [14], and several studies indicate that generative AI can improve performance on patient-care tasks [64]. These virtual assistants can help uncover clinical scenarios that might otherwise be overlooked, while final judgment remains with the physician.

In addition, several domain-specific software tools are under development or evaluation, based on advanced Natural Language Processing (NLP) technologies. Examples include AutoScribe, Suki AI, and Ambience Healthcare, which are designed specifically to support clinical practice [65].

The primary challenge in this domain is the protection of personal health data, particularly with respect to compliance with the General Data Protection Regulation (GDPR). The use of personal health information in uncontrolled environments—such as third-party chatbot interfaces—raises the risk of privacy violations, including unauthorized access, accidental data disclosure, user profiling, biased AI-generated outputs with discriminatory potential, and re-identification of pseudonymized data [66].

For these reasons, in addition to continuous training and awareness initiatives—an area where SIMI should take a leading role—the implementation of AI technologies must occur within certified, healthcare-specific software platforms. This should be accompanied by the adoption of codes of conduct and certification frameworks by healthcare institutions to ensure legal and ethical compliance.

Adopting AI solutions to streamline administrative and cognitive tasks is expected to free up clinicians’ time, allowing greater focus on direct patient care. This recommendation will likely improve workflow efficiency and reduce burnout among healthcare staff, contributing to higher job satisfaction and better patient experiences, by automating repetitive processes. Therefore, we expect AI can help optimize resource allocation and support the delivery of high-quality, timely care.

4.

AI and other technological innovations should actively contribute to reducing—rather than perpetuating—existing disparities in health and access to care. To support this objective:

●

AI systems should be trained on data from diverse and representative settings and populations;

●

Public institutions, including the Italian government and relevant health authorities, should invest in research aimed at identifying and mitigating any discriminatory outcomes associated with AI technologies;

●

Multidisciplinary collaborations involving governmental bodies, academic institutions, nonprofit organizations, and private industry should be established to promote the development of fair and unbiased AI algorithms, both now and in the future.

Possible risks

Weaknesses

Potential solutions

Algorithmic bias from non-representative datasets

Underrepresentation of minority populations in training data

Train AI on inclusive datasets and promote universal design principles

Emergence of a two-tiered healthcare system

High costs and lack of digital literacy in underserved areas

Public investment, accessible interfaces, and support personnel for vulnerable groups

Deep-rooted social inequalities persist within today’s healthcare systems, affecting both access to care and participation in basic and clinical research. In this context, AI systems present a dual potential: they may help mitigate these disparities or, conversely, exacerbate them.

Certain groups remain particularly vulnerable in terms of equitable access to healthcare. These include individuals from specific ethnic backgrounds impacted by large-scale migration, persons with disabilities, and those affected by rare diseases [67, 68]. To be effective in underserved populations, AI tools—whether general-purpose chatbots or CDSS—must be designed with these minority groups explicitly in mind. However, AI systems are often trained on standard datasets influenced by the priorities of commissioning entities, which frequently exclude or underrepresent these populations. This omission can overlook crucial genetic, epigenetic, and socio-environmental factors that affect disease development and treatment response.

As a result, algorithmic bias may arise, manifesting in generalization errors, misclassifications, inaccurate probability estimates, and flawed interpretations of clinical variables—ultimately leading to erroneous prognostic or therapeutic suggestions [69]. To address this, even correction strategies such as cross-validation, post hoc interpretability assessments, and internal checkpoints must be designed with underrepresented populations in mind. AI software developed using dedicated datasets that reflect the characteristics of minority populations may offer a path toward reducing disparities in both access and quality of care. Moreover, such efforts could contribute to generating new scientific knowledge that transcends current research limitations [70].

A related concern is the potential emergence of a two-tiered healthcare system. The costs associated with acquiring, updating, and maintaining AI systems—as well as training personnel—may be unsustainable for many healthcare institutions, particularly those in the public sector. As a result, AI systems calibrated for minority populations could become accessible only to wealthier regions or private entities, limiting availability for those most in need [71, 72]. Furthermore, even when available, these tools may be ineffective if patients lack the necessary education to understand their capabilities and risks. Cultural and material barriers to digital healthcare environments remain significant for many disadvantaged populations. Addressing this challenge requires dedicated personnel to support individuals—particularly in low-income communities—who face technological access issues [50].

In this regard, SIMI is committed to actively contributing during the design and validation phases of medical AI software, offering guidance to address the underrepresentation of vulnerable populations. It also aims to serve as an intermediary between ethics committees and developers. SIMI will promote the development of applications aligned with the principles of universal design, digital accessibility, and comprehensibility for users with cognitive disabilities. Moreover, SIMI will play a key role in clinician education—not only regarding the general principles of responsible AI use, but also in supporting access for minority populations. This includes helping patients with limited or no digital literacy benefit from AI-driven innovations.

This recommendation will help ensure that vulnerable and underserved populations benefit from AI-driven innovations, reducing disparities in access and outcomes. This is particularly important in limited sources scenarios, to support the best choose of diagnostic–therapeutic flowchart, also where medical equipment are scarce. Multidisciplinary collaboration and public investment in fair AI development will foster inclusivity and support the creation of universally accessible healthcare technologies.

5.

To foster public trust and uphold the integrity of the clinician–patient relationship, transparency throughout all stages of AI development and clinical application is needed. Whenever feasible, patients and healthcare providers should be informed when AI systems are involved in diagnostic or therapeutic processes.

Possible risks

Weaknesses

Potential solutions

Lack of disclosure may erode patient trust

Opaque AI decision-making processes

Inform patients and clinicians when AI is used; adopt explainable AI tools

Misalignment with clinical needs

Limited clinician involvement in development

Involve clinicians in AI design, development, and validation

AI must be thoughtfully integrated into clinical practice, and SIMI emphasizes the importance of explicitly disclosing its use in all procedures and clinical workflows where it is involved. At the core of healthcare lies the clinician–patient relationship, which is grounded in mutual trust. Transparency in clinical decision-making and treatment selection, including drug prescriptions, is essential to maintaining and reinforcing this trust. Conversely, a lack of transparency risks undermining it.

Human capabilities are inherently limited, and collaborative teamwork is often necessary to address individual constraints. In this context, AI can serve as a valuable ally, supporting healthcare teams in mitigating knowledge gaps. However, its use must be clearly acknowledged, akin to any other diagnostic or therapeutic tool.

Such transparency not only helps patients better understand the complexity and rigor of the diagnostic process but also allows clinicians to demonstrate the care and diligence underlying their clinical decisions—decisions increasingly supported by advanced technologies.

Furthermore, the development of AI-based tools must follow transparent and ethical standards. SIMI advocates for the adoption of explainable AI tools that produce clear, interpretable outputs which clinicians can understand and trust.

To ensure clinical relevance and safety, SIMI also recommends the active involvement of clinicians in development teams. Their expertise is crucial to aligning technological innovation with real-world clinical needs. In this regard, SIMI encourages developers to collaborate with established and recognized medical scientific societies or professional organizations throughout the design, development, and validation of AI technologies in healthcare.

By fostering transparency, collaboration, and ethical development, AI can serve as a powerful support system—enhancing, rather than replacing, the human elements that define high-quality care. This model of co-development is central to the vision of Symbiotic AI, in which human and AI are integrated within a mutually reinforcing partnership.

SIMI expects and encourages the transparency in the development and clinical application of AI. This process will strengthen public trust and uphold the integrity of the clinician–patient relationship. Informing patients and providers about the involvement of AI in care processes is expected to enhance understanding, acceptance, and shared decision-making. At the same time, the adoption of explainable AI tools will facilitate clinician oversight and accountability, supporting safer and more ethical integration of this technology in medical practice.

6.

The critical importance of safeguarding the privacy and confidentiality of both patient and clinician data during the development, training, and deployment of AI models in clinical practice should be emphasized.

Possible risks

Weaknesses

Potential solutions

Re-identification from anonymized data

Inadequate data governance frameworks

Implement strict access controls and traceability mechanisms

Misuse of patient data

Lack of infrastructure for secure hosting

Host data within national healthcare systems under privacy laws

The integration of AI into clinical practice requires the use of patient-care data for the development, training, and validation of AI models. While AI’s advanced capabilities in data processing and pattern recognition offer significant potential, they also heighten the risk of re-identification—even when datasets have been anonymized. Nonetheless, such data are essential for developing tools that effectively support clinical decision-making and improve patient outcomes. It is imperative that the use of AI does not undermine the trust that forms the foundation of the clinician–patient relationship.

Clinical data—whether relating to patients or the healthcare professionals involved in their care—are inherently reflective of the clinician–patient dynamic. The inappropriate use or absence of such data may compromise the continuity and quality of care. Given AI’s advanced analytic capabilities, the risk of re-identification from anonymized datasets remains a critical concern.

Accordingly, SIMI considers the protection of privacy a non-negotiable prerequisite for the ethical development and deployment of AI-based tools in healthcare. Both personal and clinical data must remain under the ownership and control of the patient and should be accessed and used by clinicians exclusively to enhance care, whether in routine clinical practice or ethically approved research [73, 74].

To support this principle, SIMI advocates for the establishment of robust data governance frameworks. These should include strict access controls, traceability of data usage, and full compliance with national and international data protection regulations. Where feasible, SIMI recommends that data repositories be hosted within the national healthcare system’s infrastructure, in accordance with applicable privacy laws and up-to-date, authoritative technical standards for clinical data use [73, 74].

The integration of AI into healthcare must be guided by the principles of transparency, accountability, and respect for patient autonomy. Protecting privacy is not only a legal obligation but also a foundational requirement for preserving the integrity of the physician–patient relationship in the digital era.

This recommendation should reduce the risk of data breaches and unauthorized use, ensuring compliance with legal and ethical standards. Emphasizing robust data privacy and confidentiality measures will protect patient and physician trust, which is essential for effective healthcare. This goal must be pursued by implementing rigorous governance frameworks that support the responsible use of clinical data, fostering innovation while protecting individual rights.

7.

Development, validation, and application of AI in healthcare must align with the core principles of medical ethics. These technologies should aim to enhance patient care, support clinical decision-making, strengthen the clinician–patient relationship, and promote fairness and equity within the healthcare system.

Possible risks

Weaknesses

Potential solutions

AI may prioritize efficiency over equity

Risk of replacing human judgment

Design AI to enhance fairness and support clinical autonomy

Reinforcement of structural inequities

Biased or narrowly scoped training data

Use inclusive datasets and ethical design principles

AI has the potential to enhance healthcare delivery by enabling faster and more accurate disease detection, greater personalization of treatments, and more efficient allocation of resources. However, these improvements must be guided not only by considerations of efficiency, but also by a clear ethical imperative: ensuring that every patient receives appropriate, timely, and evidence-based care. As emphasized by the World Health Organization (WHO, 2021), AI technologies should contribute to a more responsive model of medicine—one focused on the holistic well-being of patients and designed to reinforce, rather than replace, human intervention [50]. This objective should be pursued in accordance with the bioethical principles outlined by Tom L. Beauchamp and James F. Childress (1979), which are already embedded in clinical practice, and further strengthened by the principles of explicability and responsibility [75].

The enhancement of clinical decision-making is a critical area in which AI can complement clinicians’ professional expertise [42, 44, 45]. Predictive algorithms, diagnostic imaging analysis, and decision support systems should be designed to assist—rather than replace—clinical judgment, enriching the decision-making process with additional data while safeguarding the autonomy and accountability of healthcare professionals. In the context of Internal Medicine, AI should aim not only to advance technological capabilities but also to promote health equity and genuinely person-centered care.

AI technologies also offer the potential to reduce administrative and analytical burdens. However, as highlighted in the WHO’s ethical guidelines, technologies trained on biased or narrowly scoped data may reinforce existing structural inequities. AI must serve as a tool for promoting fairness and justice within healthcare systems. To this end, algorithmic models must be developed using inclusive and representative datasets, to mitigate the risk of perpetuating or exacerbating disparities based on gender, ethnicity, socioeconomic status, or geographic location. Moreover, well-designed AI systems can help bridge care gaps by improving access in rural and underserved areas.

According to this recommendation, SIMI expects to support clinical autonomy and accountability, preventing the replacement of human judgment by automated systems. The use of inclusive datasets and ethical design principles will help mitigate structural inequities and foster a more just healthcare system. Aligning AI development and application with core medical ethics will ensure that technological advances genuinely enhance patient care and promote fairness.

8.

Clinical safety, effectiveness, and equitable use of AI technologies should be prioritized by all stakeholders involved—including developers, researchers, implementers, and regulators. The integration of AI into healthcare should follow a process of “continuous improvement”, incorporating real-world testing, feedback from end users, and rigorous scientific validation across diverse clinical settings. Particular consideration must be given to the identification and management of both current and emerging risks associated with AI use in medicine.

Possible risks

Weaknesses

Potential solutions

Static AI systems may become outdated

Lack of real-world validation and user feedback

Adopt continuous improvement and multicenter clinical trials

Regulatory gaps for evolving AI systems

Inflexible approval models

Update frameworks to accommodate dynamic AI systems and human oversight

Any AI system intended for medical use should not be co