Traditional patient safety protocols are based on the historic principle of the Hippocratic oath intended to abstain from inflicting harm on our patients (”Primum non nocere”) [1]. In spite of this noble vision, patients frequently remain caught in the ‘friendly fire’ of preventable surgical complications and hospital-acquired adverse conditions, which are widely regarded as an unfortunate side-effect of modern healthcare [2]. Ironically, the plethora of stringent regulatory compliance-mandated protocols and globally disseminated patient safety checklists still fail to protect our patients from medical errors in the modern age of patient safety today [3]. Our continued quest towards “Goal Zero“ for preventable harm requires a pragmatic reassessment of the historically failed approaches in the patient safety arena [4, 5]. It is time for a paradigm shift from the traditional approach of counting and analyzing medical errors, with the intent of preventing similar occurrences to harm a different patient in the future, to effectively predicting and preventing an adverse event before harm reaches the patient [6]. The new age of artificial intelligence allows for novel machine learning approaches to support predictive analytics tools which can identify a patient at risk of sustaining harm before the adverse event occurs [7]. The latest special collection in Patient Safety in Surgery is dedicated to covering a variety of aspects pertinent to machine learning approaches for the improvement of surgical patient safety (www.biomedcentral.com/collections/MLPS).