Available online at www.sciencedirect.com

ScienceDirect Cognitive Systems Research xxx (2015) xxx–xxx www.elsevier.com/locate/cogsys

Social cognition, artefacts, and stigmergy revisited: Concepts of coordination Tarja Susi School of Informatics, University of Sko¨vde, Sweden

Abstract A number of different coordination concepts have been developed to explain how individual activities are coordinated on a social level, and the variety of concepts shows there is an interest in many domains to find such explanations. Stigmergy being one of them, has come to be increasingly applied on various kinds of human activities. In other domains we find other concepts for explaining how environmental resources contribute to work activities or how people use them to structure their work. This paper discusses different coordination concepts, including stigmergy, articulation work, coordination mechanisms, triggers, placeholders, and entry points. The first three concepts are explicitly concerned with coordination among several agents, while the last three instead concern individual activities, but arguably they can be extended to the social level. They also bring an explicitly cognitive dimension to coordination, which is not as salient in the former concepts. The concepts discussed here do have some similarities, but also important differences. They may not be interchangeable, but they could complement each other, or contribute to further elaboration of existing concepts. The stigmergic sign, e.g., could usefully be developed to recognise qualitative differences in its role as a coordination mechanism. Ó 2015 Elsevier B.V. All rights reserved.

Keywords: Stigmergy; Articulation work; Coordination mechanisms; Triggers; Placeholders; Entry points

1. Introduction Some years ago Susi and Ziemke (2001) made a comparative analysis of social/situated theories of cognition (activity theory, situated action, and distributed cognition) and stigmergy, discussing the coordination paradox, visible in both social insects and human activities. The key elements compared were agents, environment, and artefacts, and the collaborative activity emerging from their interaction. The conclusion made back then was that the considered theories and stigmergy do have some similarities, and that stigmergy could be a common denominator. However, while stigmergy, as a general principle, provides a valuable explanation to the coordination paradox (Theraulaz & E-mail address: [email protected]

Bonabeau, 1999), it may be limited in providing rich descriptions of how seemingly individual activities sum up to coordinated human activity. Hence, the aim of this paper is to discuss some concepts of coordination originating from different areas, and to relate them to stigmergy, which are discussed in the context of a case study at a workplace where most tasks are distributed and individually performed, but result in well organised and coordinated activities at the collective level. Regarding the issue of defining stigmergy, many definitions are considered vague and too general and many papers on stigmergy are criticised for leading their readers to ‘‘believe that it is a simple phenomenon that can be easily dismissed as an environment mediated, and indirect communication mechanism” (Dipple, Raymond, & Docherty, 2014, p. 90; see also, e.g., Heylighen, 2011–12;

http://dx.doi.org/10.1016/j.cogsys.2015.12.006 1389-0417/Ó 2015 Elsevier B.V. All rights reserved.

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

2

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx

Huang, Ren, & Jin, 2008). However, considering, e.g., Dipple et al.’s (2014) general theory of stigmergy, the concept of stigmergy is anything but simple; based on a thorough review of previous research, and by addressing the ‘‘what, how, why, where and when” of stigmergy, they constructed a holistic, macro-level general model thereof. Another example is Huang et al. (2008) who constructed a sign-based model of stigmergy (SBMS), a unified model that places ‘‘sign” at its core. The view on stigmergy adopted here is based on these two theories. Dipple et al. (2014) consider different concepts of stigmergy, which includes mechanisms (qualitative/quantitative) distinctions (marker-based/sematectonic), and varieties (four possible combinations of qualitative/quantitative and markerbased/sematectonic). The variety considered here is qualitative marker-based, according to which ‘‘the intentional marker is a sign left as a signal that means something to others in its single form [. . .] that facilitates coordination” (Dipple et al., 2014, p. 13). Huang et al. (2008) provide an elaborated description of sign, which is considered to have a content that is an agent’s behaviour or the product of an agent’s activities, is carried by the environment, and can be sensed or observed, and interpreted by other agents. Importantly, these theories consider stigmergy in human activities, which allows us to include a cognitive perspective on agents. In Susi and Ziemke (2001) it was argued that despite the difference in complexity between social insects and humans, the principles of coordination by use of artefacts can be applied to human activity in order to explain and understand the coordination paradox. Whether stigmergy in fact is applicable on human activity or not, has already been discussed in many papers (e.g., Christensen, 2013; Dipple et al., 2014; Parunak, 2006) and the issue will not be reiterated here. Instead it is assumed that a lot of human activities are stigmergic. After all, as Parunak (2006) states, stigmergy is ubiquitous in human interactions and it ‘‘would be more difficult to show a functioning human institution that is not stigmergic, than it is to find examples of human stigmergy” (ibid.). While stigmergy previously has attracted perhaps most interest in areas like Artificial Life (Bonabeau, 1999) and Multi-Agent Systems (Omicini, Ricci, & Viroli, 2008), it has also been applied on animals with known cognitive capacities and it is argued to provide a powerful metaphor for human interactions (Marsh & Onof, 2008; Parunak, 2006). Furthermore, to overcome the limitation of ant-like stigmergy, and to include cognizant agents, the concept of stigmergy has also been expanded to explicitly encompass a cognitively oriented perspective, as seen, e.g., in frameworks on stigmergic cognition (Marsh & Onof, 2008) and cognitive stigmergy (Lewis, 2013; Ricci, Omicini, Viroli, Gardelli, & Oliva, 2007). Cognitive stigmergy is said to preserve the benefits of ‘‘ant-biased” stigmergy of the multi-agent systems field, but also to ‘‘promote the full exploitation of the cognitive abilities of agents and of the environment articulation in

artifacts in the stigmergic process” (Ricci et al., 2007, p. 138). Another example is Parunak’s (2006) schema for analysis of human–human stigmergy, which includes four varieties of stigmergy based on a binary distinction between marker-based/sematectonic stigmergy, and qualitative/quantitative stigmergy. The schema is then applied to a number of common human activities which demonstrates their stigmergic nature. Stigmergy is also being increasingly applied in various domains concerned with human activities. Some examples are team cognition (Espinosa, Lerch, & Kraut, 2002), team work practices (Christensen, 2013), embodied cognition (Dawson, 2014), information systems and open source software development (Bolici, Howison, & Crowston, 2009; Howison, Østerlund, Crowston, & Bolici, 2012; Marsden, 2013), online creative communities (Secretan, 2013), and systems security (Lugosky & Dove, 2011). There is vast number of studies on collaboration, cooperation, and coordination of human activities, but few of them have embraced stigmergy as an explanatory concept. Besides stigmergy there are also other concepts and mechanisms for explaining coordination of human activities, but comparisons are scarce. An exception is Christensen (2013) who compared stigmergy, articulation work, awareness, and feedthrough. In his conclusion none of the mentioned concepts are interchangeable with stigmergy, but that they complement each other. Another analysis was made by Bolici et al. (2009), who compared stigmergy, boundary objects, field of work, trading zones, and community of practice. Some of the concepts are similar or clearly related to each other, but as noted by Howison et al. (2012), ‘‘they don’t take each other’s contribution into account” (referring to implicit coordination and stigmergy). The focus of this paper is stigmergy and a few concepts that are, or can be, related to coordination in human activities, used foremost in cognitive science, human–computer interaction, and computer-supported cooperative work. The concepts discussed here are articulation work and coordination mechanisms (Schmidt & Simone, 1996), triggers and placeholders (Dix, Ramduny-Ellis, & Wilkinson, 2004, and entry points (Kirsh, 2001) (concepts like boundary objects and trading zones, which were compared by Bolici et al., 2009, are not included since they concern structures for coordination between different communities, while the discussion here considers coordination within a community). These concepts do not seem to have been previously compared and related to stigmergy, although three of the concepts (triggers, placeholders, entry points) have been combined as a means for understanding the role of artefacts in social interactions (Susi, 2005). Some of the concepts were originally formulated with regard to individual activities, but the discussion will consider their possible role in coordinating activities on a collective level. For the sake of clarity, the term coordination mechanism, as used in the field of computer-supported cooperative work, will be denoted ‘‘coordination mechanisms (CMs)” to distinguish it from stigmergic coordination mechanisms.

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx

The rest of this paper is organised as follows. The next section describes a case study in a work place, which is then used throughout the paper to exemplify different concepts, and discusses articulation work and coordination mechanisms. Section three discusses the role of environmental resources in coordination, as seen in the concepts of triggers, placeholders, and entry points. The discussion in these sections is also related to stigmergy. Section four provides an overview of the coordination concepts discussed in previous sections and considers whether the work place indeed is a stigmergic systems. The last section provides a summary and conclusions. 2. Workplace coordination Before going into the details of concepts of coordination, this sections will start with a short description of a case study which will then be used to exemplify concepts throughout the rest of the paper. The case study was carried out in the control room of a middle-sized Swedish grain silo, which is a collaborative workplace with a high degree of complexity in operating the plant, and in managing grain deliveries (Susi, 2006). Although the study was carried out quite a few years ago, it very well serves as an example when discussing the current concepts in relation to coordination of real-world activities in a work place (the study is described in detail in Susi, 2006, and in short in Susi, 2005. A few of the concepts discussed here were included in the study, but they are elaborated here from a different perspective). While the workplace holds a vast number of artefacts that are vital for coordination, only one of them will be considered here, a control panel that is central to individual as well as cooperative activities and their coordination, as they are planned and unfold. Work in the control room is carried out in two or three shifts with two-three people working during each shift (seven people were employed at the time, with experiences ranging from newcomer to 26 yrs). Most activities are performed individually, but much of them are coordinated

3

through artefacts. The perhaps most important artefact in the control room is the control panel through which basically the whole plant is operated. The control panel itself is relatively easy to learn to operate. For instance, to start or stop a conveying process consists of pushing a sequence of some 10–15 buttons (depending on the state of the machinery), which can be traced by following symbols and indicators on the panel. The hard parts are to get a grip on the relation between the plant’s machinery that is represented on the control panel, and the actual underlying machinery, and whether the machinery is working properly when buttons on the control panel are pushed. Another hard part is the continuous planning ahead of work activities, and to know the best options for what grain to store in which silo; there are up to a hundred deliveries per day of up to more than 11 different types of grain, stored in 29 silos, and three transportation shafts for grain, so work has to be carefully planned and coordinated, both within and between shifts. Workers with really long experience know the plant ‘inside out’, and they also know how to plan ahead in order to not block transportation options at a later stage, which causes unwanted delays and costs. In other words, their knowledge of the plant and of best practices is crucial for coordination and fluency in the work processes. The control panel holds several representational layers (Fig. 1). There is one permanent representational layer of all the grain silos and the plant’s machinery, including transportation shafts and the sequences of actions for operating them (symbols, indicators, buttons, colours, etc.). In addition, the workers themselves have added some selfcreated representational layers on top of the permanent one. One of the self-created layers consists of small magnetic plaques with text such as, type of grain, ‘‘hatch out of order”, ‘‘open”, ‘‘closed”, ‘‘empty”, ‘‘full” and ‘‘half full”. Yet another layer consists of handwritten notes informing about, for instance, silos/grain that need to be monitored, on-going or pending repairs, and what is or will be stored in which silo. These layers are semi-permanent in

Fig. 1. The panel represents the whole plant, e.g., silos (grey circles and squares) and transportation shafts (vertical and horizontal lines). The excerpt to the right shows some of the workers’ self-made representations (magnetic plaques with text, handwritten notes). (ÓTarja Susi).

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

4

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx

that there is always a number of magnetic plaques and notes on the control panel at any given time, but the state of the layers changes during activities, and when activities come to an end and new ones start. Hence, the control panel holds an array of information readily available. The whole process – to receive a delivery of grain, to the grain being stored in a silo, or to send grain off – comprises a number of activities carried out in the control room, but also at other places within the plant. The control panel needs to be monitored and operated, grain has to be tested, there is paperwork to be done, and machinery may need to be checked to see whether it works properly. The activities are carried out individually, and may be performed by any worker (depending on their level of knowledge and experience), but they are often carried out in coordination between two or three workers. There are usually several conveying processes running in parallel, and a process may start during one shift and be completed by the next shift. While performing some activity outside the control room, the workers return every now and then, and look at the control panel to check the status of on-going processes or to see what needs doing, and they may change or add information on the panel. The control panel, then, is a crucial artefact in the workers’ ‘‘field of work” (Schmidt & Simone, 1996). Some activities that are planned are made visible on the control panel, which supports organising work ahead. For instance, when some part of the machinery needs maintenance, a note with a certain date may be put on the control panel, on the corresponding symbol for the machinery, indicating when the maintenance will take place. Making such activities visible is what Schmidt and Simone (1996) termed ‘‘articulation work” that supports organised cooperative work where individual but interdependent activities are coordinated, scheduled, integrated, etc. Christensen (2013), who compared articulation work and stigmergy, describes articulation work as a kind of meta or supra-type of work to articulate individual but mutually interdependent activities, i.e., ‘‘extra activities” aimed at coordinating work. Stigmergy on the other hand, he says, does not have this kind of supra relationship to work tasks being carried out. However, although articulation work refers to a specific kind of activity, and stigmergy refers to a coordination mechanism, the two can be closely interrelated. In this setting work can be articulated by a single person, since the workplace has a low degree of complexity with regard to the number of people working at the same time. The complexity in this setting instead lies in the work procedures and the number of activities that have to be properly planned, coordinated, and integrated. Other common examples of articulation work in the setting include handwritten notes on specific silo symbols informing that they are reserved for certain deliveries on certain dates; notes that inform about the level of moisture in the grain in a certain silo, since moisture levels above a certain percentage needs to be monitored, and; magnetic plaques like ‘‘wheat” and ‘‘empty” placed on a silo symbol, to inform that the silo is empty

and that it is reserved for wheat. Consequently, a lot of the articulation work is mediated through the control panel. As such, the control panel and the representational layers attached to it, are artefacts that function as coordination mechanisms (CMs). The concept originally developed as a generalisation of investigations of the use of artefacts for coordination purposes (Schmidt & Simone, 1996). As described by Schmidt and Simone (1996, pp. 165–166), a coordination mechanism is: a construct consisting of a coordinative protocol (an integrated set of procedures and conventions stipulating the articulation of interdependent distributed activities) [. . .] and [. . .] an artifact (a permanent symbolic construct) in which the protocol is objectified. In the current case, the control panel is a construct consisting of a coordinative protocol, which integrates procedures (i.e., how non-changeable procedures like some conveying processes must be performed) and conventions (best practices for what to do when options are available, based on the most experienced workers knowledge. An option for conveying deemed as non-effective can be crossed out with tape on the control panel). At the same time, the control panel is also an artefact, a permanent symbolic construct that objectifies the workplace procedures and conventions. Hence, the control panel functions both as a coordination mechanism and as a tool for articulation work that supports organisation of cooperative work. However, in the case of articulation work, it may also take place on different temporal scales, as exemplified by a machinery break-down; a break-down is usually signalled by an indicator light that turns on, which may cause a transportation shaft to be shut down. In such a case an on-going conveying process has to be redirected while the process is running, and work ahead may need replanning. As the redirection progresses, the information on the control panel changes as indicator lights turn on or off, and signs are moved, removed, or added, which also contributes to coordination of work tasks. This is a matter of articulation work on a short time-scale. At the same time, some of the changes constitute articulation work on a longer time-scale; for instance, a silo already reserved for a certain delivery on a given date may instead be needed right away to solve the situation caused by the break-down. An expected delivery may have to be put on hold and so the note with the reservation is moved to another silo symbol, and may also be given a new date. Hence, such incidents affect ongoing as well as future activities, which need to be coordinated on short and long term. Following Dipple et al. (2014) and Huang et al. (2008), the use of such notes is also a case of qualitative markerbased stigmergy where the sign (a note) is the product of an agent’s activity. When attached to the control panel it is an intentional sign carried by the environment, and it can be observed and interpreted by other agents. The sign means something in its single form and it supports coordi-

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx

nation. The indicator light, on the other hand, that alerted the break-down in the first place is not a sign on this account, since it is not produced by an agent. The common denominator in articulation work, coordination mechanisms (CMs), and stigmergy, should be the use of signs, or artefacts, for coordination, but there are important differences. The main difference between coordination mechanisms (CMs) and stigmergic mechanisms is their different levels of abstraction and a temporal dimension; coordination mechanisms (CMs) is a generalised concept that can include any coordinative protocols and artefacts used for coordinative purposes, while stigmergic mechanisms instead place focus on signs and their coordinative function. There is also a temporal aspect in articulation work, which may concern activities on varying time scales, whereas stigmergy lacks a distinct temporal aspect. Either there are stigmergic signs or not, and they may or may not be observed by others, but it really has no temporal aspect. To shed some further light on coordination through stigmergic mechanisms, some of the ‘‘core messages” found in Tummolini and Castelfranchis’ (2007) taxonomy of trace messages can be borrowed. However, in their definition stigmergy is ‘‘the process of indirect communication of behavioural messages with implicit signals” (ibid.), which only concerns traces implicitly used as signals for others, i.e., traces that are not marked for explicit communication. The definition then, is not suitable for the current purposes in that the messages considered here are left intentionally for explicit communicative purposes. The core messages are, hence, modified to denote explicit communication. There are four trace messages that apply to the control panel (the remaining three trace messages, ‘‘presence”, ‘‘intention”, and ‘‘ability”, are also observable in the control room, but not explicitly on the control panel):  Opportunity for action, i.e., a trace informing the conditions for an action; a note saying ‘‘out of order”, placed on a symbol for a hatch, informs about an obstacle, hence informing the conditions for further actions – what can or cannot be done.  Goal, i.e., a trace to let others understand someone’s goal; magnetic plaques saying ‘‘rye” and ‘‘empty” placed on the symbol of a certain silo informs that the goal is to fill the silo with rye.  Action accomplishment, i.e., a trace informing what someone has done; a worker who has conveyed rye to a previously empty silo informs that the task has been accomplished by placing a magnetic plaque for ‘‘rye” on the symbol for that silo.  Result, i.e., a trace informing about the result of an action; a worker who has conveyed rye to an empty silo informs the silo is no longer empty by placing a magnetic plaque for ”rye” on the symbol for that silo. As seen in the last two traces above, the same example applies to both as ‘‘different meanings can often co-exist

5

at the same time in the same trace”, and traces exploit ‘‘different interpretations of the behaviour by different observers” (Tummolini & Castelfranchi, 2007, p. 156). Likewise, different kinds of coordination mechanisms can co-exist at the same time in the same artefact; the control panel is both a coordinative mechanism for articulated work, and a stigmergic mechanism that mediates intentional, but indirect interactions through signs carried by the environment. Hence, the execution of some tasks and their coordination can be explained through different coordination concepts. Despite the seemingly similar coordinative functions though, the concepts discussed hitherto are not interchangeable (as already noted, e.g., by Christensen, 2013, in the case of articulation work and stigmergy). In many cases articulation work is an objectification of things ahead, while stigmergy is at play in the realisation of what has been planned in advance. Yet, depending on which perspective an artefact is viewed from, articulation work and stigmergic processes may be truly intertwined, and one and the same artefact can simultaneously be a coordination mechanism as well as a stigmergic mechanism.

3. Environmental resources and coordination Other concepts that clearly can be related to stigmergy and coordination are found in the field of human–computer interaction in the work of Dix, Ramduny, and Wilkinson (1998), Dix et al. (2004), which describes ways in which environmental resources contribute to work activities. Dix et al. (2004) propose trigger analysis as a way to analyse work activities, and to find out why things happen when they happen. It shows the ways external resources initiate activities, function as reminders of where in a work process people are, and ensure that tasks are carried out in full. Of interest here are the concepts of triggers and placeholders. A trigger is something that prompts an activity, something that tells you that you need to do something, while a placeholder is something that is stored and indicates what needs to be done. Two of the triggers identified by Dix et al. (1998, 2004) can be related to stigmergy, here provided with examples from the control panel (the remaining three triggers, ‘‘temporal”, ‘‘sporadic”, and ‘‘external event” are also observable in the control room, but not as intentional messages left on the control panel):  Immediate; takes place when an activity begins immediately after a previous task is completed, e.g., to convey wheat to a silo marked ‘‘wheat” and ‘‘empty”, when another silo also designated for wheat has become full, and is marked with ‘‘wheat” and ‘‘full”.  Environmental cue; explicitly coded cues in the environment that reminds that something needs to be done, e.g., a note saying ‘‘18%, 14/9” prompts the staff to continuously check the grain over the next few days, since the level of moisture is too high.

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

6

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx

Although triggers originally concern activities on an individual level, their role clearly can be related to a collective level as well since the triggers above are in fact intentional messages left on the control panel. They are also observable by any worker, and they contribute to coordination on a collective level. The function of these triggers can also be described as qualitative marker-based stigmergy in that they are intentional signs that are carried by the environment, and they are observable and they mean something to others. Basically both concepts can explain the mechanism underlying coordination, but they differ in that the former concept provides different dimensions to a sign, which allows for a differentiation between environmental modifications. Turning to placeholders (what needs to be done), they are the things that help people keep track of where in a process they are, reminding them of what to do next, thus making sure that tasks will be carried out. Placeholders are stored in different ways (Dix et al., 2004), and two of them, exemplified by the control panel, can be related to stigmergy (the last placeholder, ‘‘in people’s heads”, i.e., to remember what to do next, is not included here):  Explicitly in the environment; things that explicitly tell what needs doing, e.g., when hatches are under inspection, a note saying ‘‘check” next to a button representing a hatch on the control panel indicates which hatch is the next to be inspected.  Implicitly in the environment; things that implicitly tell what needs doing e.g., if wheat needs to be conveyed to a silo, a single magnetic plaque with ‘‘wheat” placed on a silo symbol is an implicit message that the silo has not been filled yet, and that is where wheat needs to go. Had the silo already, e.g., been filled there would have been two adjacent magnetic plaques, ‘‘wheat” and ‘‘full”. Originally placeholders concern individual activities, just like triggers, but they can also easily be extended to the collective level. The placeholders are perceptible to any worker, and they remind everyone of what needs to be done. As in the case of triggers, the function of the first placeholder above, can also be described as qualitative marker-based stigmergy. The second placeholder (‘‘implicit in the environment”) however, does not qualify as a stigmergic sign since there is no concrete single sign for that which is absent, i.e., a sign for ‘‘there is room for more wheat”. Instead, the implicit message is nested within the explicit sign ‘‘wheat” and its location on the control panel, and the interpretation of that message requires a wider understanding than what is explicated by the single sign ‘‘wheat”. However, implicit placeholders are crucial for knowing the state of processes and what needs to be done, and they significantly contribute to coordination of work tasks. Hence, the implicit placeholder provides a differentiation to mechanisms underlying coordination, but it is not a stigmergic mechanism. When considering an explicit

placeholder as a stigmergic sign, the concept of sign does not encompass temporal aspects, and the sign does not have to be part of any specific work process. A further complication with triggers and placeholders is that they can co-exist in one and the same artefact (like the previous coordination mechanisms and trace messages); an explicit placeholder, like the note saying ‘‘check”, can be both a trigger that prompts an action, and a placeholder that tells at what point an inspection process is. Dix, Ramduny, and Wilkinson (1998) pose the question what it is that makes someone notice a cue? From a stigmergic point of view one might say that we can perceive signs visible in the environment and hence, notice triggers and placeholders, and at least some of them clearly can be considered as stigmergic coordination mechanisms. Another explanation is found in a cognitive perspective on the ways people actively use environmental resources to structure their work, as discussed in Kirsh (1996, 2001). In Kirsh’s view, concepts like triggers and placeholders may be helpful for understanding work contexts, but they do not help us gain more insight into their deep structure. Instead, he argues, we need to move beyond such ‘‘surface structures”, and describe environmental elements in more abstract terms. For that end he developed the concept of entry points (Kirsh, 1996, 2001). The concept, which is influenced by Gibson (1986) ecological theory of visual perception, refers to structures or cues that invite people to do something, usually information or communication related things. In other words, an entry point is ‘‘a structure or cue that represents an invitation to enter an information space or office task” (Kirsh, 2001, p.305). Typical entry points are, e.g., calendars and to-do-lists. What people do is that they create collections of entry points that tell them what is going on, what needs to be taken care of the next day, and so on. Although such collections are personal and subjective, as preferences for the number and type of entry points differ, the entry point concept can well be expanded to a social level. An entry point made by one worker in the control room may invite someone else to act, and a collection of entry points is observable by all workers. Entry points have different properties that vary along a number of key dimensions, or characteristics, that affect the way people react to them (Kirsh, 2001). Some of the entry point properties that can be related to stigmergy, again exemplified by the control room, include:  Intrusiveness, i.e., how much attention an entry point attracts through colour, shape, etc. (determines the probability that the entry point will be approached); hand-written notes on the control panel differ in shape, colour, and size from other messages and symbols, and are likely to attract enough attention to be approached.  Richness in metadata, i.e., the amount of underlying information conveyed by the entry point (the more metadata it holds, the less needs to be kept in memory); a note saying ‘‘18%, 14/9” conveys the actual moisture

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx

level of grain in a silo, the date the reading was obtained, and the need for further monitoring of the moisture level.  Visibility, i.e., how distinct or unobstructed the entry point is (the higher the visibility, the higher the chance that the entry point will be used): a hand-written note on white paper is highly visible since the colour differs from other objects on the control panel, and it is likely to attract enough attention to be approached, so the chance of it being used is high.  Relevance, i.e., how useful an entry point is to a current activity (the more relevant the entry point is, the higher the probability it will be used); a worker is preparing for a delivery of a grain and needs to know where it should be stored, so s/he checks the magnetic plaques on the control panel. The plaques are very useful to the activity so they are likely to be used. These entry points describe characteristics of different artefacts, in this case the messages or signs that are observable on the control panel, and their functions are directly related to people’s cognitive processes. Entry points and the previously discussed triggers and placeholders, are overlapping concepts in that they all prompt actions, but there are important differences; triggers prompt a person to do something, placeholders indicate what needs doing, and entry points can explain why people take notice of triggers and placeholders. Although these concepts were developed within different domains, they complement each other, and by also extending them to a social level, they provide different dimensions to the mechanisms that facilitate coordination of human activities, from a process and characteristics point of view. Most of these concepts can also be described in terms of stigmergic coordinative mechanisms, although with less differentiation. 4. Coordination concepts and stigmergic systems As an overview of the concepts discussed in previous sections, we will take another look at the control panel and the different concepts, in the context of ‘‘silo 10” being reserved for wheat. A delivery of wheat arrives, and the worker handling the delivery checks the state of the control panel and observes the signs ‘‘wheat” and ‘‘half full” placed on the representation for silo number 10. What the different concepts can explain is the following:  Articulation work. At some earlier point in time, it was planned ahead that silo 10 would be used for storing wheat, which was made visible through the sign ‘‘wheat”, placed on the symbol for silo 10. At a later point in time, the silo has become half full, which has been made visible through the sign ‘‘half full” placed next to the sign ‘‘wheat”, which indicates the state of filling silo 10 with wheat. The articulated work organises and supports coordination of individually performed activities.

7

 Coordination mechanisms (CMs). The control panel integrates procedures, how conveying the grain has to be performed (procedures that are not changeable), and conventions, the best practices for what to do when options are available (conveying the grain may require a route including more than one shaft, and some options can be ruled out by signs on the control panel). The control panel is also an artefact, a permanent symbolic construct that objectifies the procedures and conventions. Procedures and conventions are crystallised and objectified in the different representational layers, so ‘‘wheat” and ‘‘half full” mediates articulated work.  Trigger. When checking the state of the control panel, the worker observes the signs ‘‘wheat” and ‘‘half full”, and their location on ‘‘silo 10”. The triggers indicate what needs to be done, which prompts the activity of conveying the grain to that particular silo. The trigger is an ‘environmental cue’ that reminds where the wheat is to be stored.  Place holder. ‘‘Wheat” and ‘‘half full” reminds of where in the process they are of filling silo 10 with wheat. The placeholder is of the kind ‘explicitly in the environment’.  Entry point. ‘‘Wheat” and ‘‘half full” is a cue that invites the worker to convey the grain to that particular silo. The most salient property of the entry point is ‘relevance’, i.e., the cue is highly useful for the current activity.  Qualitative marker-based stigmergy. ‘‘Wheat” and ‘‘half full” is a sign (or rather two signs combined with purpose into one sign) that has been intentionally left as a signal, and it means something to the other workers. The content of the sign is the product of a worker’s activity and it is carried by the environment. The sign is also observable for others and their interpretation of its meaning. The sign is a physical record of previous work that feed back on the workers, thus mediating indirect communication and the record facilitates organisation of activities, which in this case is to fill silo 10 with wheat. These concepts provide, to some extent, similar descriptions of indirect interactions and coordination in this work place, mediated by signs (to use a common name for the different concepts), but there are also significant differences in that they explain different aspects of ‘how and why’ of mechanisms underlying coordination. Also, clearly many functions and mechanisms can co-exist in one and the same artefact, or sign. One thing these concepts do have in common though, is that they can all be related to indirect artefact-mediated communication that facilitate coordination. At this point we might further consider the control room as a system of agents and environment, and identify whether the system in fact displays stigmergy by using four questions suggested by Dipple et al. (2014, pp. 23–24): 1. ‘‘Does the agent leave a physical and measurable difference in the environment (i.e.: a sign”? Workers leave a number of physical signs (notes, etc.) on the control

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

8

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx

panel, which is a measurable difference in the environment. 2. ‘‘Is the sign contributing to the grand purpose (i.e.: a signal)”? The grand purpose is said to describe the ‘‘obvious coordination (however paradoxical) as a result of stigmergy and its participants”. The signs on the control panel are left as signals, they mean something to other workers, and they contribute significantly to individually performed activities being coordinated at a collective level. 3. ‘‘Does the receiving agent understand the signal and react in a way expected to contribute to the grand purpose”? Agent interactions are a vital part of the system since ‘‘a shared meaning develops and we understand the communicative function of a specific stigmergy mechanism‘‘. The physical signals left on the control panel are stigmergic mechanisms and their communicative function is understood by the perceiving workers due to a shared meaning. The shared meaning and understanding of the signal allows workers to react in ways that contribute to the grand purpose. 4. ‘‘Does creating the signal introduce an emergent communication which is vital to the self-organising function of stigmergy”? The creation of a signal is to introduce an emergent communication since the signal mediates what is going on or has to take place in the system. Over time, the communication evolves and leads to the creation of new signals that again introduces an emergent communication. The communication leads to individual activities being self-organised, and hence, it is vital to the selforganising function of stigmergy. In light of these questions, the system of agents and environment in this work place does display stigmergy. In fact, stigmergy is at the core of most activities performed in this setting, which facilitates the systems ‘‘grand purpose”. However, while stigmergy provides an explanation to the achievement of coordination, it does not account for the details of signs, their characteristics, or cognitive processes and effort beyond creating, understanding, and reacting upon them. Explaining the coordination within the work place in terms of stigmergy, explains what mechanisms are involved and how coordination is achieved, but it does not seem to be able to provide more nuanced explanations. 5. Summary and conclusions Considering that the concepts discussed here have been developed within different domains it is no surprise that they ‘‘don’t take each other’s contribution into account” (Howison et al., 2012), and so provide conceptual similarities and overlapping, but they do have different foci and purposes. The variety of concepts clearly shows that there is an interest in many domains to find explanations to the achievement of coordinated behaviour, and a combination of concepts may provide more elaborate explanations

and understanding than what they can provide on their own. On the other hand, stigmergy could be seen as a common denominator, and it has even been argued that theories of interactive cognition, like situated and distributed cognition, are reducible to stigmergy (Heylighen, 2011– 12, although no explanation of how that could be is provided). However, the concepts discussed here are hardly reducible to or replaceable by stigmergy, and replacing whole theories with stigmergy seems even less likely. Coordination in human activities is very much a matter of cognitive processing and signs produced by an agent are not only ‘intentional’ and have a ‘content’; considering a work place or some other context, the agent producing a sign often has a specific intention with the sign, which also affects its content and its characteristics. The characteristics of the sign in turn affect whether the sign will be sensed or observed, and its meaning and interpretation is related to the context of activities within which the sign is produced and observed. What stigmergy currently seems to lack is a recognition of nuances in people’s intentions and production of signs, the signs contents, and the perception of signs, in relation to the context. The concept of sign could usefully be developed to recognise qualitative differences in its role as a stigmergic mechanism, which would allow the concept to gain more explanatory power, thereby providing an even better tool for analysis of coordination in human activities. Clearly, human activity is stigmergic, and stigmergy could also to a wider extent be usefully integrated with cognitively oriented concepts and theories of interactive cognition to include the dimension of coordination between agents and provide further explanations to the coordination paradox so often seen in human activities.

References Bolici, F., Howison, J., & Crowston, K. (2009). Coordination without discussion? Socio-technical congruence and stigmergy in free and open software projects. In Socio-technical congruence workshop in conj international conference on software engineering. Canada: Vancouver. Bonabeau, E. (Ed.). (1999). Special issue: Stigmergy. Artificial Life, 5(2). Christensen, L. R. (2013). Stigmergy in human practice. Coordination in construction work. Cognitive Systems Research, 21, 40–51. Dawson, M. (2014). Embedded and situated cognition. In L. Shapiro (Ed.), The Routledge handbook of embodied cognition (pp. 59–67). New York: Routledge. Dipple, A., Raymond, K., & Docherty, M. (2014). General theory of stigmergy: Modelling stigma semantics. Cognitive Systems Research, 31–32, 61–92. Dix, A., Ramduny, D., & Wilkinson, J. (1998). Interaction in the large. Interacting with Computers, 11(1), 9–32. Dix, A., Ramduny-Ellis, D., & Wilkinson, J. (2004). Trigger analysis – Understanding broken tasks. In D. Diaper & N. Stanton (Eds.), The handbook of task analysis for human–computer interaction (pp. 381–400). Mahwah: Lawrence Erlbaum Associates. Espinosa, A., Lerch, J., & Kraut, R. (2002). Explicit vs. implicit coordination mechanisms and task dependencies: One size does not fit all. In E. Salas & S. M. Fiore (Eds.), Team cognition: Understanding the factors that drive process and performance (pp. 107–129). Washington, DC: American Psychological Association.

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006

T. Susi / Cognitive Systems Research xxx (2015) xxx–xxx Gibson, J. J. (1986). The ecological approach to visual perception. Hillsdale, NJ: Lawrence Erlbaum Associates. Heylighen, F. (2011–12). Stigmergy as a generic mechanism for coordination: Definition, varieties and aspects. ECCO working paper 2011-12. . Howison, J., Østerlund, C., Crowston, K., & Bolici, F. (2012). Stigmergy and implicit coordination in software development. In CSCW’12. February 11–15, Seattle, Washington, USA. Huang, H., Ren, C., & Jin, S. (2008). ‘‘Sign” is the sign: Towards a unified view of stigmergy. In Proceedings of the 2008 IEEE international conference on information and automation (pp. 535–540). Kirsh, D. (1996). Adapting the environment instead of oneself. Adaptive Behavior, 4(3/4), 415–452. Kirsh, D. (2001). The context of work. Human–Computer Interaction, 16 (2–4), 305–322. Lewis, T. G. (2013). Cognitive stigmergy: A study of emergence in smallgroup social networks. Cognitive Systems Research, 21, 7–21. Lugosky, J., & Dove, R. (2011). Identifying agile security patterns in adversarial stigmergic systems. In: Twenty-first annual international symposium of INCOSE. Denver, CO (US), June (17 pages). Marsden, J. (2013). Stigmergic self-organization and the improvisation of Ushahidi. Cognitive Systems Research, 21, 52–64. Marsh, L., & Onof, C. (2008). Stigmergic epistemology, stigmergic cognition. Cognitive Systems Research, 9(1–2), 136–149. Omicini, A., Ricci, A., & Viroli, M. (2008). Artifacts in the A&A metamodel for multi-agent systems. Autonomous Agents and Multi-Agent Systems, 17(3), 432–456. Parunak, H. V. D. (2006). A survey of environments and mechanisms for human–human stigmergy. In: D. Weyns, H. Van Dyke Parunak, & F. Michel (Eds.), Environments for multi-agent systems II: Second

9

international workshop, E4MAS 2005 (pp. 163–186). Utrecht, The Netherlands, July 2005. Selected revised and invited papers. Ricci, A., Omicini, A., Viroli, M., Gardelli, L., & Oliva, E. (2007). Cognitive stigmergy: Towards a framework based on agents and artifacts. In: D. Weyns, H. Van Dyke Parunak, & F. Michel (Eds.), Environments for multi-agent systems III: Third international workshop, E4MAS 2006 (pp. 124–140). Hakodate, Japan, May 2006. Selected revised and invited papers. Schmidt, K., & Simone, C. (1996). Coordination mechanisms: Towards a conceptual foundation of CSCW systems design. Computer Supported Cooperative Work: The Journal of Collaborative Computing, 5(2–3), 155–200. Secretan, J. (2013). Stigmergic dimensions of online creative interaction. Cognitive Systems Research, 21, 65–74. Susi, T. (2005). In search of the Holy Grail: Understanding artefact mediation in social interactions. In B. G. Bara, L. Barsalou, & M. Bucciarelli (Eds.), Proceedings of the 27th annual conference of the cognitive science society (pp. 2110–2115). Mahwah, NJ: Lawrence Erlbaum. Susi, T. (2006). The puzzle of social activity – The significance of tools in cognition and cooperation. Doctoral dissertation. University of Linko¨ping, Sweden. ISBN: 91-85523. Susi, T., & Ziemke, T. (2001). Social cognition, artefacts, and stigmergy: A comparative analysis of theoretical frameworks for the understanding of artefact-mediated collaborative activity. Cognitive Systems Research, 2(4), 273–290. Theraulaz, G., & Bonabeau, E. (1999). A brief history of stigmergy. Artificial Life, 5, 97–116. Tummolini, L., & Castelfranchi, C. (2007). Trace signals: The meanings of stigmergy. Environments for Multi-Agent Systems III, 141–156.

Please cite this article in press as: Susi, T. Social cognition, artefacts, and stigmergy revisited: Concepts of coordination. Cognitive Systems Research (2015), http://dx.doi.org/10.1016/j.cogsys.2015.12.006