Green artificial intelligence: Pioneering sustainable innovation in health technologies

Dear Editor,

We have recently read with great interest the article entitled ‘A Bibliometric Analysis of Perioperative Medicine and Artificial Intelligence’ by Chan et al (2025) that was published in your journal. The paper offers valuable insights into the rapid expansion of artificial intelligence (AI) applications in perioperative medicine, highlighting major research trends, leading contributors and emerging areas of focus.

One of the most striking findings is the uneven geographical distribution of AI research in perioperative medicine. The data reveal that the United States, China and Italy are the leading contributors in terms of publication amount. However, contributions from low- and middle-income countries (LMICs) remain disproportionately low, underscoring a significant research gap that may compromise the global applicability and inclusivity of advances in the field. Regrettably, this trend is unlikely to shift in the near future unless proactive and inclusive measures are adopted (Salluh et al 2025). This imbalance is not just an academic issue – it has profound real-world consequences. AI-driven perioperative innovations, such as risk stratification models, automated monitoring and decision-support tools, are often developed using data from high-income health care settings. As a result, these models may not be applicable or generalisable to regions with different patient demographics, resource constraints or health care infrastructures (Yu & Zhai 2024).

The disparity in the adoption of AI in medicine between high-income countries (HICs) and LMICs risks widening further due to pressing sustainability challenges. If digitalisation in health was estimated to save up to 11 billion USD in health care costs by 2030, the development and deployment of AI-driven health care solutions require advanced digital infrastructure, substantial computational resources, and highly specialised personnel, all of which remain scarce in resource-limited health care settings (Monlezun et al 2025). A particularly critical issue is the significant energy consumption required to sustain data centres, which serve as the foundation of AI models. According to estimates by the International Energy Agency (IEA), data centres accounted for approximately 1% of global electricity consumption in 2022, excluding energy used by data networks and cryptocurrency mining (IEA n.d.). In major economies such as China, the European Union, the United Kingdom and the United States, their electricity demand was then notably higher, ranging between 2% and 4%. In addition to energy consumption, other indirect costs have been identified, such as the generation of electronic waste (Ueda et al 2024). All of this could lead to an increasing disparity between HICs and LMICs, contradicting the fundamental ethical principle of AI in health care: to reduce global inequalities. Green Artificial Intelligence is founded on the optimisation of computational efficiency, aiming to reduce energy consumption and infrastructure costs associated with training and deploying AI models (Bolón-Canedo et al 2024). As such, the adoption of a Green Artificial Intelligence approach could serve as a key strategy to facilitate the development and implementation of AI in resource-limited settings, ensuring that technological advancements are both accessible and sustainable. Recently, a set of best practice recommendations for sustainable AI in health care has been published (Ueda et al 2024). Distributed machine learning (ML) techniques, such as federated learning, energy-aware training, and model pruning, represent some of the possible techniques to significantly reduce energy consumption in ML, making AI more sustainable and accessible, even in resource-constrained environments.

In this context, international collaborations encouraged by Chan et al (2025) take on even greater value. Collaborations could mitigate these disparities through various approaches such as sharing datasets, accessing cloud-based infrastructure and developing less energy-consuming algorithms, making AI more globally accessible. Furthermore, capacity-building programmes and open-access AI initiatives could facilitate more equitable adoption, ensuring that the benefits of AI in perioperative medicine do not remain exclusive to HICs (Victor 2025). A multidisciplinary approach – involving computer engineers, data scientists, environmental engineers and other experts – is essential to identifying the most effective and sustainable solutions for each specific context. International collaborations can also strengthen LMICs in negotiations with major high-tech companies, ensuring a more balanced and ethical technology adoption (Yu & Zhai 2024).

In conclusion, we believe that the paper by Chan et al (2025) provides an interesting review of the literature on AI in perioperative medicine. One of the most compelling aspects of their analysis is the disparity between HICs and LMICs, highlighting the heterogeneous distribution of AI research and implementation. These findings prompt critical reflection on the broader implications of AI in perioperative care. From risk assessment and individualised anaesthetic planning to real-time monitoring and early detection of postoperative complications, the potential benefits of AI are significant. However, without an organised and structured incorporation into clinical routine, these improvements will remain largely confined to more technologically advanced health care systems.

The findings should inspire concrete proactives in the perioperative field – including investment in collaborative initiatives, structured education programmes and the development of adaptable, transparent AI tools that are compatible with different clinical environments. If these steps are not taken, there is a real risk that AI will deepen the divide in surgical outcomes, rather than bridging the gap and supporting the development of safe, efficient and equitable perioperative care on a global scale.