The use of Complexity Science to serve the study of sociotechnical systems began in the 90s, mainly through the work of Pavard (2003) and the COSI Training Network. This line of work was pursued by colleague and friend Nikos Zarboutis, who presented this paper co-authored with Peter Wright in 2006. Because of my military tour of duty I was unable to read it until now, but it is quite interesting.
Nikos argues for the use of Complexity Science as a tool to study sociotechnical systems, as the dominant methods of cognitive engineering (e.g. Rasmussen et al., 1994) fail to explain the causality that leads to accidents.
However, their value remained mainly descriptive, in terms of “regularities†in certain domains (e.g. healthcare) and they failed to capture the structural changes that link higher level patterns with systemic failures. Furthermore, such patterns could not be used for the enhancement of the system’s adaptive capabilities, as they lacked a methodological framework that could link such structural changes to systemic failures, in order to come about solutions that could assure the elimination of the whole class of events that stem from them and to eventually turn them into design recommendations.He then proceeds to give a more precise account of co-adaptation in worksystems(I used it more freely through the ecological metaphor of co-evolution), which sometimes can also have negative consequences:
In general, co-adaptation is the process where two agents (or agencies, depending on the chosen level of granularity) adapt to the same problem, each pursuing its own private goals. For example, during a fire in a refinery, two agents can both adapt to the same problem (e.g. some leak), when someone chooses to cut a flow off from a pipeline locally, while some other controller may be trying to restore it centrally, each following a different plan to deal with the fire, each being unaware of the actions of the other. In the bigger picture, the agents need not be human; it could be a human interacting with a technological agent (e.g. a computer) where the human may pursue different objectives than the machine, for a number of reasons.
The paper points out three recurring patterns that can obstruct regular functioning of the complex system:
- Self-reference
- Infinite Loops
- Stigmergy
He then describes an accident in the Nagoya airport that demonstrates this. An interesting alternative explanation closer to my line of work is the one highlighted by Norman (see also the recent post): The goals of human and intelligent system were conflicting. The inelligent system is another actor, just like the pilot and it had a different goal, which the pilot attempted to counter on a lower level - action.
The critical events that led to the accident, as appeared in the investigation report, (Sogame & Ladkin, 1996) begin after the First Officer inadvertently triggered the go-lever. As a result the Flight Director switched from “land mode†to “go-around†mode, initiating a process to abandon the landing and gain altitude quickly and safely so that another attempts at landing could be made. The First Officer, although eventually aware that the lever had been triggered, tried to land the aircraft by applying “nose down input†and reducing thrust. However, with the go-around mode active, the more the pilot applied nose down force to the sidestick, moving the elevator accordingly, the more the autopilot moved the Trimmable Control Stabilizer in a nose-up direction to counteracts the pilot input because, as it was clear, it was still executing commands that would ensure a safe go-around (i.e. gain power and height). After a series of human adaptations.and automatic adaptive responses, including the activation of the alpha floor function which increased thrust levels as well, the plane began to adopt a very steep climb. Despite the pilot’s final attempt to actually go-around, when it was obvious that he could not land, the plane stalled and crashed over the runway!
The re-examination of these events under the prism of the complexity paradigm reveals that many co-adaptive phenomena took place, where the human pilots and the auto-pilots were trapped in a self-referent situation interacting indirectly through stigmergy, both adapting on conflicting objectives.
References
Pavard, B. (2003). De l’ingenierie Cognitive a la Theorie des Systemes Complexes. Un Parcours d’analyse et de Modelisation de l’activite Centree sur la Conception. Paper presented at the 38ieme congres de la SELF.Rasmussen, J., Pejtersen, J. A., & Goodstein, L. P. (1994). Cognitive Systems Engineering. New York: John Wiley & Sons.
Sogame, H., & Ladkin, P. (1996). Aircraft Accident Investigation Report 96-5. Japan: Aircraft Accident Investigation Comission.
Zarboutis, N., & Wright, P. (2006). Using Complexity Theories to reveal emerged patterns that erode the resilience of Complex Systems. In E. Hollnagel & E. Rigaud (Eds.), Proceedings of the 2nd Symposim on Resilience Engineering (pp. 359-368). Juan-les-Pins, France.
technorati tags:complexity, CognitiveErgonomics, Safety, Resilience, HumanError
Leave a Reply