Digital diagnostic pathology between promise and professional erosion: artificial intelligence and the future of diagnostic sovereignty

Introduction

Diagnostic pathology is currently experiencing a period of profound technological transformation. Digital slide scanning, artificial intelligence (AI)–assisted image analysis, integrated laboratory information systems and automated workflow solutions are increasingly being introduced into routine practice. These developments hold considerable promise. They may improve efficiency, facilitate remote consultation, support quality assurance and help address the growing imbalance between rising diagnostic workload and the limited number of trained pathologists.^1–3 Industry has emerged as a key stakeholder in shaping how the future of pathology is imagined. Promotional narratives frequently describe a future characterised by real-time reporting, workflow automation, efficiency and integrated AI. While these visions reflect technological ambition, they also shape expectations regarding the organisation of diagnostics and are often presented as inevitable developments.

From the perspective of an experienced practising academic pathologist, this question merits careful reflection. The purpose of this essay is not to question the value of digital pathology. On the contrary, these technologies represent important and welcome advances. Rather than introducing new technical aspects of AI, this essay synthesises well-recognised developments into a conceptual framework that examines their implications for diagnostic sovereignty, professional identity and the evolving role of pathology.

Speed as a value—and as a norm

Few ideas capture this transformation more clearly than the emphasis on speed. The claim that the majority of pathology reports can be reviewed and released within seconds is increasingly framed not merely as an option but as a benchmark for digital diagnostics. Speed is, of course, clinically relevant. Faster diagnoses may shorten clinical decision cycles and improve patient care. But pathology is not merely a logistics service. It is a discipline built on deliberation under uncertainty.

However, speed must not compromise diagnostic quality. It is important to counter unrealistic expectations and to recognise that not every acceleration improves patient care. When velocity becomes a surrogate for quality, diagnostic reflection risks being reframed as inefficiency. Although these issues are well recognised, their integration into digital, metric-driven infrastructures may transform them from isolated concerns into structural drivers of how pathology is evaluated and practised. True diagnostic quality that translates into improved patient outcomes is difficult—if not impossible—to quantify without long-term, interdisciplinary follow-up. Hence, it is unsurprising that surrogate metrics, particularly turnaround time, have come to dominate performance assessments.⁴ Such metrics may inadvertently incentivise rapid diagnostics at the expense of diagnostic accuracy and long-term patient well-being. It should be acknowledged that automation and AI in pre-analytical processes—such as specimen tracking, slide preparation and workflow logistics—can significantly improve efficiency without directly interfering with diagnostic reasoning.

Artificial intelligence and the illusion of the ‘second pair of eyes’

AI is often introduced as a reassuring corrective: a “second pair of eyes”. The metaphor is attractive—but potentially misleading. A second human observer brings independent judgement, shared professional norms, accountability and clinical context. Algorithmic systems do not. They operate within predefined data spaces, statistical thresholds and training biases. They cannot assume responsibility, nor can they share liability. However, as AI systems become increasingly embedded in diagnostic workflows, the practical distribution of decision-making, validation and accountability may become more complex; the ultimate clinical responsibility for diagnostic interpretation remains with pathologists. If algorithmic systems begin to contribute substantially to diagnostic decisions, the question of who ultimately bears responsibility—pathologists, clinicians, institutions or technology providers—will become an important issue in the evolving governance of AI in medicine.

Automation bias may lead users to rely on algorithmic suggestions despite contradictory evidence.⁵ When AI outputs are directly integrated into diagnostic workflows, they may therefore exert a subtle anchoring effect: what is generated first is more likely to be accepted. Safety in such environments risks becoming technical rather than professional—an illusion of redundancy rather than a genuine second judgement. This illusion appears to be welcomed by many clinicians. Despite limited long-term outcome validation and unresolved concerns regarding bias and generalisability, AI-assisted systems are often perceived as increasing confidence in pathology reports, reflecting a broader phenomenon of technological trust. Of course, in settings with limited or absent pathology services, AI-assisted diagnostics may provide valuable support. However, such applications should not be interpreted as equivalent to comprehensive diagnostic practice, particularly given the unresolved challenges related to validation, generalisability and accountability.

Experience in pathology extends beyond case volume to contextual reasoning and recognition of atypical patterns. These dimensions of experience are not yet replaced by algorithmic performance and remain essential precisely in those areas where automated systems are most vulnerable. In practice, AI may also serve as a useful prompt, encouraging the pathologist to reconsider areas of potential discrepancy. When applied after an initial diagnostic assessment, such systems may function as a form of augmented intelligence—‘are you sure?’—rather than as a primary diagnostic driver. This sequential use may help mitigate anchoring effects while preserving the primacy of human judgement.

Metrics, transparency and the managerial gaze

Integrated dashboards, performance statistics, turnaround metrics, and individual productivity indicators are increasingly presented as hallmarks of modern pathology laboratories.⁶ However, when measurement becomes the dominant lens through which professional performance is evaluated, what is easily quantifiable gains authority. Speed, volume and billing efficiency become visible and comparable, creating an illusion of scalability, whereas diagnostic complexity, uncertainty, consultation and teaching remain largely invisible. Over time, this imbalance may reshape professional identity. Pathology risks being continuously evaluated—and eventually valued—not for diagnostic insight but for throughput, a tendency that may be especially detrimental in academic environments. Beyond report generation, pathology provides integrative clinical interpretation, contributes to multidisciplinary decision-making and supports molecular stratification—roles not captured by throughput metrics.

What is systematically missing in AI-based pathology visions

Experience, education and responsibility are fundamental elements of pathology, irrespective of AI. However, across many contemporary narratives of digital transformation, these dimensions remain strikingly underrepresented. This relative underrepresentation may become increasingly relevant as such narratives begin to shape implementation strategies and expectations in routine practice. Experience refers to knowledge that informs diagnostic judgement beyond formal criteria. Education reflects the formative role of pathology in medical training, where disease mechanisms are visualised and contextualised across organ systems, linking clinical presentation, morphology and pathophysiology in a way no other discipline provides. Responsibility includes the clear articulation of who ultimately decides, who answers and who bears clinical risk. Without these elements, pathology risks being reframed as a technical service rather than a medical profession.

Digital transformation requires reaffirming that pathology is a medical discipline grounded in judgement, responsibility, experience and trust. AI should support rather than redefine diagnostic judgement. Speed should remain a means rather than become a defining value. Metrics should inform practice rather than redefine professional worth. If these boundaries are not actively defended, pathology risks sliding—quietly—into pure service-orientation. Once perceived primarily as a provider of fast, standardised outputs, pathology should not be surprised if it is no longer regarded as an equal partner by clinical colleagues, but rather as an auxiliary discipline. Ironically, such developments may sometimes be accelerated by voices within the profession itself. Reframing pathology as “tissue medicine” implies a narrowing of scope that risks abandoning its broader intellectual foundation. Pathos and logos—the study and interpretation of disease, the core of academic medicine—cannot be reduced to material processing without losing their identity.

Conclusion

Digital diagnostic pathology offers extraordinary opportunities that must be thoughtfully integrated into the evolution of our medical discipline. Digital workflows and AI can improve efficiency, enhance diagnostic support, enable novel diagnostic categories and expand the reach of pathology services. These developments should therefore be welcomed. At the same time, however, the transformation of pathology into an increasingly digital discipline should not be understood solely as a technological project. It is equally a professional and institutional transformation that raises fundamental questions about responsibility, judgement, training and clinical integration.

It appears essential that digital transformation strengthens rather than replaces the professional foundations of the discipline. AI should remain assistive rather than constitutive, automation should support deliberation rather than replace it and speed should remain a means to improve care rather than become a defining value in itself.

If pathology relinquishes its diagnostic sovereignty in favour of speed, scalability and managerial optimisation alone, it risks becoming a commoditised technical service. The future of pathology must therefore be shaped not only by what is technologically feasible but also by what is professionally defensible, medically meaningful and intellectually sustainable. The central question is therefore no longer whether diagnostic pathology will become digital but whether, in becoming digital, pathology will continue to function as a sovereign medical profession grounded in judgement, responsibility and clinical interpretation. To ensure a responsible integration of AI into diagnostic pathology, several principles should be considered. First, governance structures must clearly define accountability and decision-making authority. Second, AI systems require rigorous validation across diverse clinical settings before implementation. Third, their integration into workflows should preserve independent diagnostic assessment by the pathologist, rather than replace it. Finally, outcome-based evaluation—rather than surrogate metrics alone—should guide the assessment of their clinical value.