SA Anno 1995

Introduction

In reading the contributions to this issue, it is heartening to see the deliberations on what we, as a human factors community, have to offer operators, designers, and others working with complex, dynamic systems. And indeed, we should all welcome opportunities to jointly assess whether our grasp exceeds our epistemological reach. That is, do we put more faith in some of our own ideas and models than is warranted? Epistemology asks how we know what we know. If we claim to know a lot about our subjects’ awareness on the basis of one model of situation awareness (SA; and its assorted measurement techniques), those claims should obviously be subject to some examination, as they are here. Faith in our models sometimes derives from their popular adoption and large bodies of empirical evidence generated from that. But this has little to do with inherent scientific quality. That the widespread application of a model is able to deliver facts about the existence of its own constructs is not at all persuasive—at least not since Kuhn (1962). After all, most human factors constructs can claim to be supported by an ample empirical record, whether SA, mental workload, or automation surprises. The empirical record becomes available, after all, because the models themselves generate the demand for the data and the criteria by which their evidence is defined. Thus, ever-greater accumulation of facts does not equate with scientific quality and certainly not with innovation or scientific evolution.

The 1995 model of SA represents a 17th-century ontology. Ontology is the branch of philosophy that inquires after the nature of being in the world—and for SA, as for 17th-century thinkers, like Rene Descartes and (a bit later) Isaac Newton, such being in the world means having an accurate representation of the world in the mind. A separation of mind and matter (or head and world) is fundamental to this view, and it has dogged cognitive science well into the 20th and even the 21st century (Norman, 1988). Information processing (roughly popular from the 1950s to the 1970s in cognitive psychology), for example, has faithfully replicated 17th-century ontology. SA, too, is that correspondence with the environment; it is that mirror, that mimic, that mental simile of the world outside. Protestations to the contrary—“SA . . . goes beyond traditional information-processing approaches in attempting to explain human behavior in operating complex systems” (Endsley, 2015, p. 9)—do not make it less so. SA is grafted onto the exact language, concepts, and flow charts of traditional information-processing models. And indeed, with such 17th-century dualism as its starting point, notions of “complete” or “accurate” SA become sensible and intelligible. For SA, “there is a ‘ground truth’ against which its accuracy can be assessed (e.g., the objective state of the world or the objective unfolding of events that are predicted)” (Parasuraman, Sheridan, & Wickens, 2008, p. 144). It is an ontological commitment with consequences, however. It makes sacrifices, both epistemological and ethical, that deserve some reflection.

Dynamics and the Environment

Contesting the critique that the model is not dynamic, the argument is that the “model shows a dynamic feedback loop for gathering information and acting on the environment” (Endsley, 2015, p. 11–12). This constitutes a 1950s cybernetics approach to operationalizing dynamism. The basics of interaction are there, but there is no escape from the static, linear, reductionist arrows and boxes. Since a decade or two, the human factors community has largely committed to the second cognitive revolution (the first was the move away from behaviorism and into information processing in the 1950s), which sees the environment as an active determinant of and participant in cognitive processes. SA, as John Flach (1995) would say, is largely about the A and much less about the S. If “not a ‘strictly in the head’ model” (Endsley, 2015, p. 16), it surely is a largely in-the-head model (the environment, or human interaction with it, is depicted as a single arrow flowing from “State of the Environment” toward the SA “box”; see Endsley’s [2015] Figure 1). The major preoccupation—theoretically, methodologically—is a subject’s awareness. This focus sacrifices our study of the environment, the situation, as both constitutive of and constituted by that awareness.

The commitment to awareness “in the head” is so strong that Hutchins’ (1995) work, as that of others in joint cognitive systems (JCS; Hollnagel & Woods, 2005; Woods & Hollnagel, 2006), is deemed not heretical but simply absurd. Data can be all over a situation, but awareness is exclusively in the head: “I do not call such data, residing in a report or a display or an electronic system, ‘situation awareness.’ Inanimate objects do not have ‘awareness’ of the situation or of anything else” (Endsley, 2015, p. 26). Such privileging of human consciousness is not without philosophical or religious historical precedent. But what it seems to forget is the fundamental phylogenetic differentiation that occurred through tool use—we are human not just because we are aware but because we use tools to augment, supplant, and aid our cognition, our awareness. JCS thinking tries to take that into better account than “in-the-head” models of cognition ever could. Human cognition both constitutes and is constituted by operational, design, and organizational features of work, which suggests that human–machine interaction (the system) should be the unit of analysis (not the inside mapping of the environment in an operator’s head). Salmon et al. (2008) suggest how such a JCS approach to SA differs from the 1995 SA model:

The main difference between individual and team models of SA and DSA [distributed SA] approaches relate to the treatment of SA as a cognitive construct or as a systems construct. Individual and team models suggest that SA exists in the mind of individuals, whereas DSA approaches view SA as an emergent property or a product of the system itself. (Salmon et al., 2008, p. 313)

In reaction to such a systems approach, the conclusion is that “the implication of describing SA as ‘an emergent system property’ that is distributed across the system is that it essentially becomes sufficient as long as the needed SA is distributed somewhere in the system” (Endsley, 2015, p. 25). This of course misconstrues a JCS but also the notion of emergence. By taking the (joint cognitive) system as the system that is supposed to have a particular level of SA, instead of its individual components, we can regard SA as something that “emerges” during and from cognitive work. Alas, the defense of 17th-century ontology goes on: “Salmon et al. confuse sources of SA (displays, computers, and other artifacts) with SA itself, stating that the SA is in the artifact. This would only hold true if the artifact were itself a cognizant and independent decision maker” (Endsley, 2015, p. 26).

From this, however, only more confusion follows and, beyond it, not much else: “Information that exists in the environment, but which the decision maker is not aware of is by definition not situation awareness” (Endsley, 2015, p. 26). The definition of SA becomes increasingly circular. What about information that exists in the environment—is it even information at all unless it is perceived or conceived of as information by some kind of cognizant and independent decision maker? In that case, we do not need SA or its arrow feeding in from the environment. It also no longer makes sense to talk about accuracy or completeness of SA (as its adherents like to do). The point, not just epistemological but also practical, is probably to try to understand how people perceive the environment in ways that support action and goal achievement (Weick, 1995). On that, it is unfortunate that broader conceptualizations of “information” and “awareness” are dismissed out of hand (even though the importance of goal-driven or top-down processing is acknowledged). It slims allowable cognition down to what the 1995 model dictates. “While operators can and often do create heuristics and reminders in their work environment to assist them in keeping up with information and their task status, in my view these work methods and cues do not constitute SA” (Endsley, 2015, p. 17). Retaining a narrow definition of SA just because the model says so—and in the face of much evidence for the critical role of heuristics and recognition in dynamic work (Klein, 1998), or see the notion of “level IV evidence” in medical practice (Stuebe, 2011)—is a bit dogmatic and probably unhelpful for the evolution of our science.

The Ethics of the Effects of SA

The scientific object of SA has really taken off over the last 25 years. SA—and the loss of it—is everywhere: There are books, articles, journals, conferences, training courses, operating manuals, design guidelines, and accident investigation reports addressing it. The human factors community once embraced a specific and defined object of SA to help with the design of human–machine interfaces and to explain behavior in complex, dynamic environments. But is the object of SA also used in this way today? And who is responsible for the effects if it is not? As a human factors community, it is imperative that we engage in ethical discussions, asking questions about the consequences of the explanations, classifications, and practices we propose.

The wide use of SA has not only reinforced its status—the more it was used, the greater the consensus authority on using it—but also driven a gradual move away from its original purpose. It has offered new normative standards for behavior: Pilots now describe themselves as good pilots when they are situationally aware, when their decisions are informed by “good SA.” Soon after its conception, SA started to appear in accident investigation reports, where it was given great causal power to explain accidents (see, for example, Aviation Safety Council, 2002). It has even shown up in courts. Operators get convicted for not living up to their deontological duty to (always) maintain SA (Regina v. Karl-Heinz Arthur Lilgert, 2013). Even if SA is a “useful” construct in human factors, the manner in which other discourses—most prominently, the juridical—take and use it to ask (and answer) questions about error and culpability is an ethical problem that we as a human factors community need to acknowledge some responsibility for.

As long as the self-contained individual remains the unit of analysis for understanding SA (or other aspects of cognition), we are at least not helping other communities (like the juridical) move away from 17th-century characterizations about individuals and individual minds as the fundamental elements of thought and action. We are not doing anything to point these other communities to languages of systems, to new models of accountability and responsibility that are much better adapted to the flatter, networked societies that are so prominent and rapidly expanding in the 21st century. In fact, by supplying objects like SA, which reaffirm a 17th-century ontology, we might even be abetting them. As long as it is human factors orthodoxy to reduce safety problems to their component parts (either human or technical), and to separate what was in a world (which adherents assert is “objectively apprehensible”) from what was in the mind, we are asking for ethical trouble with our models and concepts.

In reflecting on 25 years of SA, we hope that human factors, and with it, SA research, is ready to move into the 21st-century by taking the lessons from the 20th at heart. The time has come to move away from our commitment to individualism, reductionism, and dualism and to move instead toward studying work in joint cognitive systems.