Distinguishing artificial intelligence from robotic execution in total knee arthroplasty: a narrative review and evidence map

Purpose Artificial intelligence (AI) and machine-learning methods are increasingly discussed in relation to robotic-assisted total knee arthroplasty (TKA). However, robotic assistance, computer navigation, and AI-based prediction represent distinct functions that are often conflated under the single label “AI-assisted robotic TKA,” obscuring the state of evidence for each. This review clarifies these distinctions and maps the current evidence for AI-related functions across the TKA pathway, keeping AI prediction evidence analytically separate from robotic execution evidence at the level of each clinical function. Methods We performed a critical narrative review and evidence map of peer-reviewed literature relevant to robotic-assisted TKA, navigation-assisted TKA, AI/ML prediction models, digital planning, implant sizing, intraoperative decision support, postoperative monitoring, implementation, and economic evaluation. Evidence was interpreted according to clinical function, comparator structure, validation status, and translational maturity. Because the available literature is methodologically and conceptually heterogeneous, quantitative pooling and formal comparative-effectiveness claims were not attempted. Results Robotic-assisted and navigation-assisted TKA have the most mature evidence base for improving technical precision, particularly implant positioning and radiographic alignment. In contrast, AI/ML evidence is concentrated mainly in adjacent pathway functions, including radiographic interpretation, implant-sizing prediction, patient-selection support, dissatisfaction prediction, outcome modelling, perioperative risk stratification, and postoperative monitoring. Direct clinical evidence demonstrating that AI-enabled robotic TKA provides incremental patient-level benefit beyond conventional robotic assistance remains limited. Current studies are constrained by inconsistent terminology, short follow-up, limited external validation, incomplete reporting of AI components, heterogeneous outcomes, cost uncertainty, and uncertain generalizability across patient populations and healthcare settings. Conclusion AI should not currently be treated as a single validated surgical intervention in TKA. Rather, it represents a family of emerging pathway-level functions that may support preoperative planning, intraoperative decision support, patient selection, risk communication, and longitudinal recovery monitoring. Future studies should explicitly define the AI component being evaluated, distinguish AI functionality from robotic or navigation assistance, use standardized comparators, and assess long-term patient-reported outcomes, implant survivorship, safety, cost-effectiveness, and equity across multicenter prospective cohorts or randomized trials.