Disentangling objects' contextual associations from perceptual and conceptual attributes using time-resolved neural decoding.

The rapid adoption of artificial intelligence (AI) in health care is changing clinical workflows across various medical specialties. From diagnostic imaging analysis to ambient documentation scribes, AI-powered tools are increasingly integrated into routine clinical practice. According to the American Medical Association, physician AI use nearly doubled from 38% in 2023 to 66% in 2024, reflecting rapid integration compared to historical health care technology adoptions []. These tools are intended to alleviate the administrative burden that has been identified as a primary driver of clinician burnout. This clinician crisis peaked at 62.8% prevalence in 2021, though rates have since declined to approximately 45% as of 2023 []. Algorithmic solutions are considered practical largely because they can automate cognitively demanding tasks, particularly clinical documentation. A landmark time-motion study demonstrated that physicians spend 49.2% of their office day on electronic health record and desk work combined, with only 27% on direct clinical face time, and an additional 1‐2 hours of electronic health record work each evening []. Early evidence from ambient AI scribes suggests meaningful reductions in documentation time, and these tools are frequently promoted to reduce administrative burden []. However, it remains unclear whether these systems reduce overall cognitive burden or shift it from content generation to verification. Recent commercial deployments in 2024 and 2025 have accelerated this enthusiasm. Tierney et al [] reported substantial reductions in documentation burden following an enterprise-wide rollout of an ambient AI scribe to over 3000 clinicians, and Albrecht et al [] described similar quality-improvement gains in a multispecialty implementation. These 2 enterprise deployments illustrate the pace of commercial scaling but do not contribute outcome data to this review because neither used a validated cognitive-workload instrument, such as the National Aeronautics and Space Administration Task Load Index (NASA-TLX), or the Physician Task Load Index, or a validated burnout instrument as a prespecified primary outcome. Consequently, neither met the eligibility criteria (detailed in the Methods section). Throughout this text, commercial deployment data, theoretical frameworks, and prior reviews provide background only; the primary, secondary, and exploratory outcomes reported in the Results section derive solely from the 21 studies that met the eligibility criteria.