Safety Science
Safety science is the interdisciplinary study of how organizations, technologies, and socio-technical systems fail — and how they can be designed to fail less catastrophically. It draws on sociology, psychology, engineering, organizational theory, and systems theory to understand the structural, cognitive, and cultural conditions that produce accidents, and to develop interventions that reduce their frequency and severity.
The field has evolved through three major phases, each corresponding to a different theory of accident causation: the human error phase (1900s–1960s), the organizational accident phase (1970s–1990s), and the resilience phase (2000s–present). Each phase retained insights from the previous one while expanding the analytical frame to include larger systems and longer timescales.
Phase I: Human Error
The earliest safety science focused on individual operators. Accidents were understood as the result of human error — fatigue, inattention, lack of skill, or violation of procedures. The solution was better training, better selection, and better enforcement of rules.
This framework produced genuine improvements in safety, particularly in domains like aviation and manufacturing where standardization was possible. But it had a critical limitation: it could not explain accidents in which every individual operator followed every procedure correctly, yet the system still failed. The Three Mile Island accident (1979) was the canonical case: operators followed their training, yet the reactor melted down because the system's design had not anticipated the specific combination of failures that occurred.
Phase II: Organizational Accidents
The second phase, pioneered by James Reason, Charles Perrow, and Jens Rasmussen, shifted the analytical frame from individual operators to organizational systems. Reason's Swiss cheese model showed that accidents occur when multiple defensive layers fail simultaneously — and that these failures are often latent conditions created by organizational decisions (budget cuts, production pressure, design choices) long before the accident occurs.
Perrow's Normal Accidents theory pushed the analysis further, arguing that some accidents are structurally inevitable — the normal output of systems that are both interactively complex and tightly coupled. The policy implication was radical: for such systems, safety cannot be achieved through better procedures. It requires structural redesign.
Rasmussen's dynamic risk management model showed that organizations migrate toward the boundaries of safe operation under production pressure. Safety is not a static equilibrium but a dynamic process in which the organization continuously trades safety margins for production efficiency. The accident occurs when the migration goes too far — when the organization's operating point crosses the boundary into the region of unacceptable risk.
Phase III: Resilience
The third phase, associated with David Woods, Erik Hollnagel, and the resilience engineering community, shifted the frame again — from why do accidents happen? to how do systems succeed under uncertainty? This reframing, called Safety-II, treats safety not as the absence of failures but as the presence of capacity — the capacity to absorb disturbance, adapt to change, and recover from perturbation.
Resilience science studies how organizations maintain function in the face of variability, surprise, and stress. It focuses on adaptive capacity — the ability to reconfigure structures and behaviors in response to novel conditions — rather than on error elimination. The methods include resilience engineering, high reliability organization studies, and the analysis of work-as-done (how operators actually perform work) rather than work-as-imagined (how managers think work is performed).
The Synthesizer's Take
The three phases of safety science are not a progression from wrong to right. They are a progression from simple to complex — each phase addressing failures that the previous phase could not explain. Human error is real and important; organizational accidents are real and important; resilience is real and important. The mistake is to treat any one phase as sufficient.
The most dangerous tendency in contemporary safety science is the belief that technology can solve safety problems. Better sensors, better algorithms, better automation — these are valuable, but they do not address the fundamental challenge: that safety is a property of socio-technical systems, not of technologies alone. A perfectly designed automated system can fail because the organization around it has not maintained the capacity to intervene when the automation encounters conditions outside its design envelope.
The future of safety science lies in integration: combining the precision of human factors research, the structural insight of organizational sociology, and the dynamical perspective of systems theory into a unified framework that can address the complex adaptive systems we now build — systems that learn, evolve, and generate behaviors that no designer anticipated.
Safety is not a product. It is a process — a continuous negotiation between the system's desire for efficiency and its need for resilience. The organizations that succeed are not those that eliminate risk but those that learn to live with it, to detect it early, and to recover from it quickly.