#1
See my recent blog highlighting an article on the future of pathology and pathologists in an increasingly AI-driven world. (El-Khoury & Zaatari). Here.
#2
See a Chat GPT history of autopilots in airlines. Here.
#3
Below, what do we get from #1 and #2 together? I asked Chat GPT to compare the history of autopilots in airlines (1940-2025) with the projected future of pathologists in an AI environment (2025-2050).
###
AI CORNER
###
Below is a compact, first-time-reader–friendly summary of each uploaded document, followed by a contrast-and-compare section applying the lessons from 100 years of aviation automation to El-Khoury & Zaatari’s vision of a potential “pathologist-free” diagnostic model.
I cite each document explicitly. First, we provide an executive summary and then we go over the data and resources in more detail.
#######
Automation, Autonomy, and the Future Role of the Pathologist
Pathology stands at an inflection point not unlike the moment in aviation when the autopilot shifted from a simple stabilizing device to a fully integrated flight-management system. As El-Khoury and Zaatari describe, pathology’s technical arc—microscopes, histotechnology, digital slides, deep learning systems, and now multimodal foundation models—has advanced to the threshold where a “pathologist-free” diagnostic workflow can at least be imagined. The question is no longer whether AI can read certain slides as accurately as pathologists; it demonstrably can. The question for leadership is how far autonomy can go, how quickly, and what role human pathologists will play in the new diagnostic ecosystem. Aviation’s century-long experience with automation provides a strikingly apt framework for thinking about this future.
Airlines acquired technically capable autopilots decades before anyone considered removing pilots from the cockpit. By the 1960s, autopilots could fly an airliner from shortly after takeoff to just above the runway threshold. By the 1970s and 1980s, digital flight computers could manage navigation and engines as a unified whole. In recent years, autonomous taxi–takeoff–landing sequences have been demonstrated, and emergency “no-pilot” landing systems already exist in certified form on small aircraft. Yet pilots remain indispensable. Their role evolved from manual controllers to systems integrators, safety supervisors, exception managers, and the human face of accountability. Technology reached near-autonomy long before institutions, regulators, insurers, or the travelling public were prepared to operate without human professionals.
Pathology is now entering the comparable integration phase. Digital pathology has solved the logistical and technical barriers that previously constrained automation; task-specific deep learning systems reliably detect metastases, grade tumors, estimate biomarkers, and identify patterns that even experts may miss; and emerging foundation models—such as those capable of integrating histology, IHC, genomics, radiology, and unstructured clinical text—are beginning to function like interactive copilots. These systems can draft reports, propose ancillary stains, and even defend their reasoning when prompted. In isolation, each capability looks narrow; in aggregate, they begin to resemble a diagnostic autopilot system.
El-Khoury’s “pathologist-free workflow” pushes this thought experiment further, envisioning oncologists or other referring physicians interacting directly with AI diagnostic platforms. The thought exercise is provocative precisely because the technical elements are now visible: the slide scanner as the sensor array, the foundation model as the flight computer, and the clinician–AI conversational portal as the human–machine interface. But aviation teaches us that the existence of a capable machine does not eliminate the need for an accountable, domain-trained human supervisor. Tasks can be automated; professions tend to evolve rather than disappear.
Two realities anchor this conclusion. First, diagnostics—like flying—contain rare and unpredictable failure modes that cannot be exhaustively encoded in training data. Edge cases, domain shifts, staining anomalies, artifacts, bias, unexpected clinical contingencies, and out-of-distribution histology samples remain intrinsic to real-world practice. Second, professional liability, regulatory oversight, accreditation, and public trust generally require a clearly identifiable human expert responsible for safety-critical decisions. Aviation never eliminated pilots despite nearly full technical autonomy; instead, pilots became responsible for detecting when automation was wrong. Pathology’s future is likely to mirror this trajectory.
For pathology leadership, the key strategic insight is that the profession is not being replaced—its center of gravity is moving. The field is shifting from manual microscopic interpretation toward roles that emphasize supervisory responsibility, multimodal data integration, AI governance, quality assurance, and rare-case expertise. Pathologists will increasingly resemble systems-level diagnosticians who adjudicate, contextualize, and validate machine outputs, rather than primary engines of image-level pattern recognition. Departments that embrace this transition will need to invest in new competencies such as AI literacy, workflow engineering, data curation, and validation science, while maintaining deep expertise in classical morphological judgment for the cases where automation fails.
In sum, the pathologist of the future is unlikely to vanish; instead, they will inhabit a role analogous to the modern pilot—less at the joystick, more at the center of a complex safety-critical information system. The leadership task is not to defend the past but to architect this future consciously: to define the skills, workflows, educational models, governance standards, and regulatory relationships that ensure pathologists are not bystanders to the rise of AI but active designers of the next era of diagnostic medicine.
######
1. Summary of the Autopilot History Document
The autopilot history describes how aircraft automation evolved from simple mechanical stabilizers in the 1930s to the digitally integrated, AI-assisted flight-management ecosystems of today.
1930s–1960s.
Airliners adopted basic autopilots much earlier than most people expect: 1930s aircraft like the Boeing 247 already had heading/attitude hold. By the 1940s–50s, three-axis autopilots were standard on long-haul jets. The 1960s brought revolutionary ILS-coupled systems capable of near-automatic landing.
1970s–2000s.
Digital flight computers replaced analog servos; flight directors, autothrottles, and triple-redundant systems matured. In the 1980s the Flight Management System (FMS) unified navigation, engines, and autopilot into one coordinated computer, enabling gate-to-gate automation.
2000s–2025.
GPS, RNP, synthetic vision, and AI-based anomaly detection enabled unprecedented precision. Research projects such as Airbus ATTOL achieved fully autonomous taxi–takeoff–landing sequences. Garmin’s general aviation Autoland demonstrated a zero-pilot emergency landing system.
Overall arc: Automation grew from “pilot aid” → “flight manager” → “supervised autonomy,” with pilots gradually shifting from manipulators of controls to systems-level supervisors and safety integrators.
2. Summary of El-Khoury & Zaatari (2025) on AI in Pathology
“The Rise of AI-Assisted Diagnosis: Will Pathologists Be Partners or Bystanders?” (here)
This comprehensive review traces pathology’s technological evolution—from microscopes, microtomes, H&E, IHC, and digital slide scanning—to the rise of deep learning and emerging multimodal foundation models.
The authors highlight:
Digital pathology as prerequisite for AI
Scanning glass slides into whole-slide images solved logistics, archiving, and collaboration challenges and created the data substrate for AI.
Task-specific deep learning success
AI systems have matched or exceeded pathologists in narrow tasks: metastasis detection, Gleason grading, MSI prediction, biomarker quantification, etc. Multiple systems (Paige Prostate Detect, Ibex, Owkin MSIntuit) have regulatory approvals or large-scale deployments.
General-purpose “foundation models”
Vision transformers and multimodal models (UNI, Virchow, PLIP, CONCH, THREAD, PathChat) integrate histology, text, and molecular signals. These “AI copilots” can draft reports, propose stains, or generate differential diagnoses.
The provocative question
Could a pathologist-free diagnostic workflow emerge?
The authors sketch a hypothetical model where:
-
Tissue is processed by lab staff.
-
Slides are scanned.
-
AI analyzes the WSI.
-
The referring physician—not a pathologist—interacts with a multimodal AI like PathChat.
-
Diagnosis and treatment decisions are made without a pathologist.
They note advantages (speed, consistency, cost) but emphasize unresolved issues:
• fragility to domain shift
• demographic bias
• black-box opacity
• legal liability
• the risk of deskilling the field
• unclear professional identity in an AI-dominant era
They outline three future models:
symbiotic, transformational, and the long-term, speculative disruptive “pathologist-free” model.
3. Comparison: What Pathology Can Learn from the 100-Year History of Autopilot
El-Khoury’s speculation about a “pathologist-free workflow” resembles past predictions about “pilot-free aircraft.” Aviation history provides a robust analogy—one that both supports and cautions against the feasibility of removing the human professional.
Here are the major parallels and divergences.
A. Both fields show the same three-phase automation arc
1. Assistance phase (1930s–1960s aviation ↔ present pathology)
-
Aviation: autopilot kept wings level and reduced fatigue.
-
Pathology: AI highlights regions of interest, counts mitoses, grades tumors.
Humans still do the integrative work.
2. Integration phase (1980s–2000s aviation ↔ emerging foundation models)
-
Aviation: FMS unified navigation, engines, and flight controls.
-
Pathology: foundation models unify histology, IHC, text, and genomics; AI copilots draft reports and propose ancillary tests.
Cognitive load shifts from “manual” to “supervisory.”
3. Supervised autonomy phase (2020s aviation ↔ speculative pathology future)
-
Aviation: autonomous taxi–takeoff–landing demonstrated; Garmin Autoland can land the plane with zero human input—but pilots remain mandatory.
-
Pathology: El-Khoury's “pathologist-free” model imagines that clinicians, with AI copilots, finalize diagnoses without a pathologist.
Key aviation lesson:
Even after the development of full technical autonomy, regulators, insurers, and the public kept pilots in the loop.
B. Where automation succeeded: high-constraint, rule-bound, sensor-rich tasks
Autopilot thrives because flying an aircraft is governed by:
-
continuous numeric telemetry
-
strict environmental physics
-
standardized maneuvers
-
globally harmonized procedures
-
rich sensor redundancy (radar, GPS, IRS, pitot, vision systems)
-
stable failure modes
Many narrow pathology tasks share these properties (e.g., quantifying Ki-67, detecting metastases).
C. Where autonomy stalled: socio-technical factors, not technology
Even when autopilots could handle every phase of flight, three barriers prevented pilot removal:
-
Regulatory accountability
Someone must be legally responsible for safety-critical decisions. -
Rare event management
When unexpected combinations occur—sensor icing, volcanic ash ingestion, structural failures—human adaptability remains essential. -
Public trust and liability
The public is not comfortable boarding a pilot-less plane.
Pathology exhibits stronger versions of these constraints:
-
Diagnostic liability cannot be delegated to an algorithm with non-explainable reasoning.
-
Rare events, unusual tumors, artifacts, and preparation errors require context-dependent judgment.
-
Clinicians lack pathology-specific training, unlike pilots who are deeply trained professionals.
Aviation demonstrates that “full technical capability” does not create “full social acceptance.”
D. The most important divergence: feedback & ground truth
In aviation, sensors measure the world in real time. Ground truth is physics.
In pathology:
-
Ground truth is disease, often ambiguous, multi-modal, and clinically contextual.
-
Errors may not be discovered for months or years.
-
Training labels are themselves imperfect (interobserver variability).
Therefore, the feedback loops that made aviation automation robust do not exist at the same scale in pathology.
This means autonomy in pathology is harder, not easier.
E. The “referring physician replaces the pathologist” scenario resembles 1930s idea that radio operators could replace pilots
Early aviation futurists argued that:
“If the autopilot can fly, the pilot becomes optional—let radio technicians run the aircraft.”
This never occurred because:
-
The pilot holds domain-specific judgment
-
Expertise cannot be transferred to an adjacent specialty
-
When automation fails, the wrong expert in the cockpit is catastrophic
El-Khoury’s provocative workflow (oncologist + AI without pathologist) runs into the same wall: the substituting professional lacks the domain expertise needed to detect when the AI is wrong.
In aviation this is called the automation paradox.
The more reliable the system becomes, the less skilled the human operator becomes—until, in an emergency, the human must suddenly perform a task they have not practiced.
A pathologist-free workflow amplifies this paradox:
-
If AI rarely fails, clinicians lose the habit of verifying.
-
When AI catastrophically fails, they are the wrong specialists to intervene.
4. Synthesis: What new perspectives emerge from comparing both histories?
1. Technical capability arrives decades before social, regulatory, and professional acceptance.
Aviation teaches us that even perfect automation does not eliminate human professionals.
Pathology will likely follow the same pattern.
2. The future role of the pathologist is supervisory, integrative, and safety-critical.
Just as pilots became “systems managers,” pathologists may become:
-
AI system supervisors
-
diagnostic integrators
-
data-quality guardians
-
explainability adjudicators
-
rare-case experts
-
authors of AI training sets
This aligns with El-Khoury’s “transformational model,” not the “disruptive pathologist-free” model.
3. Eliminating the pathologist creates new failure modes, just as eliminating pilots would.
Aviation’s worst automation disasters arise when:
-
automation is trusted too deeply
-
oversight skill has atrophied
-
the supervising human lacks domain expertise
A pathologist-free workflow creates the same structural risk.
4. The analogy strengthens a counterargument to replacing pathologists:
Even when automation is mathematically superior at flying, humans remain responsible for judgment, oversight, and edge cases.
By analogy:
Even if AI eventually outperforms pathologists in most image-based tasks, professional oversight remains essential.
5. Aviation history shows that autonomy disrupts tasks, not professions.
Tasks shift:
-
Pilots no longer manually navigate using celestial sextants.
-
Pathologists may no longer count mitoses or grade tumors manually.
But the profession remains because:
-
accountability
-
edge-case complexity
-
multimodal integration
-
public trust
-
regulatory requirements
El-Khoury’s “pathologist-free” proposal therefore becomes less plausible when mapped onto the aviation trajectory.
Bottom Line
Autopilot history strongly suggests that even complete technical autonomy does not remove human experts from safety-critical systems.
El-Khoury’s article raises the possibility of a “pathologist-free” model, but the aviation analogy reveals that:
-
Full automation is technologically feasible sooner than it is socially acceptable.
-
Humans remain indispensable for oversight, rare events, edge cases, liability, and public trust.
-
Professions adapt rather than vanish.
Therefore, the more realistic future—mirroring aviation—is not:
AI replaces the pathologist,
but:
AI replaces parts of what the pathologist does, while elevating the pathologist into a supervisory, integrative, more cognitive role.
