Decision making within military systems is set to become more distributed and team-based given the push for a Hardened Networked Army. Defence science, particularly within the human factors field, will need to adapt its research methods to meet this change. The Distributed Cognition (DCog) analysis is a context-sensitive approach to analysing and evaluating real-world cooperative work settings. A DCog analysis was carried out on a Royal Australian Artillery Regiment. The Regiment’s ‘cognitive system’ was represented by artefact, physical, and information flow models to allow an investigation into the possible effects of digitization. An Australian Artillery Regiment consisting of over 300 personnel was observed while on exercise. A large repository of video, photo and voice data was produced, which was subsequently analysed using a DCog analysis. The artefact, physical and information flow models allowed the recognition of specific issues which contributed negatively or positively to the overall performance of the system. The DCog description of the Regiment illustrates likely consequences of technology and automation for Artillery and the Australian Army.
Gunner Jones is a Gun Number in Australia’s Royal Regiment of Australian Artillery Corps. Like most members of the Royal Australian Artillery, the Gunner is proud to be associated with his Regiment and more importantly his specific Battery (unit of artillery guns), often boasting of their speed and efficiency. Interestingly however, the Gunner sees himself and his fellow Gun Numbers as different to the rest of the Regiment. He has not met and is generally unaware of any members outside his work area, though he may work alongside them at any given time. His world is truly bounded by his own Battery and the other Gun Numbers associated with it. Anything outside this world has little meaning in comparison.
In Hutchins’ (1995) in depth analysis of ship navigation, certain groups on board a Navy ship were seen to be culturally distinct but working toward a solution to a common problem. The complex task of navigating large vessels through busy ports was shown to be achievable only through the complex coordination of people and artefacts. No one person could navigate the ship at one time, but rather, effective navigation stemmed from a collaborative effort of individuals propagating partially encoded solutions. These partially encoded solutions were often passed between distinctly different groups and individuals, and as such, each group or individual would often only be cognisant of their own partial solution. This cognitive division of labour allows highly complex tasks to be completed relatively efficiently, and is also a key aspect in the world of Artillery.
1.1 The Field Gunnery Problem
The system goal of an Artillery Regiment is the efficient and accurate solving of the field gunnery problem. The field gunnery problem relates to the calculation of accurate targeting data such that first round accuracy can be achieved, culminating in target neutralisation or destruction. At a macro level, the problem is solved by following a series of well defined doctrinal steps: (1) Fix and orientate the guns, (2) determine the location of the target, (3) calculate map bearing, (4) map range and angle of sight, (5) convert map data to predicted data by compensating for any non-standard conditions, and, (6) issue orders to apply these data to the guns.
1.2 Distributed Cognition Theory and Analysis
The DCog approach assumes cognition is not a localised phenomenon within a specific individual, but rather, a distributed phenomenon including people, media, information and artefacts. The term artefact refers to any object or tool which is used by people to accomplish a task. The concept that knowledge may be as much in the world as it is in the head was popularised by Norman’s (1988) work on design. People are said to use ‘external cognition’, where information is created and used in the world around us, rather than purely inside our minds (Preece, Rogers & Sharp, 2002). We do this to reduce memory load (like using written reminders), simplify cognitive effort (like offloading computationally heavy problems onto external media) and trace changes (like erasing a destroyed enemy on a paper map). This idea of external cognition is central to the DCog approach.
Hutchins’ (1995) presentation of DCog focuses on how information passes through a system, and how that information is represented and transformed. Solving a problem consists of transforming representations of a problem through a problem space into a goal state (Perry, 2003). Here the critical system function and focus is on the transformation. In Artillery terms for example, a Forward Observer having visual contact of a target is only one representation of the problem, which has to translate to the Gun Numbers so a projectile can accurately reach the target. It is the propagation of this representational state of the problem which offers key insights into a system and its processes (Perry, 2003). In the current system for example, the Forward Observer would need to convert his visual identification of a target to map data before passing on to any other individual. This representational state therefore starts as an internal state (within the mind) and is then transformed to an external media (co-ordinates on a map).
2.1 Participating System
A Medium Artillery Regiment was the subject of the current DCog analysis. Basic elements of an Artillery Regiment include the following:
Joint Offensive Support Coordination Centre (JOSCC). The JOSCC provides the coordinating function for the Brigade’s offensive support requirement.
Regimental Command Post (RCP). The RCP coordinates regimental fire missions, and commonly issues orders and reports to JOSCC and BCPs.
Battery Command Posts (BCP). The BCPs are responsible for target prediction and are considered the most active of all Regimental elements given their workload. BCPs compute and transform targeting information and forward orders on to the gun line.
Joint Offensive Support Teams (JOST). The JOSTs are normally attached to one of the Brigade’s manoeuvre arms (Armour for example). The JOSTs identify enemy targets and forward that information on to the BCP for fire missions.
Battery Gun Lines. The gun line is where the Regiment’s Artillery assets sit. The gun line fires the actual round to target and is controlled by the BCPs.
Each Regimental element has key actors who are critical to system performance. These roles are outlined below:
JOSCC: Commanding Officer, signallers (radio operators) and aviation liaison staff.
RCP: Regimental Signals Officer, and an Operator Command Post (OPCP) who receives and passes regimental fire missions.
BCP: Gun Position Officer (GPO), an OPCP, various supervision staff and at least 2 signallers. The GPO gives the order for fires and informs the RCP of engaged targets and unavailable guns.
JOST: Forward Observer (FO), at least 2 signallers, and an FO assistant.
Gun Line: Each gun line normally holds six guns. The gun line personnel or “gun numbers” carry out all technical duties and fire orders from the GPO. Their duties are manual labour intensive.
2.2 Data Collection
The main data collection effort occurred when the author travelled to a military training camp to observe the Regiment on Exercise. The author was permitted unhindered access across the Regimental groups over four days of the Exercise. A large repository of video, photo and voice data was produced, which was subsequently analysed at DSTO Edinburgh.
An Artillery Regiment, just like any other functional human system, includes properties which are directly observable. By bounding certain portions of the system included in the analysis, allows us to expose the representations, their states, transformations and their coordination throughout the system. In Artillery terms, the system’s overall goal is to consistently and efficiently solve the field gunnery problem. Inputs are initiated by the JOST, with their requests for fire acting as a representational state traversing through the system via various Regimental groups and media. The output is the firing of round for effect, at the originally observed target.
3.1 The Information Flow
The information flow model provides a description of how information travels around the system, between regimental elements. For a fire mission, the system’s input will generally be target data from a FO. Information such as target location (map grid reference), target description, required ammunition, current weather and time propagate through the system in the task of destroying or neutralising a target (the system’s output). An information path for a Fire Mission Battery (FM BTY) would thus start at the JOST, with the FO observing a target. The FO will first determine the target’s location. The FO then passes his orders to the JOST signaller, within the same vehicle. The JOST signaller then simultaneously records the data in his logbook and transmits the order over the radio. Upon hearing the requesting order, the signaller and GPO within the BCP are alerted to attention. The BCP signaller begins writing down the data/orders in his logbook. The GPO at this point will select the specific guns to take part in the mission, and transfer the data to a paper ‘check map’, checking various safety factors.
With confidence in the data, the actors proceed to convert it within the BCP. The OPCP enters the map data into the gunnery computer, which then converts the map data to elevation and bearing data for the gun line. Using this transformed representational state, the GPO issues orders to the gun line, who then proceed to load and prepare the gun for firing. Knowing the orders have been passed, the JOST will maintain observation on the target waiting for the rounds to impact. On impact the JOST will assess the round’s accuracy.
In the FM BTY example given above, we begin to see the coordination required to achieve the system goal. Throughout the process, representational states, whether they be coordinates on a map or elevation data for a gun, are being propagated through the system as partial solutions. Each actor within each Regimental element has their own task in achieving the system’s goal.
At the beginning of the mission, the representational state is with the FO as a perceptual construct in his mind. He is observing the target, and thus the representational state is real world visual. The FO will then convert this visual into map data. At this point, each representational state will be based on a purely imagined situation within the heads of the system’s actors. No key actor other than the FO will have visibility of the target. The map and coordinate system are key artefacts which play an important role in representing the problem and eventually making the solution transparent.
After the FO’s representational state transformation, the JOST signaller will retain the representational state but record and propagate it on different media. The BCP signaller will conduct the same process being sure to read back the representational state verbally over radio. At this point the representational state has been transformed once and propagated across two media. It is only upon reaching the BCP that the representational state is transformed again. The GPO receives the FO’s partially encoded solution both in written (signaller’s notebook) and in verbal (over the radio) form. This partially encoded solution is map-based allowing the GPO to plot the target on a check map. With the OPCP instructed to convert that information to bearing and elevation data via the gunnery computer, we observe another transformation of representational state (elevation and bearing data in a computational state). The representational state enters the BCP in a form readily intelligible by the GPO, and leaves in a form understandable by the gun line. The BCP thus plays a coordinating function of representational states, rerepresenting information in multiple forms to solve the field gunnery problem.
With the BCP making multiple transformations of the representational state via multiple media, the system is provided with multiple opportunities to identify and recover from errors. However, it also provides more inputs of potential error and greater time consumption. Here we see the trade-off of efficiency for reliability. The system is reliant on reliable solutions of the field gunnery problem so as not to cause any wastage or even worse, incidents of fratricide. With this conscious focus on reliability comes a division of labour which simplifies roles to repetitive and relatively straightforward tasks; a principle first put forward by Adam Smith in his 1776 treatise, The Wealth of Nations. Within the job design literature this is termed ‘job simplification’ (Parker & Wall, 1998), and is associated with ‘Taylorism’, a term that characterises rigid, clearly divided production line-type work.
Dividing work within a system to maximise individual efficiency and reliability are known within industry to create low levels of job satisfaction (Parker & Wall, 1998). Looking at the present system, information travels through each element via a specific signaller, which is then observed or overheard by another actor who then proceeds to further propagate that representation. As previously alluded to, everyone’s role is rigid and predefined to offer the most efficient command and control architecture for the system. The trade-off is one of job satisfaction for efficiency. This is a common dilemma often researched in production lines around the globe. Given the criticality of the actors’ roles within this particular system, efficiency is a very high priority and yet is not being addressed at the most important element, the BCP.
The process of information filtering and buffering is an essential property in a system’s fluid operation. Figure 1 illustrates the main flow properties of the system, including the most frequent communication links (represented by dashed arrows). In Figure 1, the JOSTs act as filtering elements, deciding on which targets to engage and what information to propagate through the system. Both the JOSCC and RCP manage Regimental fire missions. The JOSCC is the central authority of the system and can decide to fully control fires, or delegate authority to other elements. Most commonly, the JOSCC and RCP will monitor requests on the regimental command net for any regimental fire missions, in which case the RCP takes over. Additionally, the JOSTs may liaise with JOSCC for permission to fire on critical targets.
Figure 1: The main flow properties of the system
The BCP is the only point where information is held and buffered from proceeding down the line. This is because key transformations need to be made to the representational states so that orders can be passed to the gun line in an intelligible form. Ultimately, the gunners are blind to any preceding representational states, with only the BCPs being able to pass down orders to the gun line. Without the BCP, the gun line would be bombarded by fire missions, which would eventually queue and clog the system. Consequently, the BCP is also the most saturated element of the system in terms of cognitive workload.
3.2 The Artefact
In this section we describe the design of individual artefacts which support the work of an Artillery Regiment. We pay particular attention to those artefacts which are central to the system’s performance and which are used to assist in coordination of complex tasks. Our focus is on how these artefacts empower and shape cognition at either an element or individual actor level.
One of the most prominent artefacts observed was the map. Its use provided the actors a visual representation of the problem, and a common picture of understanding. The map also acts as a redundancy measure for the system in case of any computer failures. By transforming representational states to locations on a paper map, the actors were tracing changes and reducing memory load by using a form of external cognition. Representational states could thus be encoded and stored within the map artefact, allowing actors to build a shared understanding of the state of the problem on a visual medium, and allowing regimental elements to transform and propagate partially encoded solutions.
The logbook was another frequently used artefact which assisted actors in propagating representational states. The logbook was also a visual medium, but served to assist actors in a different context. Most commonly the logbook was used to record radio transmissions such as target information for a FM BTY. Here again we see a form of external cognition, with the logbook acting as a memory aid and verification tool to help maximise error recovery. Logbooks also serve as a very useful tool and official record for retrospective analysis of firing errors, assuming entries were recorded correctly. This backing–up of representational states provides a secondary checking measure, but can just as easily allow propagation of corrupt representational states. Writing down the wrong grid reference for example can corrupt the rest of the representational chain, if not corrected at the source.
Another key mediating artefact which was essential in the BCP’s coordinating role, was the gunnery computer. This artefact, like that of the map, demonstrates the use of external cognition to assist actors in complex tasks. Rather than manually converting targeting data (grid reference for example) into predicted data (bearing and elevation for the guns), the system uses an artefact to simplify cognitive effort. Each partial solution or representational state with the OPCP is first encoded within a computer and then transformed outside the human mind. Again we see cognition as an interaction between actor, artefact and activity.
All artefacts can be seen as repositories of partially encoded solutions, which propagate through the system. Whether it is a radio, a map or computer, partial solutions are often stored outside the human mind. By storing partial solutions within artefacts we become reliant on those artefacts for solving problems. For repetitive and well-defined problems such as the field gunnery problem, this reliance is justified given its complexity. If however, all partial solutions or a significant proportion of them are embedded within artefacts, the system looses its resilience and becomes only as effective as its artefacts allow.
- Discussion and Conclusion
4.1 Digitisation and Automation
Under Project LAND 17, Australia has chosen the Advanced Field Artillery Tactical Data System (AFATDS) to act as their OS command and control system. AFATDS is known to feature digital tactical pictures, collaborative planning and fire mission execution facilities. Similarly, under Land 146, a digitised targeting system will be procured to aid in the conversion of targeting data by the FO. This would provide a representational state readily interpretable down at the guns. It is currently unknown however, as to how the Royal Australian Artillery will employ AFATDS on the battlefield. What is known however is the general functionality of the software and its ability to automate some of the process behind solving the field gunnery problem. Here we will discuss the impact of AFATDS and digitisation in respect to this automation.
Wickens (1984) suggested that automated technologies are usually developed for 3 reasons. One, to perform functions that human can not (complex calculations for example). Two, to perform functions which humans are capable of performing but only at the cost of high workloads. And three, to provide assistance in areas where humans exhibit limitations. By decentralising some of the target engagement logic within the field gunnery problem, AFATDS’ design covers all three reasons and subsequently allows actors to handle a larger number of targets. Traditionally however, the introduction of automated technology within a system also alters the roles played by key actors. Quite often actors go from traditional operators to supervisory controllers (Hawley, Mares & Giammanco, 2005; Parasuraman, Sheridan & Wickens, 2000).
As it stands, the artillery system is Taylorist in nature, and has clearly divided roles with relatively well-defined tasks. The system is arranged for dealing with threats in a repetitive and serial manner. With the introduction of AFATDS, actors will move from operating against single threats in time, to managing a resource pool in order to defeat a complex enemy force. Hopkin (1992) classed this shift in work as a move from a tactical to a strategic orientation. This strategic orientation may be difficult for some actors to adapt to given the change in nature of their work. This change may affect multiple elements of the system. For example, the FO can currently determine the ammunition and distribution of fire (or spread) using his contextual knowledge based on his direct observations. AFATDS however provides such a judgment based on inputs provided by sensors including the FO. Though the presence of this functionality within AFATDS does not force the system to use it, the FO’s role can quite easily revert to one of a sensor with little participation in the decision making process. A similar change may be observed in the gun line. With the possible purchase of modern self-propelled howitzers, the gun numbers will no longer be required to manually orient the gun or even to load it. They will become less coupled from the gun itself, exercising less direct control, and become a crew who exercise control from a computer intermediary. Thus, involvement within the system becomes one of a monitor and controller of a computer, which in turn completes the controlling tasks.
Role changes are not the only implications of introducing automation. Other issues can surface which affect the overall performance of the system. The most cited are automation bias, the phenomenon of operators becoming complacent with computer generated solutions, failing to question them; and job simplification, where operators go from active participants in a system to passive monitoring agents of a computer.
Automation has been used to support decision making in various areas. These most commonly include information acquisition, prioritization, information analysis, decision selection and action implementation (Parasuraman et al., 2000; Wright & Kaber, 2005). Skitka, Mosier and Burdick (1999) investigated an automated system which provides decision recommendations for pilots. Using a monitoring task within a flight simulation they found that participants in the non-automated setting outperformed their colleagues who were given a very reliable decision aid. Participants with the aid were found to make 2 key categories of errors. Errors of omission concerned events where participants missed key events which were not prompted by the automated aid, and errors of commission concerned events where participants did what the aid recommended even when it contradicted their training and other flight indicators. These results have been replicated with computer assisted route planning (Chen & Pritchett, 2001; Johnson, Ren, Kuchar & Oman, 2002; Layton, Smith & McCoy, 1994) and resource allocation optimisation (Cummings, 2003). Other studies however have found that automation can be an effective performance enhancer when reliability is extremely high and when applied to limited aspects of a task (Metzger & Parasuraman, 2005; Wright & Kaber, 2005).
We suggest that introducing a system which generates decision logic to the extent which AFATDS does, may exhibit the same phenomenon of automation bias within artillery system actors. Additionally, applied research has indicated that teamwork training and explicit automation bias training do not produce any significant improvements (Mosier, Skitka, Dunbar & McDonnell, 2001). Having AFATDS producing target judgments and resource allocation decisions may be efficient, but comes at the cost of heightened risk with the removal of the human in the control process. There may however be an optimum level or application of automation by AFATDS within the artillery system.
The Taylorist nature of the artillery system is not unique within the military. Most of the military is designed under the principle of simplicity and reliability, which permits a more predictable work pattern in high stress and fatigue environments. Designing work to be quick and simple allows systems to operate efficiently while diminishing the chances for error. By introducing automation however there may be a further simplification of work and a simultaneous increase in cognitive demands. This may ultimately affect job satisfaction and even performance (Parker & Wall, 1998).
If we assume the introduction of AFATDS will require monitoring roles, we must also assume that there will be an increased requirement of attentional demands. Controlling any system via a computer screen demands a high level of vigilance (Van Cott, 1985), which fundamentally alters the balance between the physical and cognitive demands of work. The introduction of AFATDS not only affects the tasks to be completed by the operators, but may also decrease job autonomy and control by further restricting discretion (Klein, 1989). Currently the system allows some sort of change of work procedures at the actor’s discretion. The FO for example may decide the priority of a mission, the method of engagement and whether to control the firing order at his command or allow the GPO to control firing. Alternatively, he may decide to leave these decisions for the BCP. There is some form of control and autonomy in the process. AFATDS however may require set criteria of inputs by the FO, which do not change at any situation. The FO would thus be forced to input the same variables for all fire missions, restricting his discretion within the system’s process. A more efficient and goal oriented representational state is created with AFATDS, but at the cost of actor discretion.
One must also be cognisant of the potential for a significant increase in what the job design literature call “production responsibility” (Parker & Wall, 1998). Production responsibility is the potential costs associated with performing an error within the system. Currently the system has checks and more critically, multiple representational state transformations at which an actor could pick up a potential error. With an automated system simplifying the work of all Regimental elements, it may be easier to propagate an error through the system at far greater speeds, resulting in less points of verification. Thus, the system gains a performance benefit in terms of efficiency, but may lose some error visibility due to the automation making state propagation incredibly simple.
We suggest that one of the most promising aspects of AFATDS within the current system will be its role in improving performance visibility. Earlier in the paper we mentioned how the gun line have very little visibility of much of the artillery system’s processing. Currently, the gun numbers will only find out the results of a fire mission if the GPO passes down this information from the FO. This may arrive with a significant delay or alternatively not at all if the BCP is busy with other missions. With an OS battle management system and modern self-propelled howitzers however, the gun numbers will sit within the gun in front of computer screens presenting live tactical maps. This not only affords the gun numbers visibility of the results of their actions, but also allows them to become involved within the same situational picture as the rest of the artillery system. Battle damage assessment will unlikely be real-time, and may still rely on the FO’s discretion, but with digitisation will provide greater visibility of battlefield performance.
Under the planned changes, Gunner Jones will still have little visibility as to what the other regimental elements of the system are doing. His work, just like everyone else’s within the system will involve cognition within the head and outside, in the world. The system will still require human involvement and representational states will still require transformation and propagation. By creating an enriched and automation friendly system, Gunner Jones and his colleagues of the Regiment may increase their levels of efficiency and effectiveness in solving the field gunnery problem.
Hutchins, E. (1995). Cognition in the Wild. Cambridge, MA: MIT Press.
Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics. Part A: Systems and Humans, 30, 286-297.
Preece, J., Rogers, Y., & Sharp, H. (2002). Interaction design: beyond human-computer interaction. New York, NY: J. Wiley & Sons.
Perry, M. (2003) Distributed cognition. In J. M. Carroll (Ed.) HCI Models, Theories and Frameworks. Towards a Multidisciplinary Science. London: Morgan Kaufmann.
Parker, S. K. and Wall, T. D. (1998). Job and Work Design: Organizing Work To Promote Well-Being and Effectiveness. California: Sage Publications.
Hawley, J. K., Mares, A. L., & Giammanco, C. A. (2005). The Human Side of Automation: Lessons for Air Defense Command and Control (ARL-TR-3468). Army Research Laboratory.
Hopkin, V. D. (1992). Human factors issues in air traffic control. Human Factors Society Bulletin, 35(6). 1-4.
Skitka, L. J., Mosier, K. L., & Burdick, M. (1999). Does automation bias decision-making? International Journal of Human-Computer Studies, 51, 991-1006.
Chen, T. L., & Pritchett, A. R. (2001). Cockpit decision aids for emergency flight planning. Journal of Aircraft, 38, 935-943.
Johnson, K., Ren, L., Kuchar J., & Oman, C. M. (2002). Interaction of automation and time pressure in a route replanning task. International Conference on Human-Computer Interaction in Aeronautics (HCI-Aero2002), Cambridge, MA, American Association for Artificial Intelligence.
Layton, C., Smith, P. J., & McCoy, C. E. (1994). Design of a cooperative problem-solving system for en-route flight planning: An empirical evaluation. Human Factors, 36, 94-119.
Cummings, M. L. (2003). Designing decision support systems for revolutionary command and control domains. PhD Dissertation, Systems and Information Engineering, University of Virginia, Charlottesville.
Metzger, U., & Parasuraman, R. (2005). Automation in future air traffic management: Effects of decision aid reliability on controller performance and mental workload. Human Factors, 47(1), 35-49.
Mosier, K. L., Skitka, L. J., Dunbar, M., & McDonnell, L. (2001). Aircrews and Automation Bias: The Advantages of Teamwork? International Journal of Aviation Psychology, 11(1), 1-14.
Van Cott, H. P. (1985). High technology and human needs. Ergonomics, 28 (8), 1135-1142.
Klein, (1989, March-April). The human cost of manufacturing reform. Harvard Business Review, 44, 61.
Wright, M. C., & Kaber, D. B. (2005). Effects of automation of information processing functions on teamwork. Human Factors, 47(1), 50-66.
Cortexia: Leading Human Factors Consultancy Australia