We have argued previously (Johnston and Milton, 2002b) that existing information systems implicitly support the deliberative theory of agency. According to this theory (Johnston and Brennan, 1996), purposeful action proceeds by an agent building an abstract model or representation of the external objective world from sense data and then reasoning about this model to determine actions that will achieve goals. For example, in traditional transaction-based information systems, ‘transactions’ are gathered that represent changes in the world. Data models that correspond to the representation scheme are used to design operational databases that are affected by the transactions. In extreme cases, such as MRPII (Wight, 1981), application programs also deduce goal-attaining actions and human actors are only required to define the goal state, execute the actions in reality by following automatically generated schedules and provide sense data by recording transactions. More typically, applications programs help human actors to make decisions by providing information about objects from reality using data gained through transactions. Decision support systems are good examples of this type of system.

In the past 30 years a number of methodologies have been developed to assist in designing such systems. These are often called information engineering methodologies (IEM), with Structured Systems Analysis and Design Methodology (SSADM), the British government standard, being a typical example. These design methodologies share the ontological assumptions of the deliberative theory, namely, that systems should represent the world in which the system acts in terms of external, independent and objective entities, properties and relations (Wand et al., 1995). Given this focus on symbol/object representation, use of these methodologies encourages designs for socio-technical systems in which the information systems form the representational scheme which mimic the deliberative approach to agency.

On the other hand, disciplines other than information systems have considered an alternative approach called the situational theory of agency. In robotics, specifically, this alternative theory has been motivated by the brittle performance and computation intensity of artefacts based on the deliberative approach. The key to this alternative theory is to provide an agent with largely reactive responses based on sense data obtainable directly from the agent’s ground view of the world, and to introduce the agent’s goals and perspective explicitly in the representation schemes implicit in the theory. In the situational theory, agents respond reactively to ‘situations’ without deliberation. Situations are descriptions of the world centred on the agent and only include features of the world that relate to the agent’s purposes (Agre and Chapman, 1987). These features consist of the relations of things to the agent given its goals. Actions are selected from a repertoire used to respond to situations. This approach to action selection leads to goal attainment only if the agent’s environment exhibits structure (‘affordances’) that obviate the need to plan (Agre and Chapman 1987). An affordance is a structural aspect of the environment that makes it possible for an agent to reach a desired situation by merely reacting to its current situation. Analysis and exploitation of environmental structure is an important part of designing situated agents (Agre and Horswill, 1992; Hammond et al., 1995; Horswill, 1995; Agre and Horswill, 1997). An activity in this theory is a grouping of situations and associated actions that together lead to a reliable reaching of a desirable situation.

We can see the differing roles of representation in the situational theory. Situations are agent-centred and intention-laden. Representation of situations on the basis of a symbol/object isomorphism is neither possible nor necessary. An agent responds to being in a situation by taking an action. An agent needs to notice that it is in a situation and does so by sensing aspects of its environment. Consequently, aspects of situations are needed to fire situation-action responses. Agre and Chapman (1987) argue that the representational scheme is ‘indexical’ and ‘functional’ in nature. Indexical representations describe things relative to the agent and functional representations select things according to their relevance for the purposes of the agent or concern the activities in which the agent is engaged. Further, Agre and Chapman (1987) argue that to eliminate the computational complexity of action selection inherent in using aerial world models, indexical/functional representations of situation features that are relevant to the agent’s goals are rebound ‘on the fly’.

The reliance of the situational theory on indexical/functional rather than symbol/object representation shows it is built over different ontological categories: situations, aspects of situations, actions, activities (groups of situation/action pairs), environmental structure, and environmental affordances.

To illustrate the situational theory and how it differs from the deliberative theory, consider a rat searching for food in a connected maze without cycles (Figure 14.1, “The two views of ‘Rat World’.” (a)). The rat could:

The first requires the rat to gather and hold a representation of the maze, as viewed from above and which includes all objects in the maze, before deducing a plan of action that is to be effected by it. The aerial view in Figure 14.1, “The two views of ‘Rat World’.” (a) shows this (deliberative) view. The second is a situational approach, shown in the ground view in Figure 14.1, “The two views of ‘Rat World’.” (b), where the rat notices a limited range of situations relevant to its activity and of which it becomes aware by sensing these aspects – the absence and presence of walls. To act, the rat only needs to be aware of the absence and presence of walls near it, and it is not interested in anything else in its environment. In this way representation is purely indexical (centred on the rat) and functional (for its acting).

In the ‘ground view’ there are three aspects, numbered 1, 2 and 3. These three aspects completely determine the situation the rat is in, at least in relation to its seeking food. The rat will then select the action appropriate for the situation. All possible situations and their associated responses can be grouped into an activity (called ‘seeking food in a maze’).

As stated earlier, myopic-situated actions rely on environmental affordances for their efficacy. In this example, the environmental affordance is that the maze is singly connected and does not have cycles. It is the existence of this structural property of the maze that ensures that if the rat invokes the activity it will reliably reach food. This maze-navigating example illustrates the general point made by advocates of the situational theory (Agre and Horswill, 1992) and ecological theories of behaviour (Gibson, 1977; Schoggen, 1989) that environmental structures, or affordances, make a significant contribution to the production of goal-directed behaviour of real agents in real environments. As Agre and Horswill (1992) put it: ‘it is almost as if these surroundings were an extension of one’s mind’.

There are three ways in which a situational system is brought into being. First, a situational system could evolve so that agent actions and the effects of actions knit perfectly with the environment and situations to make activities reliable. Biological organisms are excellent examples of evolved situational systems. In many cases, such as social activities, the activity and its environment may have co-evolved. Second, an agent may learn an activity by seeing the effect of actions in specific situations. In this case, trial and error is used to find the action rules that best exploit the structures in the environment, but also environments might be chosen because of their particular affordances for action. Third, and this is the approach we propose for information systems, a system can be designed so that actions in response to situations have desired effects. Depending on constraints, either or both the action rules and the environment structures will be deliberately designed to ensure the reliability of an activity. It is for this purpose that we propose our methodology, and it is a distinctive feature that ‘environmental engineering’ is part of it. We assume that some level of iteration may be needed.

When an agent is confronted with an unknown situation, or when an existing activity is not reliable in an environment, there are at least three ways for the agent to respond. The most extreme response is to deliberate from first principles, much like the deliberative theory of agency. According to Heidegger, in his analysis of ‘breakdown’ (Dreyfus, 1991), an agent will resort to an ascending hierarchy of situated practices of repair before resorting to pure deliberation. For instance, an agent might first engage in another activity that is closely related to the failed routine activity but suited to a slightly different environment. An example of this would be using a different maze-solving routine. Alternatively, the agent could reason about the activities themselves without necessarily building a complete external world-model, which would amount to invoking a routine of problem solving.

We have used the situational theory of agency, as it is discussed in robotics and other disciplines, to determine the concepts central to an agent-centred situational system. Whereas the deliberative theory suggests information systems design should emphasise modelling the world using objects, properties, relations and states, and deduction upon these models to determine action (such as decision support and planning), the situational theory would make central the notions of an activity, situations that comprise activities, actions that are a reaction to situations, and aspects that allow situations to be detected. Also the situational theory would emphasise the importance of proper structuring of environments of action, which is largely ignored in the deliberative approach. Thus, a methodology for designing situational systems must:

The final methodology would consist of detailed documented guidelines for performing these steps together with appropriate representational analytical tools.

Thus far, the only experience in analysing situational systems in the literature is for designing robots and software agents. It has not been explicitly applied to socio-technical systems. However, there exist evolved routine work systems involving human actors and we are interested in examining their workings with respect to each of the theories of agency and drawing conclusions about how they work based on the examination. Should they be found to be consistent with the situational theory of agency, we may add depth to our understanding of the characteristics of situational systems. Consequently, in the following section we discuss three cases of evolved manual systems supporting routines. Manual systems are used in this paper because they are examples of effective situational systems. The design methodology for situational information systems assumes that agent environments, sensing mechanisms for situations, and actions may need to be designed from scratch and are thus ‘blue sky’. Fine-tuning will be required where any unreliable activities are found: essentially ‘tweaking’ them to make them more effective. Fine-tuning will involve further moulding of the environment and improvement in sensing situations.