Artificial intelligence (AI) has progressed from a theoretical concept to a fundamental component of the entire perioperative patient care continuum^[1]. This advancement has been primarily driven by two factors: improvements in algorithmic capabilities and the clinical integration of surgical robotic systems.

Machine learning (ML)-based algorithms serve as the foundation for achieving high diagnostic accuracy and enabling tailored treatment strategies. In thoracic surgery, many new models are being developed to support lung cancer diagnosis, predict lymph node metastasis and assess the risk of postoperative complications^[2]. For example, by leveraging clinical data, a machine-learning model can predict the risk of prolonged air leak after a lung resection with an Area Under the Curve (AUC) of 0.84. Such accurate prediction could substantially aid surgeons in proactively managing PAL in patients undergoing video-assisted thoracoscopic surgery (VATS)^[3]. As validation study results continue to accumulate, leading medical centers are progressively incorporating these models into their clinical platforms.

Operationally, as the essential hardware platform for implementing AI, surgical robots are experiencing steadily increasing rates of installation and utilization. In the United States, the proportion of lobectomies performed via Robot-Assisted Thoracoscopic Surgery (RATS) increased from 19.2% to 34% between 2015 and 2018^[4]. This upward trend continues globally, with robotic surgery gradually transitioning from an alternative approach to a mainstream modality in thoracic surgery^[5].

The Da Vinci Surgical System, based on a “master-slave” design^[6], enhances surgical stability and dexterity. Nevertheless, the quality and efficiency of surgery remain dependent on the surgeon’s expertise and endurance. The incorporation of increasingly advanced AI as a core processor is expected to endow these systems with greater autonomy, enabling a phased transition from teleoperation (L0) and task assistance (L1-L3) to conditional autonomy^[7]. The integration of advanced algorithms with hardware platforms has laid a solid foundation for the development of an intelligent surgical model characterized by tripartite collaboration. This model, designed to enhance the surgeon’s strategic decision-making while leveraging AI and robotic systems as capability extensions, constitutes a dataflow-driven closed-loop system that is fundamentally transforming the paradigm of clinical practice in thoracic surgery.

The following sections discuss the specific applications of this model in preoperative planning, intraoperative guidance, and postoperative management, highlighting both current applications and future prospects of this tripartite collaboration [Figure 1].

Stage III lung cancer, characterized by its high heterogeneity and substantial therapeutic challenges, remains the most formidable issue in contemporary lung cancer management and represents a particularly promising application scenario for the intelligent surgical model. Beyond optimizing the surgical procedure itself, this model has the potential to support the development of personalized therapeutic strategies for patients with locally advanced disease. With a properly sequenced regimen of neoadjuvant therapy and surgery, patients may achieve long-term survival.

For patients who have undergone neoadjuvant therapy, VATS presents increased technical difficulty. In contrast, RATS systems have demonstrated superior clinical outcomes, including reduced operative time, a lower conversion rate to open thoracotomy, and enhanced efficiency in lymph node dissection^[16]. Moreover, AI shows considerable potential in assisting surgeons with complex or controversial surgical decision-making. Previous studies have demonstrated that a CT-based deep learning approach (the NeoPred model) can predict response to immunotherapy in patients with non-small cell lung cancer (NSCLC)^[17]. However, a critical gap remains: no study has yet integrated clinical characteristics, CT imaging, and pathological features to develop a comprehensive multimodal model. By combining AI predictions with clinical expertise, clinicians may more accurately identify patients with complete treatment response. For these individuals, surgeons may adopt personalized strategies, such as deviating from the standard of care based on pre-therapy tumor staging by avoiding radical pneumonectomy. Furthermore, the definition of “borderline resectable” remains subjective, often reliant on the surgeon’s personal judgment^[18]. Through analysis of extensive clinical datasets, AI holds the potential to uncover complex patterns associated with R0 resection and long-term survival, thereby facilitating the development of personalized resectability scoring systems. Going a step further, AI could support feasibility assessments for conversion surgery in patients initially deemed unresectable following induction therapy.

Although the intelligent surgical model holds substantial clinical promise, key barriers remain at both the research and implementation stages. In particular, most studies of machine-learning predictive models are single-center and retrospective, and therefore provide only limited evidence. The lack of high-level, multicenter, prospective clinical validation hinders their adoption in clinical practice and prevents their incorporation into clinical guideline development. Furthermore, predicted outcomes often focus on perioperative short-term metrics, lacking assessment data on patients’ long-term survival and quality of life.

The critical obstacles in the implementation phase are primarily manifested in technological, ethical, and humanistic aspects.

For instance, inequitable access to robotic and AI technologies is driven largely by their high cost. A fundamental question of social justice is: who stands to gain from these technological advances? Their concentration in economically developed regions may exacerbate global health disparities and result in tangible harm^[19]. Mitigating this requires cost reduction through international collaboration. This gives rise to a second issue: data bias. Information from developed regions may lack generalizability, thereby compromising algorithmic fairness^[20]. Future initiatives should focus on establishing high-quality global surgical databases. Furthermore, reliance on “black-box” algorithms creates major concerns regarding transparency^[21]. Developing explainable AI (XAI) and establishing a robust medical AI validation and regulatory framework may help increase physicians’ confidence in these new technologies.

As robotic autonomy advances, the ethical issue of accountability becomes more prominent. Similar to the issue of determining liability in an accident involving an autonomous vehicle^[22], it is imperative to develop a robust legal and ethical framework to delineate rights and obligations in human-robot surgical collaboration and to ensure informed patient consent. The development of this framework should be guided by comprehensive societal dialogue^[23].

As AI cannot address emotional problems, empathetic surgeons must retain the central role within the intelligent surgical model. Complete delegation of patient care to AI could diminish therapeutic adherence and negatively impact treatment outcomes. Addressing this challenge necessitates a redefinition of the surgeon’s role and a restructuring of medical education. Cancer treatment involves divergent patient priorities, ranging from accepting higher risks for potential survival gains to prioritizing quality of life. Fully autonomous, AI-driven surgery is fundamentally limited in addressing this spectrum of values. Accordingly, the empathy and emotional support provided by human physicians therefore remain an indispensable component of patient-centered care.

For clarity, the AI-related topics discussed in this perspective are summarized in Table 1. In the foreseeable future, AI is more likely to augment rather than replace surgeons. A surgeon-led intelligent surgical model is emerging, in which AI and robotic systems empower and extend the capabilities of surgeons. High-quality real-world data will, in turn, support the development of more advanced algorithms and increasingly autonomous robotic systems. By addressing the challenges outlined above, this approach may foster a more natural, transparent, and trustworthy model of collaboration.