Introduction With advancements in minimally invasive surgery (MIS), surgical video has evolved from historical record to integral component of the operation, closely intertwined with the surgeon's actions. Any integration of Artificial Intelligence (AI) further transforms surgical video into a dynamic element that provides real-time intraoperative influence on the surgeon. 1 Such innovation is intended to enhance the capabilities of surgeons via augmented and potentially automated operative components. 2 However, AI-adapted surgical video can introduce new risks especially as, during operations, decision-action times are short with often irreversible consequence. This article examines the transcending role of surgical imaging, through a cognitive neuroscience lens, to consider differently the clinical and medico-legal implications of AI-augmented perception in surgery versus the traditional view of surgical video and indeed AI as just tools.with reality through perception which includes neural inputs from their a priori knowledge and experience. Surgical insight in MIS only fully emerges when operative video as sensory input is transformed by the surgeon's perceptual processing into scene pattern recognition and meaning (the surgical camera's digitalisation of the internal operative scene into a pixel stream represents inert structured signals until interpreted). Like all perception, surgical perception is a predictive process in which the surgeon's cognition repeatedly generates hypotheses to reconcile against incoming sensory signals. In operations, the surgeon's own well-trained mind, including anatomical expectations, judgement and both positive and negative previous experiences and heuristics, is essential to process the video display imagery with the required speed and capacity to flow sequential decisions under uncertainty. Under the "extended mind" theory (Figure 1), surgical equipment, including video feeds, are therefore viewed as extensions of the surgeon's mind. 3 Operative interpretations so are not created in the surgeons' eye or mind but in combination with the surgical video imagery as substrate for the cognitive scaffold that the surgeons uses to reason. This is equivalent to the concept "writing is thinking" 4 or the example of solving a mathematical equation on a chalkboard-the writing tool and canvas are intrinsic to the cognitive process.While surgical expertise enhances surgeons' perception and cognitive capacity for surgical decision making, surgeons (like all humans) are susceptible to cognitive illusions which can cause misperception. 5 These can stem from interpretation biases, like attentional and confirmation blinding, ambiguity resolution and expert "overconfidence," which can lead a surgeon to see what they expect to see rather than what is actually present. Such well-recognized limitations on human perception in other high-risk industries, like aviation, have led to the development of systems (such as slowing down, checklists or double verification for pilots) that counterbalance flawed heuristics to prevent human error. 5 Perioperative checking and timeouts are current surgical examples however until now there has been little system sense-checking of intraoperative surgical decisions. Digital assurance of perfusion sufficiency or critical anatomy using near-infrared fluorescence guided surgery is increasingly accepted as one such adjunct paving the way for AI-augmented surgical video (although the tradition of implicit confidence in surgeon's decisions meant it arguably faced much greater resistance and required much greater rigor in its testing then dexterity-enhancing innovative technologies such as robotic-assistance). 6,7 Role for intraoperative AI in surgery today As much as representing reality to the surgeon, surgical video so creates data amenable to computational and/or AI analysis. Some AI systems that work on surgical video data in real-time are simply surgical tools that act via passive externalism (i.e. digitally diminishing cautery smoke obscuration of the surgical field). Other AI systems directly serve as partners in decision-making at critical points in the surgery (i.e. active externalism). 2 When AI systems move into this space and become tightly coupled with the surgeon's decision-making, they form part of the "Surgical Extended Mind". An approach that treats all AI applications in surgery as "just another tool" without recognizing this important distinction underserves both the potential risks and benefits of AI in surgery. On the other hand, realising that AI can affect a surgeon's consciousness and behaviour (such as the surgeon becoming dependent on or uncritical of the AI) enables approaches that mitigate risks, including those arising from poorly trained or deliberately poisoned algorithms, and define accountability for AI systems.The integration of artificial intelligence into surgical decisionmaking may also be interpreted through the lens of exbodiment, a developing theoretical framework in cognitive science and theoretical biology that extends embodiment and extended-mind perspectives emphasizing the inseparability of perception, action, and environment 8,9. Within this view, cognition is not confined to neural processes alone but can be partially outsourced to engineered matter, such that external artifacts function as computational collaborators and co-evolve with human cognition as components of a single adaptive system. Organisms are understood to construct niches that compensate for biological constraints; in this sense, AI systems may be regarded as technologically constructed cognitive niches that enable surgeons to mitigate limits of perception, memory, and bias. The long-term aim of surgical AI integration would therefore not be replacement of human judgment, but the emergence of support systems sufficiently fluid that their use approaches incorporation into skilled action. Consistent with extended-mind accounts, repeated interaction with external tools can lead to internalization of their operational logic, reflected neurally in shifts from effortful prefrontal engagement toward premotor and sensorimotor representations as expertise develops. Notably, exbodiment also emphasizes the productive role of constraint: imperfect or limited tools can reorganize cognition by prompting novel strategies and forms of problem solving.Interaction with imperfect but informative AI systems may therefore foster new modes of surgical reasoning, provided that responsibility, interpretability, and human oversight remain clearly structured.technologies capable of influencing the cognitive processes that drive intra-operative decisions impact the legal standard of care used to assess surgeon's liability (see supplementary Table 1). At its core, the standard of care seeks to determine whether the surgeon's decisions were objectively reasonable. 10 As AI extends into the surgeon's cognitive process to aid decision making, liability law may need to shift in its reasonableness assessment. On the one hand, advanced AI systems that help surgeons overcome limitations of human perception may mean that misperceptions that were once reasonable may become unreasonable and lead to liability. For example, anatomical misperceptions that cause a surgeon to injure a patient's bile duct in a laparoscopic cholecystectomy may breach the standard of care if (1) AI technology was available to improve intraoperative dissection decisions and (2) a reasonable surgeon should have used that technology. On the other hand, if AI systems influence surgeon perception to cause cognitive errors that lead to unreasonable surgical decisions, surgeons may be liable for failing to realise this. In any case, intraoperative reliance on AI must be assessed for reasonableness regardless of whether the AI-influenced decision turned out to be right. 10 Reasonable use of AI for cognitive support in surgery also requires the surgeon understands and mitigates its risks including overreliance and automation bias. AI developers and healthcare organizations also have responsibility for safe design and implementation of AI systems (and should share liability when it fails).Surgical video, like other data can also be recorded for later review although typically is not at present routinely made part of the medical chart (the surgeon's operative note as summary of the surgical act takes this place). As the capability to store and process large amounts of data becomes easier, operative recordings are now seen as rich data sources.However if surgery is embodied cognition-in-action, surgical recordings can be seen as external memory of the actors. Ascribing ownership and rights to such recordings becomes layered (attribution is important for legal discovery and increasingly too as a source for AI model training and testing).While the governing medical institution can claim to own the video recording it captured through its resources and policies (much like how medical records are viewed), patients or surgeons (present via their actions derived from their perceptions, interpretations and decisions) may also claim access or indeed credit and any usage is restricted by personal privacy laws (such as GDPR in the EU). Furthermore, subsequent viewings, including analysis of operative perception and performance quality, bring in new cognitive inferences from the new viewer including hindsight bias and retrospective clarity when the outcome is know free from real-time perceptual limitations and pressures and so create a "liability trap" unless properly appreciated. While AI can provide re-analysis of operative recordings, we need be careful not to allow it retrospectively ascribe "missed" anatomy or mistakes that the unaided surgeon could not have realistically detected given normal human constraints (reasonable human misperceptions may not breach the standard of care which demands reasonable, not perfect or super-human, performance).Surgical cognition should therefore be understood as a distributed system spanning neural processing, technological mediation, and institutional responsibility. Recognizing intraoperative AI not merely as an instrument but as a component of cognitive architecture reframes both surgical expertise and accountability. Future governance of intraoperative AI should therefore align technological design, clinical training, and legal standards with this emerging model of distributed intelligence. perform functions typically done in the "head" and meet certain criteria including being constant and easy and directly useability with outputs that are automatically endorsed (Fig 1(a)). In such coupled systems, the external element is crucial to cognition (i.e. removing the external component reduces cognitive competence akin to losing part of the brain). Surgical AI decision-support systems fulfil these criteria and so are distinct to even similar systems being deployed postoperatively (e.g. for video analysis). Surgical extended mind systems need therefore (Fig1b) augment human cognition by being (i) deserving of trust and user friendly to alleviate mental strain (and not add to it by providing excessive or irrelevant recommendations)(ii) share error responsibility and accountability with the clinician along with other system framework components (e.g. the data feed) as suboptimal environments may inhibit perception and (iii) operate with seamless integration (Fig 1b).The role of technology in surgery has evolved from external instruments supporting embodied skill toward cognitively integrated systems that participate in perception and decision-making. AI-assisted surgery may represent a transition from extended cognition toward exbodiment, in which human expertise, engineered systems, and institutional structures co-adapt within a distributed cognitive framework.
Cahill et al. (Thu,) studied this question.