From promise to practice: A roadmap for artificial intelligence in critical care

Artificial intelligence (AI) has regained strong momentum in medicine, driven by unprecedented computing power and the availability of massive clinical datasets. Intensive care units (ICUs) are at the forefront of this movement, given their unique combination of high data density, decision-making under uncertainty, and the vulnerability of critically ill patients. Yet despite the abundance of proof-of-concept studies, the clinical translation of AI tools remains strikingly limited, with fewer than 2 % of published algorithms prospectively evaluated in real-world ICU settings.

In this editorial, we discuss the roadmap proposed by Workum et al., which outlines a progressive, risk-aligned framework for the integration of AI in critical care. Beyond model performance, the authors emphasize fundamental values such as fairness, explainability, and accountability, while highlighting the practical challenges of data interoperability, infrastructure, governance, and liability. Their work reminds us that AI adoption is not primarily hindered by algorithms themselves but by the surrounding ecosystem of data quality, regulatory clarity, and clinician trust.

We argue that the true promise of AI in the ICU lies not in rapid technological breakthroughs but in careful, evidence-based, and human-centered implementation. Whether these systems will ultimately improve patient-centered outcomes remains uncertain. The roadmap by Workum et al. should therefore be read as a call for cautious progress: to begin with low-risk applications, to invest in infrastructure and interdisciplinary collaboration, and to rigorously evaluate clinical benefit before moving toward high-stakes medical decision support.