Abstract / Summary
Artificial intelligence (AI) in surgery and anesthesiology encompasses different tasks: risk prediction, hypotension alerts, video analysis, technical-skill assessment, surgical-duration estimation, and postoperative support. Treating them as one promise of “AI surgery” obscures essential distinctions among retrospective performance, external validation, prospective clinical utility, impact on processes or outcomes, and routine implementation. This whitepaper synthesizes selected applications rather than exhaustively cataloguing all subspecialties, robotic platforms, or simulation techniques. The evidence closest to clinical evaluation in anesthesiology is concentrated in intraoperative hypotension prediction. A systematic review and meta-analysis included eight randomized-trial reports; its published total of 568 participants includes overlap between a study and its substudy and does not represent unique individuals. Seven reports evaluated the Hypotension Prediction Index and one evaluated fuzzy-logic control. The hypotension meta-analysis including the largest trial did not find a statistically significant difference and showed high heterogeneity. Favorable results from small trials do not establish universal benefit: this is a narrow family of systems, populations, and protocols. In surgery, phase recognition and technical-skill assessment have systematic reviews, whereas models for complications, delirium, length of stay, and surgical duration remain dominated by retrospective development or limited validation. Latin America has direct evidence that should not be replaced by silent extrapolation. A Bogotá study evaluated models for surgical-duration estimation; Brazilian studies developed models for postoperative delirium and length of stay after coronary artery bypass grafting; and LASOS described perioperative outcomes across 17 countries. These sources demonstrate regional research and data capacity and a measurable perioperative burden; they do not demonstrate general clinical effectiveness of AI, routine implementation, or transportability across countries. The directed search identified few regional AI records compared with the international literature; statements about infrastructure, interoperability, and governance therefore also rely on regional sources and are labelled as implementation context or inference, not model performance. Central finding: AI can currently support bounded perioperative surveillance, prediction, analysis, training, and operational tasks. It cannot yet responsibly promise independent intraoperative decisions, replacement of surgical or anesthetic