Tag Archives: Agent-based simulation

Towards contextualized social simulation: Complementary use of Multi-Fuzzy Cognitive Maps and MABS

By Oswaldo Terán1 and Jose Aguilar2

1Escuela de Ciencias Empresariales, Universidad Católica del Norte, Coquimbo, Chile, and CESIMO, Universidad de Los Andes, Mérid.

2CEMISID, Universidad de Los Andes, Merida, Venezuela; GIDITIC, Universidad EAFIT, Medellin, Colombia; and Universidad de Alcala, Dpto. Automatica, Alcala de Henares, Spain.

Abstract.

This work suggests to complementarily use Multi-Fuzzy Cognitive Maps (MFCM) and Multi-agent Based Simulation (MABS) for social simulation studies, to overcome deficiencies of MABS for contextually understanding social systems, including difficulties for considering the historical and political domains of the systems, variation of social constructs such as goals and interest, as well as modeler’s perspective and assumptions. MFCM are a construction much closer than MABS to natural language and narratives, used to model systems appropriately conceptualized, with support of data and/or experts in the modeled domains. Diverse domains of interest can be included in a MFCM, permitting to incorporate the history and context of the system, explicitly represent and vary agents’ social constructs, as well as take into account modeling assumptions and  perspectives.  We briefly describe possible forms of complementarily use these modeling paradigms, and exemplifies the importance of the approach by considering its relevance to investigate othering and polarization.

1. Introduction

In order to understand better issues such as othering and polarization, there is a claim in social simulation for research that includes the important domains of history, politics and game of power, as well as for greater use of social science data, make more explicit and conscious about the models’ assumptions, and be more cautious in relation to the interpretation of the simulations’ results (Edmonds et al., 2020). We describe a possible form of dealing with these difficulties: combining Multi-Agent based Simulation (MABS) and Multi Fuzzy Cognitive Maps (MFCM) (or other forms of cognitive maps), suggesting new forms of dealing with complexity of social behavior. By using MFCM an alternative modeling perspective to MABS is introduced, which facilitates expressing the context of the model, and the modelers’ assumptions, as suggested in Terán (2004).  We will consider as a case studying othering and polarization, given the difficulties for modeling it via MABS (Edmonds et al., 2020). Our proposal permits to explicitly represent social constructs such as goals, interest and influence of powerful actors on, e.g., people’s othering and polarization, and so in better contextualizing the simulated model. Variations of social constructs (e.g., goals, othering, polarization, interests) can be characterized and modeled by using MFCM.

Combined use of MABS and Fuzzy Cognitive Maps (FCM) (MFCM, multi FCM, are an extension of FCM, see the Annex) has already been suggested, see for example Giabbanelli (2017). MABS develop models at the micro level, while FCM and MFCM permits us to create models at the macro or contextual level; the idea is to use one to complement the other, i.e., to generate rich feedback between them and enhance the modeling process. Additionally, Giabbanelli propose FCM as a representation closer than MABS to natural language, allowing more participatory models, and better representation of the decision making process. Giabbanelli recommend forms of combining these two modeling approaches, highlighting key questions modelers must be careful about. In this line, we also propose a combined usage of a MFCM and MABS to overcome deficiencies of MABS modelling in Social Simulation.

Initially (in section 2) we offer a description of human societies from a broad view point,  which recognizes their deep complexities and clarifies the need for better contextualizing simulation models, allowing modeling of diverse agents’ constructs, and making explicit modelers’ assumptions and perspectives. Afterwards (in section 3), we briefly review the drawbacks of MABS for modeling some of these deep complexities. Then (in section 4), MFCM are briefly described, supported on a brief technical account in the Annex. Following (in section 5), we suggest to complementarily use MABS and MFCM for having a more comprehensive representation of human societies and their context, e.g., to better model problems such as othering and polarization. MFCM will model context and give a conceptual mark for MABS (allowing to model variation of context, e.g., changes of agents’ interests or goals, making explicit modelers’ perspective and assumptions, among other advantages), which, in turn, can be used to explore in detail specific configurations or scenarios of interest suggested by the MFCM.  Finally (in section 6), some conclusions are given.

2. (A wide view of) Human societies and influence of communication media on actual culture

As humans and primates, we recognise the social groups within which we develop as people (e.g., family, the community where we grow up, partners at the school or at work) as part of our “large home”, in which its members develop a common identity, with strong rational and emotional links. Other groups beyond these close ones are “naturally” estrangers for us and its members “instinctively” seen as others. In large civilizations such as western society, we extend somewhat these limits to include nations, in certain respects. In groups we develop perspectives, follow certain myths and rites, and have common interests, viewpoints about problems, solutions for these, and give meaning to our life. Traditionally, human societies evolve from within groups by direct face to face interaction of its members, with diverse perspectives, goals, interest, and any other social construct with respect to other groups. Nowadays this evolution mainly from natural interaction has been importantly altered in some societies, especially western and western influenced societies, where social media has introduced a new form of communication and grouping: virtual grouping. Virtual grouping consists in the creation of groups, either formally or informally, by using the internet, and social networks such as Facebook, Instagram, etc. In this process, we access certain media sites, while discarding others, in accordance with our preferences, which in turn depends on our way of thinking and preferences created in social, both virtual and direct (face to face), interaction. Currently, social media, and traditional media (TV, newspapers, etc.) have a strong influence on our culture, impacting on ours myths, rites, perspectives, forms of life, goals, interests, opinion, reasoning, emotions, and othering.

Characteristics of virtual communication, e.g., anonymity, significantly impacts on all these constructs, which, differently from natural interaction, have a strong potential to promote polarization, because of several reasons, e.g., given that virtual environments usually create less reflexive groups, and emotional communication is poorer or lack deepness. Virtual interaction is poorer than direct social interaction: the lack of physical contact strongly reduces our emotional and reflexive connection. Virtual social interaction is “colder” than direct social interaction; e.g., lack of visual contact stops communication of many emotions that are transmitted via gestures, and prevents the call for attention from the other that visual contact and gestures demands.

Even more, many times sources and veracity of information, comments, ideas, and whatever is in social media, are not clear. Even more, fake news are common in social media, what generate false beliefs, and behavior of people influenced and somewhat controlled by those who promote fake news. Fake news can in this sense generate polarisation, as some groups in the society prefer certain media, and other groups choose a different one. As these media may promote different perspectives following interest of powerful actors (e.g., political parties), conflicting perspectives are induced in the different groups, what in turn generates polarization. Social media are highly sensitive to manipulation by powerful actors worldwide, including governments (because of, e.g., their geopolitical interests and strategies), corporations (in accordance with their economic goals), religious groups, political parties, among many others. Different groups of interests influence in direct and indirect, visible and hidden, forms the media, following a wide diversity of strategies, e.g., those of business marketing, which are supported by knowledge of people (e.g., psychology, sociology, games theory, etc.). Thus, the media can create and contribute to create visions of the word, or perspectives, in accordance with the interest of powerful international or national actors. For more about all this, see, e.g, Terán and Aguilar (2018).

As a consequence, people following media that promotes a world view, related with some powerful actor(s) (e.g., a political party or a group of governments) virtually group around media that support this world view, while other people do the same in relation to other media and powerful actor(s), who promote(s) a different perspective, which many times is in conflict with the first one. Thus, grouping following the media sometimes promotes groups with conflicting perspectives, goals, interests, etc.,  which generates polarization. We can find examples of this in diverse regions and countries of the world. The media has important responsibility for polarization in a diversity of issues such as regional integration in Europe, war in Ukraine, migrations from Middle East or Africa to Europe, etc. Consequently, media manipulation sometimes allow powerful actors to influence and somewhat control perspectives and social behavior. Even more, the influence of social media on people is sometimes stronger than the influence of direct social interaction. All these introduce deep complex issues in social human interaction and behavior. This is why we have chosen polarization as the case study for his essay.

Consequently, to comprehend actual human behavior, and in particular polarization, it is necessary to appropriately take into account the social context, what permits to understand better the actual complexity of social interaction, e.g., how powerful international, national, and local actors’ influence on media affects people perspectives, goals, interest, and polarization, as well as their strategies and actions in doing so. Contextualized modeling will help in determining social constructs (goals, interests, etc.) in certain situations, and their variation from situation to situation. For this, we suggest complementing MABS with MFCM. For more about the consequences of virtual interaction, see for example Prensky (2001a, 2001b). Prensky (2009) has also suggested forms to overcome such consequences: to promote digital wisdom. MABS and MFCM models will help in defining forms of dealing with the problems of high exposure to social networks, in line with Prensky’s concerns.

3. Weakness of the MABS approach for modeling context

Edmonds et al. (2020) recognize that MABS models assume a “state of the world” or “state of nature” that does not include the historical context of the agent, e.g., in such a way that they explicitly present goals, interests, etc., and pursue them via political actions, sometimes exerting power over others. For instance, the agents can not change their goals, interests or desires during the simulation, to show certain evolution, as a consequence of reflection and experience at the level of desires, allowing cognitive variations. The models are strongly limited in relation to representing the context of the social interaction, which in part determines variation of important factors of agents’ behavior, e.g., goals. This, to a good extent, is due to lack of representation of the agents’ context. For the same reason, it is difficult to represent modelers assumptions and perspectives, which might also be influenced by social media and powerful actors, as explained above.

The Special Issue of the Social Science Computer Review. (Volume 38, Issue 4, August 2020, see for instance Edmonds et al. (2020) and Rocco and Wander (2020)), presents several models aiming at dealing with some of these drawbacks of MABS, specifically, to relate models to social science data, be more aware about the models’ assumptions, and be more cautious in relation to the interpretation of the simulations’ results. However, in these works diverse difficulties are not addressed, e.g.,  having appropriate representation of the context in order to explicitly consider diverse constructs, e.g., goals and interests, as well as having a wide representation of modelers perspectives and assumptions so that diverse perspectives can be addressed and compared, among other important matters.

MABS represent social interaction, i.e., the interaction in a group, where the agent’s goal, and other social constructs are assumed given, not variable, and to understand the context where they appear is not of interest or is out of reach (too difficult). However, as explained above, agents are in diverse social groups, not only in the simulated one, and so goals, interests, and beliefs in the modeled group are shaped in accordance to their interactions in diverse groups, and the influence of multiple, virtual and natural groups in which they participate. In order to represent variations of such elements, the context must be taken into account, as well as to elaborate models from narratives. MFCM is naturally close to narratives, as it is elaborated from conceptual frameworks. In this sense, MFCM might represent an intermediate step towards MABS models. In a MFCM and in the steps towards elaborating the MABS, modelers’ perspectives and assumptions can be made explicit. In addition, MABS presents limitations to determine the conditions for which a certain behavior or tendency occurs (Terán, 2001; Terán et al. 2001), i.e., for making strong inferences and theorem proving of tendencies for subsets of the theory of the simulation, which could more easily be performed in the MFCM. Hopefully, exploring configurations of the MFCM the proof could be carried out indirectly, in a higher level than in MABS, as has already been suggested in previous papers (Aguilar et al., 2020; Perozo et al., 2013).

4. Multi-Fuzzy Cognitive Maps (MFCM)

We suggest conceptual or cognitive maps as a more flexible form than MABS to represent context of a social situation, and in particular MFCM, as implemented by Aguilar and others (see e.g., Kosko, 1986; Aguilar 2005, 2013, 2016; Aguilar et al., 2016, 2020; Contreras and Aguilar, 2010; and Sánchez et al., 2019; Puerto et al., 2019). A brief description of Fuzzy and Multi-fuzzy cognitive maps, following Sánchez et al. (2019), is given in the Annex.

Fuzzy cognitive maps help us in describing the context via qualitative (e.g., very low, low, medium, high, too high) and quantitative variables, as indicated in the annex. The system is represented by the network of concepts (variables) interrelated via weights (also given by variables). The high level of the MFCM paradigm, differently from a MABS, permits us to explicit different elements of the models such as the agents constructs (goals, interests, etc.), as well as the modelers assumptions and perspectives, as suggested in Terán (2004). MFCM will facilitate to explicit the accumulated set of assumptions (“abstraction-assumptions, design-assumptions, inference-assumptions, analysis-assumptions, interpretation-assumptions and application-assumptions”, as these are summarized in Terán, 2004).

In a MFCM, a particular situation of the system is given by a specific configuration of the weights (see, e.g., Sánchez et al., 2019). Suppose we are dealing with a model similar to that elaborated in Sánchez et al. (2019) to study the quality of opinion in a community. Sánchez et al. examine the capabilities of the MFCM for knowledge description and extraction about opinions presented in a certain topic, allowing the assessment of the quality of public opinion. Special attention is offered to the influence of the media on public opinion. The evolution of the concepts and relationships is presented. Concepts define the relevant aspects from which public opinion emerges, covering diverse domains, for instance, the social, technological and psycho-biological ones. The MFCM permits to identify the media preferred by the public in order to better understand several issues, including the high esteem that the new communication media hold.

In line with this, let us assume that we want to understand the quality of public opinion in a community of Europe about diverse issues during 2022. This network, with a certain configuration of the weights, but unspecified concepts, represents a social system with a certain structure (as the weights are given) that is in some sense general, as the values of the concepts can still vary. Variation of the concepts represents different scenarios of the social system (with a given structure, defined by the weights); e.g., the model of a European community considered in relation to three scenarios regarding the state of public opinion in relation to specific issues: 1: climate change, 2: situation of tourism in the community, 3: secondary effects of the COVID-19 vaccines. The weights of the network are determined by using a variety of scenarios; i.e., the network is trained with several scenarios, for which all possible values of the concepts are known. Once the network is trained, it can be used to infer unknown specific values of the concepts for other scenarios (following Aguilar et al. 2020; Sánchez et al. 2019; Terán y Aguilar, 2018); e.g., the state of public opinion in relation to the involvement of EU in the war in Ukraine. Even more, by exploring an appropriate set of scenarios, proofs about the state of certain concepts can be developed; e.g., that a majority of people in the community is against direct EU involvement in the war in Ukraine. The proof could be carried out for a subset of the possible configurations of a domain, several domains, or part of a domain, e.g., for the psycho-biological domain. Additionally, having an appropriate elaboration of the model would allow evaluating how polarized is the opinion of the community in relation to the involvement of EU in that war.

Diverse configurations of the MFCM can represent different modelers’ perspectives and assumptions, as well as various agents’ constructs, such as goals, interests, etc., allowing to deal with the above described drawbacks of MABS to cope with complexity of social systems.

5. Combined use of MABS and cognitive maps

The combined usage will give at least two levels of modeling: the inner, defined by the agents’ interaction concreted in a MABS, and the outer or contextual one, given by the MFCM. These will be the two last levels in the description given in 5.1. Interaction between these models occurs as the modeler interprets each model outputs and feedbacks the other. Ideally, we would have direct automatic feedback between these models.

5.1 Levels of description of the System

In order to contextually model a social system and investigate problems such as polarization, we suggest below five levels of description of the system. The first three levels are not directly associated to computational models, while levels four and five are descriptions that assist development of the computational models: MFCM and MABS, respectively. Each level gives context to the following one (the first gives context to the second, etc.). A lower level (e.g., 1. in relation to 2.) of description corresponds to a more general language, as suggested in Terán (2004). Each level must take into account the previous levels, especially the immediately superior level, which gives the most immediate context to it. This description is in line with the suggestions given in Terán (Idem). Each description makes certain assumptions and is shaped by the modeler’s perspective, which in part is coming from those actors given information to build the model. Assumptions and perspectives introduced in level of modeling i, i = 1, …, 6, can be called Assumptions-given in (i) and Perspectives-given in (i). At levels of description j, Assumption i = 1, 2, …, i are accumulated, and can be called Assumptions(j), as well as holistic Perspective(j) based on Perspectives-given in (i), i = 1, 2, .., i.  These assumptions and perspectives correspond to those defined in Terán (2004) as “abstraction-assumptions, design-assumptions, inference-assumptions, analysis-assumptions, interpretation-assumptions and application-assumptions”.

  1. Describe in natural language the system, including its relevant history, with emphasis in culture (practices, costumes, etc.) and behavior of individuals and groups relevant to the object of the study, e.g., from a historical-ontological perspective. Here, how the system has reached its actual situation is explained. This will give a global view of the society and the general form of behavior, problematics, conflicts, etc.
  2. Describe the diverse relevant domains given context to the system of interest in accordance with the study, e.g., political, economical, dominant actors, etc., and the relationships among them. Concrete specifications of these domains sets scenarios for the real system, i.e., possible configurations of it.
  3. Describe the particular social group of interest as part of the society explained above, in 1., and the domains given in 2., showing its particularities, e.g., culturally, in terms of interests, situation and social interaction of this group in relation with other groups in the whole society, in accordance to the problematic addressed in the study.
  4. Elaborate a cognitive map of the situation of the social group of interest, following the description given in 2 and 3. This is a description to be represented in a computational language, such as the MFCM tool developed by Contreras and Aguilar (2010).
  5. Describe the MABS model. The MABS model is then represented in a simulation language.
  6. The computational MFCM (or other cognitive map) and the MABS developed following 4. and 5. are then used to generate the virtual outputs and simulation study.

5.2 Possible combined uses of MFCM and MABS.

The MFCM (in general a cognitive or conceptual map) gives context to the MABS model (modeler’s assumptions and perspectives are added in the process, as indicated above), while the MABS model represents in detail the interaction of the agents’ considered in the MFCM for a specific scenario of this, as indicated in the levels of modeling given above. With this idea in mind, among the specific forms a combined usage of ABM and MFCM, we have:

i) Offering feedback from the MFCM to the MABS. A MABS in a certain configuration can be used to generate either directly or indirectly (e.g., with additional verification or manipulations) input for a simulation model. For example, in parallel to the model presented in Sánchez et al. (2019), where the domains social, biological-psychological, technological and state of the opinion are displayed, a MABS model can be developed to represent the interaction between social entities, such as people who receive information from the media, the media itself, and powerful actors who design the agenda setting of the media. This MABS model might use diverse methodologies, e.g., endorsements or BDI, to represent social interaction, or a higher level of interaction where actors share resources, on which they have interest, as in a System of Organized Action (see, e.g., SocLab models Terán and Sibertin-Blanc, 2020; Sibertin-Blanc et al., 2013). Constructs required as inputs for the model, e.g., goals, interests, values to define the endorsements schema, etc. could be deducted from the MFCM, as direct values of some concepts or functions (mathematical, logical, etc.) of the concepts. These operations could be defined by experts in the modeled domains (e.g., media owners, academics working in the area, etc.). Ideally, we would have an isomorphic relation between some variables of the MABS and variables of the MFCM – however, this is not the usual case –. In this process, the MABS is contextualized by the MFCM, whose modeling level also permits to identify modeler’s assumptions and perspectives. Also, in a narrative, and then in the MFCM, goal, interest, and other constructs can be explicitly represented, then varied and their consequence understood in order to feedback the MABS model.

ii) Giving feedback to the MFCM from the MABS. Inputs and outputs values of the MABS simulation can be used as an input to the MFCM, e.g., as a set of scenarios to train the network and determine a certain structure of the MFCM, or to determine a specific scenario/ configuration (where both, the weights and the concepts are known).

iii) Determining conditions of correspondence among the models. By simulating the MABS associated to certain scenarios of the MFCM, or, vice-verse, by determining the scenarios of MFCM related to certain MABS, the consistency among the two models and possible errors, omissions, etc. in one of the models can be detected, and then the corrections applied. Even more, this exercise can hopefully determine certain rules or conditions of correspondence among the MABS and the MFCM.

iv) Using a model to verify properties in the other model. Once certain correspondence among the models has been determined, we can use one of the models to help in determining properties of the other. For instance, a proof of a tendency in an MABS (this has been an important area of research, Terán, 2001; Terán et al. 2001; Edmonds et al., 2006) could be developed in a much easier form in the corresponding MFCM. For this, we need to characterize the set of configurations of the MFCM corresponding to the set of configurations of the MABS for which we want to perform the proof.

These possible combined uses of MFCM and MABS do not exhaust all potentials, and diverse other alternatives could appear in accordance with the needs in a particular study. Even more, automatic feedback between MFCM (or other cognitive or conceptual map) and MABS could be implemented in the future, to facilitate the mutual contributions between the two modeling approaches. This would cover modeling requirements the MABS in itself does not support at present.

5.3 A case: Modeling othering and polarization, the case of “children with virtually mediated culture”.

We outline a possible model that considers othering and polarization. In section 2 we described a society. In a society, as virtual groups become homogeneous in beliefs, motivations, intentions and behavior, certain sort of endogamy of ideas and opinions appear, constraining the variety and richness of perspectives from which people observe and judge others, making them generally less tolerant to others, more restrictive in accepting opinions and behavior of others, so less inclusive. This has diverse additional effects, for instance, increase of polarization between virtual groups regarding a diversity of themes. Problems such as polarization occur also in children with strong usage of virtual social networks (see, Prensky, 2001a, 2001b, 2009).

To investigate this problem and support the MABS, as a case, we suggest a MFCM with four levels (see Figure 1). The goal of the models (MFCM and MABS) would be to better understand the differences between the communities of children whose interaction is basically virtually mediated and the community of children whose interaction is face to face, or direct, people to people. In general, it is of interest to determine the state of othering and polarization for diverse configurations of the MFCM. As we explained above, there are clear differences between virtual and face to face interaction, consequently the upper layers in Figure 1 are the two possible niches of cultural acquisition (costumes, points of view, etc.), during life of people (ontogenesis), namely, the virtual mediated culture and the direct, face to face, cultural acquisition. These two layers involve interaction among diverse actors (e.g., people, media and powerful actors are present in layer 2). Layer 2 represents technological actors, while layer 1 represents social interaction, but both of them might involve other elements, if required. The third level represent those biological aspects related with behavior, which are created via culture: the psycho-biological level. Both levels, 1 and 2, affect the third layer, as emotions, reasoning, etc., are founded on people interaction and have a cultural base. Constructs of behavior such as goals, interests, desires, polarization, etc., appear and can be explicitly represented at this level. This third level, in turn, impacts on the overall state of the community, e.g., on the auto-generative capacity of the society, finally affecting global society/community’s othering and polarization, as our emotions, reasoning, etc., impact on our view point, on othering, etc. These last are variables defined in terms of the previous levels. In this model, the definition of concepts such as “othering” and “polarization” is crucial, and indicates basic modeling assumptions and perspective. Finally, the overall state of society/community impacts back on the cultural niches (layers 1 and 2).

In a specific situation, the whole interaction (1) of the society or community is divided between the two niches given by layers 1 and 2, a proportion of interaction frequency occurs as virtual communication, and the compliment (one minus the proportion of virtual interaction), occurs as direct, face to face, contact. This is represented in Figure 3 by the variable “Proportion of virtual interaction type”. Changes of this variable allows us to explore diverse configurations or scenarios of interaction, ranging from total virtual interaction (null direct contact) (the variable takes the value 1), to null virtual interaction (total direct contact) (the variable takes the value 0).

The four levels of MFCM

Figure 1. The four levels of the MFCM. The two alternative niches structuring the psycho-biology of people are at the top of the process. The overall state represents general measures such as the auto-generative character of a social system, and attitudes including othering and polarization.

Example of possible variables in some levels (Figure 1) are:

i) Face to face interaction, and ii) virtual interaction or technological: The next variables are candidates to be at these levels. Degree of:

  •  Coherence of the interaction (possible state: good, etc.);
  •  Identification of the others in the interaction (good or clear, etc.);
  •  Richness of the interaction (high or good, etc.);
  •  Truthfulness of the messages (fairness) (e.g., good: messages and communication are fair);
  •  Openness of the community (e.g., high: usually people is open to interact with others);
  •  Speed of the interaction (e.g.: low, medium, …);
  •  Intentioned influence and control of the communication by powerful actors (e.g., high, medium, low, ..);

iii) Psycho-biological level

  •  Reflection (state: good means that people question their experiences, and observed phenomena);
  • Closeness of interpretations, attitudes, desires, intentions, and plans (a high value means that people’s interpretations, etc., are not very different);
  •  Emotion and mood;
  •  Empathy;
  •  Addiction to virtual interaction;
  • Goal;
  •  Interest;
  •  Immediatism (propensity to do things quickly and constantly change focus of reasoning).

iv) Overall state of people and society:

  •  Auto-generative capacity of the society;
  •  Capacity of people to reflect about social situations (and autonomously look for solutions);
  •  Othering;
  •  Polarization; 

Concepts at each layer impacts concepts at the other layers. E.g., concepts of level three have a strong impact on concepts of the fourth layer, such as “polarization”, and “auto-generative character of the society”.

As indicated above, to understand the dynamics of the MFCM we can develop a wide range of scenarios, for instance, varying the switch “Proportion of virtual interaction” in the interval [0, 1], to explore a set of scenarios for which the degree of virtual interaction increases from 1 to 0, as the proportion of direct interaction decreases from 1 to 0 (the real case corresponds to an intermediate value between 1 and 0). These experiments will help us in understanding better the consequences of virtual mediated culture. Even when the outline of the model presented here might need some adjustments and improvements, the present proposal keeps its potential to reach this goal.

The MFCM will be useful to deal with many issues and questions of interest, for instance:

  • How social networks affect basic social attitudes such as: i) critical rationality (people’s habit for questioning and explaining their experience (issues/phenomena in their life)), ii) tolerance, iii) compromise with public well-being, and iv) othering and polarization
  •  How social networks affect social feelings, such as empathy.

The MABS model will be elaborated in accordance with the description of the MFCM indicated above. In particular, different values of the social constructs at levels 1 and 2 (e.g., goals and interests of the actors), and the corresponding state of layer 3 (e.g., Polarization), imply diverse MABS models.

The whole network of concepts, the particular network of concepts at each level, and the definition of each concept, offers a perspective of the modelers. Different modelers can develop these elements of the model differently. Assumptions can be identified also at each level. Both, perspectives and assumptions come from the modelers as well as from the theories, consult to experts to create the model, etc.. An specific model is not part of this essay, but rather a subject of future work.

6. Conclusion

Social simulation has been widely recognized as an alternative to study social systems, using diverse modeling tools, including MABS, which, however, present some limitations, like any other research tool. One of the deficiencies of MABS is their limitations to contextually modeling social systems; e.g., to suitably include the historical and political contexts or domains; difficulties to represent variation of agents’ constructs, e.g., goals and interest; and drawbacks to made explicit modeler’s assumptions and perspectives. In this paper, we have suggested to mutually complement MABS and MFCM, to overcome MABS drawbacks, to potentiate the usefulness of MABS to represent social systems.

We argue that the high level of the MFCM paradigm permits us to express different elements of the models such as the agents’ constructs (goals, interests, etc.), as well as the modelers assumptions and perspectives, as suggested in Terán (2004). Thus, MFCM facilitates the identification of the accumulated set of assumptions during the modeling process. Even more, diverse configurations of a MFCM can represent diverse modelers’ perspectives and assumptions, as well as diverse agents constructs, such as goals, interests, etc., allowing to deal with the above described complexity. This permits us to more realistically elaborate models of a wide diversity of social problems, e.g., polarization, and consequences of the influence of social networks in culture.

Among the forms MFCM and MABS complement each other we have identified the followings:  mutual feedbacking of variables and concepts between the MFCM and the MABS, determining conditions of correspondence among the models, what facilitate other modeling needs, e.g., using a model to verify properties in the other model (e.g., proofs required in a MABS could be carried out in a corresponding MFCM).

A case study was outlined to exemplify the problematic that can be addressed and the advantages of using MFCM to complement MABS: Modeling othering and polarization, the case of children with virtually mediated culture. Combined use of MFCM and MABS in this case will contribute to understand better the problems created by the high use of digital interaction, especially social networks, as described by Prensky (2001a, 2001b, 2009), given that virtual interaction has strong influence on our culture, impacting on ours myths, rites, perspectives, goals, interests, opinion, othering and polarization, etc. Characteristics of virtual communication, e.g., anonymity, significantly impacts on all these constructs, which, differently from natural interaction, have a strong potential to promote certain tendencies of such constructs, e.g., polarization or our opinions, because of several reasons, for instance, given that virtual environments usually create less reflexive groups, while emotional communication is poorer or lack deepness. It is difficult to represent all these dynamics in a MABS, but it can be alternatively expressed in a MFCM.

The purpose of the work was to give an outline of the proposal; future work will be conducted to wholly develop concrete study cases with complementary MABS and MFCM models.

References

Aguilar, J. (2005). A Survey about Fuzzy Cognitive Maps Papers. International journal of computational cognition 3 (2), pp. 27-33.

Aguilar, J. (2013) Different Dynamic Causal Relationship Approaches for Cognitive Maps”, Applied Soft Computing, Elsevier, 13(1), pp. 271–282.

Aguilar, J. (2016) “Multilayer Cognitive Maps in the Resolution of Problems using the FCM Designer Tool”, Applied Artificial Intelligence, 30 (7), pp. 720-743.

Aguilar J., Hidalgo J., Osuna F., Perez N.(2016) Multilayer Cognitive Maps to Model Problems”, Proc. IEEE World Congress on Computational Intelligence,  pp. 1547-1553, 2016.

Aguilar Jose, Yolmer Romero y OswaldoTerán (2020). “Analysis of the effect on the marketing of the sport product from the “par Conditio” principle in Latin American football and baseball competitions”. Submitted to International Journal of Knowledge-Based and Intelligent Engineering Systems.

Contreras, J. Aguilar, J. (2010). “The FCM Designer Tool”. Fuzzy Cognitive Maps: Advances in Theory, Methodologies, Tools and Application (Ed. M. Glykas), Springer, pp. 71-88, 2010.

Edmonds, B. and Hales, D. and Lessard-Phillips, L.(2020). Simulation Models of Ethnocentrism and Diversity: An Introduction to the Special Issue. Social Science Computer Review. Volume 38 Issue 4, August 2020, 359–364, https://journals.sagepub.com/toc/ssce/38/4

Edmonds Bruce, Oswaldo Terán y Grary Polhill (2006). “To the Outer Limits and Beyond –characterising the envelope of social simulation trajectories”, Proceedings of theThe First World Congress on Social Simulation WCSS”, Kyoto, 21-25 Agosto, 2006.

Giabbanelli Philippe, Steven Gray and Payam Aminpour (2017), Combining fuzzy cognitive maps with agent-based modeling: Frameworks and pitfalls of a powerful hybrid modeling approach to understand human-environment interactions, Environmental Modeling and Software, September 2017,  95:320-325 DOI:10.1016/j.envsoft.2017.06.040

Kosko, B. (1986). Fuzzy Cognitive Maps. International Journal of Man-Machine Studies 24, pp. 65-75.

Perozo Niriaska, Jose Aguilar, Oswaldo Terán y Heidi Molina (2013). A Verification Method for MASOES, IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS, Vol 43, num 1, February, pp. 64-76,  ISBN: 1083-4419. DOI: 10.1109/TSMCB.2012.2199106

Prensky, Marc (2001a). Digital Natives, Digital Immigrants Part 1, On the Horizon, Vol. 9, No. 5, September, pp 1-6. DOI: 10.1108/10748120110424816

Prensky, Marc (2001b). Do They Really Think Differently? Digital Natives, Digital Immigrants Part 2, On the Horizon, Vol. 9, No. 6, October 2001, pp 1-6. DOI: 10.1108/10748120110424843

Prensky, Marc (2009). “H. Sapiens Digital: From Digital Immigrants and Digital Natives to Digital Wisdom,” Innovate: Journal of Online Education: Vol. 5 : Issue 3, Article 1. Available at: https://nsuworks.nova.edu/innovate/vol5/iss3/1

Puerto E., Aguilar J., Chávez D., López C. (2019) Using Multilayer Fuzzy Cognitive Maps to Diagnose Autism Spectrum Disorder, Applied Soft Computing Journal, 75, pp. 58–71.

Rocco Paolillo, and Wander Jager (2020). Simulating Acculturation Dynamics Between Migrants and Locals in Relation to Network Formation, Social Science Computer Review. Volume 38 Issue 4, August 2020, pp. 365–386. https://journals.sagepub.com/toc/ssce/38/4

Sánchez Hebert, Jose Aguilar, Oswaldo Terán, José Gutiérrez de Mesa (2019). “Modeling the process of shaping the public opinion through Multilevel Fuzzy Cognitive Maps”, Applied Soft Computing, Volume 85. https://doi.org/10.1016/j.asoc.2019.105756.

Sibertin-Blanc, C., Roggero, P., Adreit, F., Baldet, B., Chapron, P., El-Gemayel, J., Mailliard, M., and Sandri, S. (2013). “SocLab: A Framework for the Modeling, Simulation and Analysis of Power in Social Organizations”, Journal of Artificial Societies and Social Simulation (JASSS), 16(4). http://jasss.soc.surrey.ac.uk/

Terán Oswaldo (2001). Emergent Tendencies in Multi-Agent Based Simulations Using Constraint-Based Methods to Effect Practical Proofs Over Finite Subsets of Simulation Outcomes, Doctoral Thesis, Centre for Policy Modelling, Manchester Metropolitan University, 2001.

Terán Oswaldo  (2004). Understanding MABS and Social Simulation: Switching Between Languages in a Hierarchy of Levels, Journal of Artificial Societies and Social Simulation vol. 7, no. 4. http://jasss.soc.surrey.ac.uk/7/4/5.html

Terán Oswaldo, Edmonds Bruce and Steve Wallis (2001) “Mapping the Envelope of Social Simulation Trajectories”, Lecture Notes in Artificial Intelligence (Subserie de Lecture Notes in Computer Science): MABS, Volume 1979, pp. 229-243, Springer, Alemania, 2001

Terán Oswaldo y Jose Aguilar (2018). “Modelo del proceso de influencia de los medios de comunicación social en la opinión pública”, EDUCERE, 21 (71), Universidad de Los Andes, Mérida, Venezuela. https://www.redalyc.org/toc.oa?id=356&numero=56002

Terán Oswaldo and Christophe Sibertin-Blanc (2020). Impact on Cooperation of altruism, tenacity, and the need of a resource,  accepted in IEEE Transactions on Computational Social Systems. DOI: 10.1109/TCSS.2021.3136821.

Annex. Fuzzy Cognitive Maps (FCM) and Muti-Fuzzy Cognitive Maps (MFCM).

Cognitive map theory is based on symbolic representation for the description of a system. It uses information, knowledge and experience, to describe particular domains using concepts (variables, states, inputs, outputs), and the relationships between them (Aguilar 2005, 2013, 2016). Cognitive maps can be understood as directed graphs, whose arcs represent causal connections between the nodes (concepts), used to denote knowledge. An arc with a positive sign (alternatively, negative sign), going from node X to node Y means that X (causally) increases (alternatively, decreases) Y. Cognitive maps are graphically represented: concepts are connected by arcs through a connection matrix. In the connection matrix, the i-nth line represents the weight of the arc connections directed outside of the  concept. The i-nth column lists the arcs directed toward , i.e., those affecting .

The conceptual development of FCMs rests on the definition and dynamic of concepts and relationships created by the theory of fuzzy sets (Kosko, 1986). FCM can describe any system using a causality-based model (that indicates positive or negative relationships), which takes fuzzy values and is dynamic (i.e., the effect of a change in one concept/node affects other nodes, which then affect further nodes). This structure establishes the forward and backward propagation of causality (Aguilar, 2005, 2013, 2016). Thus, the concepts and relations can be represented as fuzzy variables (expressed in linguistic terms), such as “Almost Always”, “Always”, “Normally”, “Some (see Figure 2).

The value of a concept depends on its previous iterations, following the equation (1):

Screenshot 2022-05-23 at 15.10.24

Cm(i+1) stands for the value of the concept in the next iteration after the iteration i, N indicates the number of concepts, wm,k represents the value of the causal relationship between the concept Ck and the concept Cm, and S(y) is a function used to normalize the value of the concept.

ot-fig2

Figure 2. Example of an FCM (taken from Sánchez et al., 2019).

MFCM is an extension of the FCM. It is a FCM with several layers where each layer represents a set of concepts that define a specific domain of a system. To construct a MFCM, the previous equation for calculating the current status of the concepts of a FCM is modified, to describe the relationships between different layers (Aguilar, 2016):

Where F(p) is the input function generated by the relationships among different layers, and p is the set of concepts of the other layers that impact this concept. Thus, the update function of the concepts has two parts. The first part, the classic, calculates the value of  concept in iteration  based on the values of concepts in the previous iteration . All these concepts belong to the same layer where the “m” concept belongs. The second part is the result of the causal relationship between the concepts in different levels of the MFCM.


Terán, O. & Aguilar, J. (2022) Towards contextualized social simulation: Complementary use of Multi-Fuzzy Cognitive Maps and MABS. Review of Artificial Societies and Social Simulation, 25th May 2022. https://rofasss.org/2022/05/25/MFCM-MABS


Artificial Sociality Manifesto

By Gert Jan Hofstede1*, Christopher Frantz2, Jesse Hoey3, Geeske Scholz4, and Tobias Schröder5

*Corresponding author, 1Information Technology, Wageningen, 2Department of Computer Science, Norwegian University of Science and Technology, 3School of Computer Science, University of Waterloo, 4Institut für Umweltsystemforschung, Universität Osnabrück, 5Potsdam University of Applied Sciences

Table of Contents

Approach

Ambition

With this position paper the authors posit the need for a research area of Artificial Sociality. In brief this means “computational models of the essentials of human social behaviour”; we shall elaborate below. The need for artificial sociality is justified by the encroachment of simulations and knowledge technology, including Artificial Intelligence (AI), into the fabric of our societies. This includes smart devices, biosensors, facial recognition, coordination apps, surveillance apps, search engines, home and care robots, social media, machine learning modules, and agent-based simulation models of socio-ecological and socio-economic systems. It will include many more invasive technologies that will be invented in the coming decades. Artificial sociality is a way to connect human drives and emotions to the challenges our societies face, and the management and policy actions we need to take. In contrast to mainstream AI research, artificial sociality targets the social embeddedness of human behaviour and experience; we could say the collective intelligence of human societies rather than the individual intelligence of single agents. Human sociality has characteristics that differ from other varieties of sociality, while having variation across cultures (Henrich, 2016). In this piece, we concentrate on the incorporation of human sociality into agent-based computational social simulation models as a testbed for the integration of the various elements of artificial sociality.

The issue of artificial sociality is not new, as we’ll discuss below in the “State of the art” section. Our evolutionary perspective, we feel, offers new possibilities for integrating various strands of research. Our ambition is mainly to find a robust ontology for artificial human sociality, rooted in our actual evolutionary history and allowing to distinguish cultures. We hope that efforts at engineering computational agents and societies can benefit from this work.

Why is sociality so important?

Humans are eusocial

Sociality is a word used across various sciences. Neuroscientist Antonio Damasio makes it a central concept, arguing that it is present in all social creatures, even long predating multicellular organisms (Damasio, 2018). In agreement with Wilson & Holldöbler (Edward O. Wilson & Hölldobler, 2005), Wikipedia defines it in a biological way: “Sociality is the degree to which individuals in an animal population tend to associate in social groups (gregariousness) and form cooperative societies”. The site continues: “The highest degree of sociality recognized by sociobiologists is eusociality. A eusocial taxon is one that exhibits overlapping adult generationsreproductive division of labor, cooperative care of young, (…).” Obviously, this definition holds for humans. We are a eusocial primate species.

Why are we in the world?

A grand question in philosophy is “Why are we in the world?”. Evolutionary biology would answer “because our ancestors reproduced, ever since the beginnings of life”. The next question is “Why did our ancestors reproduce?” Well, they did so because “they were fit, and conquered natural and human-made hazards”. Thirdly, “Why were they fit?” This third “why” question takes us to sociality. Being eusocial gave our ancestors the fitness they needed. It allowed them to cooperate and divide tasks in groups. Millions of years ago, early hominins gathered, hunted, defended themselves, cared for the weak, exchanged goods and foods (G. Hofstede, Hofstede, & Minkov, 2010), chapter 12.

Sociality integrates elements of all possible sciences that are useful in comprehensively modelling human (or non-human) social behaviour, drives, and decision making. It spans from the “what” to the “why” to the “how”. The notion of sociality changes the meaning of the concept of intelligence into something that could be group-level, not individual-level. The most astounding fact about humans is the high degree of social or collective intelligence. Because of the protection it affords, collective intelligence even raises the tolerance for individual ineptness (Diamond, 1999).

Artificial sociality

Artificial sociality is the study of sociality by means of computational modelling. This could take many forms, e.g. social robotics, body-worn devices. In this paper we focus on computational social simulation with a particular focus on sociality. The application to computational social simulation sets purpose and limits to the selection of potentially relevant knowledge. Artificial sociality will be concerned with building blocks and primitives that are chosen so as to be reusable for a multitude of applications. In this sense it is a transformative endeavour. It offers a systematic integration of the existing insulated approaches sponsored by diverse disciplines to understand and analyse the human condition in all its facets. The primitives developed for artificial sociality should have the potential to be used by a great many applied scholars. More importantly, the dedicated integrated treatment of disciplines is increasingly recognised as necessary to produce sufficiently accurate insights, such as the impact of cultural aspects on the assessment of social policy outcomes (Diallo, Shults, & Wildman, 2020). Applications that benefit from a systematic consideration of artificial sociality include models of human collective action in society, in socio-environmental, socio-economical, or socio-technical systems. Typically, these models would be used to support policy making by achieving a better understanding of the dynamics of target systems.

The history of sociality

Early hominins were mentioned above. In the evolution of life, sociality is actually much older than that. To properly appreciate its importance, we’ll present a brief history of sociality.

Sociality is as old as slime moulds, primitive organisms (“Protista”) that are usually monocellular (e.g. Dictyostelium). Slime moulds know collective action and large-scale division of labour. Social insects such as bees and ants are a more familiar case of successful sociality. Among mammals, there are the burrowing mole rats who live in eusocial colonies. These, or similar, life forms are linked to us by an unbroken chain of life. Sociality has an ancient path dependency.

Hominins

Limiting ourselves to the last million years, our hominin ancestors have brought sociality to a new level. In contrast to other primates, humans have not radiated into distinct species, but merged into one genetically closely related pool, with tremendous cultural variation. They did this through a combination of migrating, fighting, spreading of diseases, cross-breeding, and massive copying of inventions. Some of the latter are mastery of fire, language, script, law, agriculture, religion, weapons and money. Our present-day sociality is the outcome of an unbroken chain of reproduction, all the way since the origins of life until today. At present, fission-fusion dynamics happen all the time in all human societies. Divisions between groups of people are deeply gut-felt. They range from stable across generations to ephemeral; but they are not genetically deep, nor absolute. Yet they matter greatly for the behaviour of our policy-relevant systems. Religions, political alliances, trade networks, but also social media hypes and terrorist movements are cases in point.

Victims of reason

In recent centuries, humans have tended to forget that for all our cleverness and symbolic intelligence, humans are also still social mammals with deep relational drives. Our relational drives tell our intelligence what to do, and do so generally without being transparent to us (Haidt, 2012; Kahneman, 2011). A purely cognitive or rational paradigm cannot capture all of these drives. Thus, when trying to understand our collective behaviours, we can be “victims of reason”. To quote Montesquieu: “Le Coeur a ses raisons que la raison ne connaît point” (‘the heart has its reasons unknown to reason’) (Montesquieu, 1979 [1742]). Artificial sociality goes beyond reason, identifying the unknowns of underlying relational motives. Yes, expected profit is an important motive; but it is relational profit that matters, influenced by gut feelings and emotions such as love, hate, pride, shame, envy, loyalties. Financial profit for the individual is just a special case. As theorized eloquently by Mercier and Sperber (Mercier & Sperber, 2017), reason is used by humans for social acceptance far more than it is used for accuracy. Basically, reason is used for arguing and justifying a position in a social group to enhance influence on, and acceptance by, the group.

Watch the lake, not just the ripples

When we create policy, we tend to run from incident to incident, often forgetting to consider the patterns of path dependence linking these incidents. Causal chains of things happening today run backwards into deep history. The French revolution for instance, while seemingly showing limited impact on life nowadays, has changed and shaped the conception of the nation state and of rights that modern citizens comfortably assume to be omnipresent. Similarly, present-day individualism can be traced back to the marriage policies of the medieval catholic church (Henrich, 2020). For both these examples, it stands to reason that even older sources exist, hidden on the unbroken path of history. Across undoubted and transformative change, there is a continuity to history, especially where sociality is concerned. Sociality is about understanding the lake of human nature, in order to better anticipate the ripples on it.

Why artificial sociality?

Fully understanding sociality is vital for our survival. Artificial sociality, by showing sociality in action, can help. Here we propose a list of principles that indicate how vital it is to understand sociality better. Therefore, they justify developing artificial sociality.

  • Systems over disciplines – The earth in the Anthropocene is one system, of which key aspects are ecology, economy, and technology. All of these are known by our intellect. Their development is driven by our sociality. To understand these systems, including human sociality, we need to integrate knowledge across disciplines. This includes both natural and social sciences.
  • Multi-level systems – Grand challenges are multi-level. They are about water, climate, contagious diseases, migration, peace. They involve people and groups in systems combined of natural, institutional and economic subsystems. They have dynamics and feedback cycles, often leading to unanticipated and undesired outcomes. They may or may not be subject to policy, but they are unavoidably subject to sociality.
  • Emotions AND Rationality – In disciplines concerned with modelling human behaviour, there is a tendency to work on the assumption that “we are our brains” (Swaab & Hedley-Prole, 2014). A broad cross-disciplinary perspective, as well as life experience, make it clear that this is not really the case. Sociality has reason for breakfast: we are subject to gut feelings, we are driven, or get carried away, by emotions. Artificial sociality can bring these things to life.
  • Interaction over Individuals – The behaviour of our systems strongly hinges on the sociality of the people in them. Key issues have to do with gut feelings, emotions, trust, communication, hierarchy, group affiliation, power, politics, geopolitics. All of these rest not in the individual but emerge from social interactions
  • Explainability over black boxes – While data-driven modelling experiences great popularity, models purely based on data render limited insight into the conceptual inner workings of a social system and its meaning for a target system (i.e., the social reality it represents). Artificial sociality needs to seek a balance of theory, data, and understanding. Analysing policy without understanding interaction effects limits scientific and practical value.

For whom?

  • Interdisciplinary researchers can use artificial sociality in models for understanding their target systems.
  • Policy makers can create better ideas and policies if they are helped by plausible systemic models of the issues they face and the dynamics those issues exhibit.
  • Citizens can act as policy makers, taking their fate into their hands.
  • Designers of intelligent systems can integrate knowledge about social dynamics.

With whom?

  • All disciplines in the social and life sciences. In order to articulate artificial sociality, all disciplines that study human life can potentially contribute. This ranges across levels of integration: anthropology, artificial intelligence, behavioural biology, behavioural economics, cultural psychology, evolutionary biology, history, neurosciences, psychology, small group behaviour, social geography, social psychology, sociology, system biology…the spirit is one of consilience (Edward O Wilson, 1999).
  • Non-academic stakeholders (e.g., governments, the general public). Not only can participatory approaches help uncover hidden rules and drivers of behaviour, but also can artificial sociality be an educational tool for an enlightened society to raise its self-reflection and awareness of its inner workings.

How?

  • We recognize the integration of various disciplines’ involvements, the diversity of their respective data, theories, concepts and methods, as a challenging endeavour. In many instances we are struck by gaps between involved disciplines, and the ability to integrate data and theory in a systematic manner. Just because one theory is right does not mean that another one is wrong; often, there is complementarity, if one is willing to search for it.
  • Simulation and levels of abstraction. To this end, simulation offers the necessary capabilities, since its approach has the potential to traverse disciplines by offering broad accessibility, modelling at abstraction levels that correspond to the analytical levels within different disciplines (e.g., micro, meso and macro level in sociological research). Its unique ability to afford the systematic integration of theory and data (Tolk, 2015), deductive and inductive reasoning has rendered social simulation as a “third way of doing science” (R. Axelrod, 1997), while available computational resources allow us to explore artificial sociality at scale.
  • Creative spark. Computational simulation requires a design effort that links its various contributions into mechanisms. These constitute an original, interdisciplinary contribution. They can themselves be validated.
  • Disciplinary contributions. Social simulation is conceptually a method embedded in life sciences, complexity, and social-scientific disciplines. Each of our models creates a miniature world. These worlds need all kinds of input from various sources and disciplines.
  • Practicable outputs. Agent-based social simulation typically intends to produce practicable outputs, using theory, data and intuitions as its inputs regardless of their origin (Tolk, 2015) (Edmonds & Moss, 2005). Therefore, social simulation, in particular agent-based modelling, and artificial sociality, should institutionally be fed by many disciplines. All researchers from all disciplines are welcome.
  • Dynamics. Agent-based models are eminently suitable to help understand the dynamics of systems. They allow one to investigate unintended collective patterns arising from individual motives, intra- and intergroup dynamics. In other words, they can link disciplines at different levels of aggregation, from the individual to the society. Sensitivity analysis of these dynamics is an integral part of the method.

State of the art on sociality across disciplines

Research into human behaviour has been carried out for a long time, and in many disciplines. Such research, usable or even intended for modelling, has been picking up in recent decades. It would be presumptuous to try and give a full review of developments. Yet we believe that it is useful to give a brief overview of what is happening in various disciplines. To avoid distracting from our purpose, the details are in the appendix.

What we need from the disciplines

Given our position that every living thing that exists today, has evolved and continues to evolve, we need contributions of various types for making sense of sociality. Let us, for one moment, consider life as a game of chess. In such a model, we need to know the what, the why and the how. In our proposal, these elements will become intertwined.

  • What: the constitutive elements (chessboard, pieces); the starting position, the rules of the game (formal and etiquette).
  • Why: the motivation of the players during a game: typically, this would be “capturing the enemy king”, but other motives could occur. For instance, I might want to lose, for motivating a junior opponent, or win, for challenging him or her.
  • How: the configurations that are meaningful, and sequences of moves that can make these configurations happen. Limitations in players’ skills can reduce the space of possibilities.

These questions also obtain for sociality:

  • What: medical- and neurosciences study our constitutive elements. History, institutional economics and anthropology study what collective behaviours occur in groups of people.
  • Why: evolutionary biology studies the origins of sociality. Psychology studies human motivation today, for instance in leadership, organizational behaviour, clinical -, social- and cultural psychology. Ethology does the same for non-human creatures.
  • How: Sociology tends to describe the how of sociality, for instance patterns, their causes and their sequences. Computational branches in biology, economics, and sociology construct artificial worlds. Simulation gaming, and experimental economics do the same with real people, in artificial incentive structures.

For computational modelling we will need input on all three of these questions. The models will require

  • A “what”: agents in an environment.
  • A “why”: motivation for the agents: drives, urges, goals.
  • A “how”: perceptions and actions for the agents, and coordination of these across space and time. This will lead to emergent pattern.

The three questions are really highly intertwined; we take them apart only for the sake of exposition. Also, the emergent pattern of one branch of science, or of a simulation model, can become the input, taken as given, of another. For instance, some models could investigate the emergence of institutions, norms, or culture; others could use such concepts as input variables.

The take-home message of this section is that our modelling efforts will be best served by an eclectic mode that draws from a broad variety of sciences.

What the disciplines tell us

We shall now attempt a synthesis of work on sociality across disciplines presented in the appendix that are important for the research agenda of artificial sociality. To structure it, we stick to our what, why and how questions. Admittedly, our synthesis is partial; this is done for the sake of purposefulness, not because other perspectives could not have merit.

What

Sociality is not a human invention. It is absolutely central to life on earth, and has been since billions of years, in an unbroken chain of reproduction. Sociality has served to preserve homeostasis in populations, enabling some to reproduce (Damasio, 2018). It is as old as monocellular organisms, many of which are known to coordinate their behaviours in response to external stimuli, particularly at the service of reproduction. Human sociality is special in a few ways (Henrich, 2016). We coordinate in many ways with many people we do not personally know. For achieving joint action, we have basically two mechanisms. In evolutionary terms these are prestige and dominance (Henrich, 2016).  In sociological terms:  status and power (Theodore D. Kemper, 2017). Also, groups of followers are able to curtail the power of leaders. For these functions we evolved intense emotional lives (J. E. Turner, 2007). Emotions are the proximate indicators of our sociality that our organisms provide to us. We’ll return to these issues under “how”.

Selective pressures do not just operate between individuals, but at many levels. There is selective pressure between individuals, human groups, forms of coordination, even ideas. Models can concentrate on any of these levels.

Why

Sociality, in terms of status and power motives in multiple, changing groups, and attending feelings and emotions, is necessary for solving coordination problems, e.g. dividing food, reproducing, bringing up children, or avoiding traffic congestion; and for solving collective action problems and social dilemmas, e.g. selecting a leader, disposing of a dysfunctional leader, or distributing resources across the citizens of a country. This holds in small groups and families with informal social bonds as well as in large groups or societies that rely on formal, depersonalised interaction patterns. Without sociality there can be neither Gemeinschaft (community) nor Gesellschaft (society). Sociality shapes our moral sense.

How

In Humans, sociality develops very early in life, preceding speech and walking. It requires intense care, play, and education during many years; we are a neotenous species, remaining juvenile for many years and even keeping some brain plasticity during adult life. For a baby, the organism has precedence. After just a few months, giving and conferring status becomes important. Between 11 and 19 months, power use develops (Eliot, 2009). During childhood, the social world grows, and various reference groups become distinguished. We learn the dynamics of prestige / status giving and claiming. At puberty, sociality more or less plateaus; just like we speak with the accent of our childhood, we act with its culture. Our hormonal systems are aligned with the dual nature of prestige / status and dominance / power; more on this in the appendix.

This phenomenon of a flexible beginning then stable existence also holds for groups of people. Once formed, societies, groups, organizations and companies, have cultures that tend to remain stable over time, despite many perturbations (Beugelsdijk & Welzel, 2018; G. Hofstede et al., 2010).

Sociality happens. Every action in which several people are present or imagined provides an instance to mutually imprint sociality through status-power dynamics in a world of groups. This ranges from glances and nonverbal involuntary movements, to explicit verbal communication, to social media posts and likes, to elaborate rituals involving prestige and social roles, to coercive acts involving life and death. All of these constitute as many claims for, and accord or refusals of, status; and some of them include power moves.

Groups in society are endlessly variable. They change at various timescales, from life-long to context-dependent and ephemeral. They can be nested or overlapping. Their salience is socially and situationally determined.

Collective results of social acts need not be intended. Much of our societies’ behaviour largely emerges unplanned. A few frequent, archetypical patterns can often be seen in this unplanned system-level behaviour. Agent-based modelling is privileged as a method by allowing to generate these unplanned patterns.

Key theories

There is such a wealth of theoretical work in so many disciplines that even the brief overview above may seem a bit unorganized. Therefore we briefly mention a few of the theories that we’ll most use in our proposal.

  • Kemper’s status – power – reference-group theory of relations. This comprehensive sociological theory also touches on neurobiology and psychology. This makes it compatible with evolutionary theories of human sociality. The appendix has a more elaborate treatment.
  • Heise’s Affect Control Theory (ACT). This theory shares a lot of elements with Kemper’s status – power dynamics but is targeted to small group interactions.
  • Tajfel & Turner’s Social Identity Approach (SIA). This theory elaborates on elements of group and intergroup dynamics, somewhat similar to Kemper’s reference groups.

Work to do in artificial sociality

The synthesis above suggests that sociality is about things that we do, and things that happen between people, in any of the contexts of their lives. Artificial sociality can reproduce sociality using modelling techniques that make life happen: “generative social science” (Epstein, 2006), or, with a newer word, computational social simulation. The task for artificial sociality is first and foremost a modelling task with the ambition to understand sociality-in-action better.

Principles

Ontologically, our perspective is one of consilience. Since there is only one world, findings that align across different sciences are particularly interesting to use in models of sociality. This is the case with the match between neurobiology, emotions, and the status-power theory of relations discussed in the appendix, for example.

Vocabulary

One of our tasks is to generate better understanding and a common vocabulary. At present, many modellers criss-cross the same conceptual space, but with different maps from different reference disciplines.

Open world hypothesis

In order to be able to talk with one another and build shared vocabulary, researchers should maintain an open world hypothesis: if your model differs from mine, then we can talk. What is the difference, is it really a difference, what does that allow or disable? Such discussion allows us to enrich our ontology. It is unrealistic, anyway, to expect everyone to agree. Artificial sociality is heavily loaded with worldview, and people disagree on worldviews. This is actually something that artificial society should help explain; unfortunately, we can predict that such an explanation will not please everyone.

Realms to model

Our sociality operates in a world with non-social elements such as space, time, objects. On a scale from content-based to relational, we can distinguish four realms that need to be modelled.

  1. This means the bio-physical and the institutional world, divorced from what people might feel about it.
  2. Cognitions about content. This includes knowledge, opinions, norms, and values that influences our perspectives on the content realm. They are partially conscious, the less so the more they are shared (and therefore cultural). This realm binds the relational to the non-relational world.
  3. Cognitions about relations. We have ideas about the status (“social importance”) and power that others have, about our own status and power in groups. These are normally unconscious.
  4. Cognitions about our own organism. This includes all kinds of organismic feelings, again often not fully conscious, and may include meta cognitions (e.g. “thinking about thinking”). Emotions link the organism with the relational world, often unconsciously. For instance, an insult is an attack on our status, and may bring the blood to our cheeks.

Artificial sociality requires considering all these elements. To which extent we consider each of them can be case-dependent. Depending on the application, some might have to be further elaborated. It is possible to model only one or several of these realms. For instance, Kemper’s theory posits the organism as one of the relevant reference groups, merging 3. and 4. Hofstede’s GRASP world has only sociality (3.) and no content (Gert Jan Hofstede & Liu, 2020). The general-purpose link from emotion as coherent dynamic social meaning, to content as objects and actions in institutional frames proposed in BayesACT may provide a link between (1.), (2.) and (3.) (Schröder, Hoey, & Rogers, 2016). Ultimately all of the realms will be needed in combination.

Theories and realms

Theories from the social sciences tend to concentrate on a subset of these realms. Table 1 indicates this.

Table 1: theories and realms to model (legend: from – not included, … to +++ central to this theory)

Theory modelled realms
  Content Cognitions
    on content on relations on organism
Affect Control Theory (Heise, 2013) + ++
Reasoned action approach (Fishbein & Ajzen, 2010) + +
Social identity approach (H Tajfel & Turner, 1986) + ++ ++
Status-power theory of relations (Theodore D. Kemper, 2017) +++ +
BayesACT (Schroeder et al., 2016) + + ++

Sources: theory, data, and experience

Models are integration devices, built from a variety of sources. Theory, data, and real-world experience all contribute to the usefulness of models that include artificial sociality (figure 1). The figure positions computational social simulation as a meeting place of these three elements. Different mixtures are possible, depending on the aim of modelling (Edmonds et al., 2019). Models range from purely theory-based ones that can illustrate core concepts, to models developed in participation with stakeholders that reflect real life, to highly complicated, data-fed models that can describe existing data or predict (generalize to) future measurements.

Artificial sociality as we propose it is, in the first instance, a theoretical concept. We believe that it has strong face validity in real life. This is by virtue of the empirical basis and broad scope of the theories involved. Integrating our concepts with data, for instance the never-ending stream of social media data, is a major challenge for the coming years.

Manif - Picture 1

Figure 1. Social simulation as a meeting place of theory, data and real life (Gert Jan Hofstede, 2018).

Model architectures

In artificial sociality we cannot get away with ideas only. Implementations are also needed, and functional computer code. In computer code, all the capabilities of our virtual world and of the agents that populate them, have to be unambiguously specified. This raises the issue of architecture. For instance, do agents have a body, a brain, and a soul? Do groups have common agency, or is that delegated to individuals? If the world is spatial, do we have instinctive reactions to moving objects? Is there “fast and slow” thinking as per many author’s writings  (e.g. (Kahneman, 2011) (Zhu & Thagard, 2002) (Glöckner & Witteman, 2010)?

Currently, a thousand flowers are blooming in the computational modelling of human behaviour. This is a good way to search. We believe that one architecture will not cover all needs; in all likelihood many streams of research will dry up, and we’ll be left with a limited number of rather general-purpose architectures for different purposes. Many existing models and architectures deserve to be taken into account.

State of the art

Artificial sociality, by design, is integrative across its contributing disciplines. Scientists have tried to integrate research on human behaviour and society across disciplines as long as we know. This has, however, become progressively harder as disciplines have branched. Aristotle was still a polymath, but today this is hardly possible any more.

Some attempts that are meaningful for artificial sociality in our view merit mention here.

Conte and Gilbert and their legacy

In social simulation, the concept of sociality was introduced in the nineteen nineties. Psychological computer scientists Kathleen Carley and Allen Newell published their extensive essay “The Nature of the Social Agent” in 1994, in which they proposed that compared with “omniscient” economic agents, social agents have more limited processing capabilities, but a richer social environment. They will turn to socio-cultural clues instead of raw data (Carley & Newell, 1994). Cognitive psychologist Rosaria Conte and sociologist Nigel Gilbert are founders of the notion of “artificial societies” (Gilbert & Conte, 1995). They set out to define artificial sociality as a challenge for computational social simulation. Their reflections were crowded out of the public eye by the advent of the Web, and the increasing ubiquity of data as sources for modelling. Yet computational social modelling has remained focused on human social behaviour.

Flache et al. in a position paper explicitly dedicate their work to Conte, who died prematurely in 2016 (Flache et al., 2017). They plead for more research on the question that Robert Axelrod posed in 1997: “If people tend to become more alike in their beliefs, attitudes, and behavior when they interact, why do not all such differences eventually disappear?“(Robert Axelrod, 1997). Flache et al. discuss several models, the currency of which is opinions.

Jager also builds on a statement by Conte when he pleads for “EROS”, or more attention to social psychology in computational social simulation (Wander Jager, 2017). He reviews a number of theories that have been used in social simulation, none of which includes emotions. The most generic of these might be Ajzen’s Theory of Planned Behaviour (the most recent version of which the author calls Reasoned Action Approach (Fishbein & Ajzen, 2010).

Other efforts

Work on active inference and a hierarchical (deep) Bayesian probabilistic view of the mind has led to more integrative models including of interpersonal inference (Moutoussis, Trujillo-Barreto, El-Deredy, Dolan, & Friston, 2014) and culture (Veissière, Constant, Ramstead, Friston, & Kirmayer, 2020). These models consider a long-standing view of human intelligence as being largely predictive rather than descriptive. That is, the mind is set up to seek information, and to interpret evidence, in ways that confirm prior beliefs.

A mid-range approach to sociality is taken by Shults and colleagues. They take domain-directed social scientific theory and develop agent-based models with agents embodying the theory. These tend to contain instantiated sociality elements such as fear. This includes terror management theory (Shults, Lane, et al., 2018) and intergroup dynamics under anxiety (Shults, Gore, et al., 2018).

Some computational modellers have built models of human behaviour suiting their purpose. This includes empathic agents, care robots, and the military. These models include some sociality, without necessarily using that term. Space forbids to deal with them at length. Interesting pointers are (Balke & Gilbert, 2014; Schlüter et al., 2017).

Consumat architecture

An example of an architecture that is appealing because of its simplicity, while including both content and a bit of sociality, is the Consumat (Wander Jager & Janssen, 2012; Wander  Jager, Janssen, & Vlek, 1999). Consumats live in one group or network, not necessarily but possibly in a spatial world, in which they have repeated decisions to take about which they are more or less certain. In addition, they are more or less “happy” based on the outcome of their previous decisions. “Happiness” and “uncertainty” combined determine what they will do: repeat, imitate someone else, deliberate on content issues, or do a more elaborate social comparison. The currency of “happiness” is not further specified, making the Consumat model quite flexible. Embedding fundamental concepts of sociality (e.g., allusions to reference groups and uncertainty), Consumat takes the individual as a unit of concern, rendering it a flexible starting point for richer developments of artificial sociality that have a stronger emphasis on the structure the agent is embedded in. It has found quite a few applications. A more elaborate follow-up effort on Consumat called Humat is now being developed into publications.

FAtiMA

An engineering approach to sociality with considerable fidelity is FAtiMA (Mascarenhas et al., 2021). This open-source toolkit for social agents and robots includes prestige / status dynamics and social emotions. Status dynamics are called “social importance” in FAtiMA (Mascarenhas, Prada, Paiva, & Hofstede, 2013).

GRASP

The GRASP meta-model for sociality (Gert Jan Hofstede, 2019) is an attempt at capturing the bare essentials of sociality: Groups, Rituals, Affiliation, Status, and Power. GRASP is deliberately content-free. Its relational currency is status and power. It is based on the works of Kemper mentioned here, and on Hofstede’s and Minkov’s work on national cultures. Culture modifies the rules of the status-power action choices (G. Hofstede et al., 2010; Gert Jan  Hofstede & Liu, 2019). A showcase model using GRASP, GRASP world (Gert Jan  Hofstede & Liu, 2019; Gert Jan Hofstede & Liu, 2020), pictures the longevity of social groups based on the ease with which agents can leave a group in which they are subjected to power or receive insufficient status. The resulting patterns resemble social dynamics in different cultural environments.

Contextual Action Framework (CAFCA)

The CAFCA meta-model (figure 2) allows to disentangle levels of sociality and context. It was created to add on to Homo economicus models, and allows to classify existing model ontologies. Sociality implies moving to the bottom right of the model. CAFCA shows how far we still have to travel. One could extend it: a relational perspective is not included so far, nor is a multi-group world.

Manif - Picture 2

Figure 2: CAFCA, the Contextual Action Framework (Elsenbroich & Verhagen, 2016).

We can conclude that in response to Conte’s and Gilbert’s challenge, explicit opinions have received a lot of attention in computational social simulation, but emotions and feelings have not. We believe that this still leaves some phenomena unexplained. Opinions need not always be taken at face value, but can be manifestations of social feelings and emotions, e.g. love for one’s group. Computational agents are still often “autistic”, whereas real people have sociality at their core (Dignum, Hofstede, & Prada, 2014). Sociality can give them “biases”, “perspectives”, or “relational rationality” (Gert Jan Hofstede, Jonker, Verwaart, & Yorke-Smith, 2019) that can be derived from various theories.

Bayesian Affect Control Theory (BayesACT)

BayesACT is a dual process model that unifies decision theoretic deliberative reasoning with intuitive reasoning based on shared cultural affective meanings in a single Bayesian sequential model (Hoey, Schröder, & Alhothali, 2016; Schröder et al., 2016).  Agents constructed according to this unified model are motivated by a combination of affective alignment (intuitive) and decision theoretic reasoning (deliberative), trading the two off as a function of the uncertainty or unpredictability of the situation. The model also provides a theoretical bridge between decision-making research and sociological symbolic interactionism. Bayes ACT is a promising new type of dual process model that explicitly and optimally (in the Bayesian sense) trades off motivation, action, beliefs and utility, and integrates cultural knowledge and norms with individual (rational) decision making processes. Hoey, et al. (in publication: Jesse Hoey, Neil J. MacKinnon, and Tobias Schroeder. Denotative and Connotative Management of Uncertainty: A Computational Dual-Process Model. To appear in Judgement and Decision Making, 16 (2), March 2021.2021) have shown how a component of the model is sufficient to account for some aspects of classic cognitive biases about fairness and dissonance, and have outlined how this new theory relates to parallel constraint satisfaction models.

Proposal: a relational world

We now put forward our own proposal for an architecture, not because we believe this is the only way to go, but in order to give an example of where a more radical take on sociality can lead.

Theory base

Theory versus data

We assume that data provide no more than a partial perspective on the phenomenon they are captured from. Only in concert with a theoretical concept will they attain meaning. For instance, consider today’s vast quantities of data on social media usage. Our communication on social media does not reflect all of our relations. Linking data and Kemper’s theory, we presume that people will use social media to claim status (e.g. show pictures of successes and important rituals), to confer status (e.g. like and follow others), and to use power (e.g. insult high-status others). There are also many relational motives that will not show in social media. People will hide shameful actions (e.g. failing, being exposed); they will protect some of their behaviours from some of their reference groups (e.g. their parents or spouses). People may fear the power of their own government, and stay away from some social media. Often people will seek information and interpret evidence in a way that confirms group acceptance, rather than in a way that confirms facts (Mercier and Sperber, 2009). Which members of a society go on which social media, and just how they select which things to show and which not to, are dependent upon relational dynamics that the data cannot show without help from theory. A theory is needed about the “why” of behaviour.

Building blocks: Complicated vs complex

We are aware of the tension between complicatedness of model structure and complexity of model outcomes (Sun et al., 2016). According to these authors, complex behaviour can be represented either by a model with few simple primitives, or by a very elaborate model. Our intuition is that a bottom-up approach with strong theory base and simple ontology is most promising. An analogy can illustrate this (figure 3). A complicated model architecture tends to be difficult to adapt. The price to pay for a simple, adaptive architecture is abstraction. To build a valid, versatile model with few primitives, just a few types of building blocks could suffice; only, one needs a great many of them.

Manif - Picture 3

Figure 3: giraffe models in Lego. From left to right: 1) Model that is valid but made of a complicated piece; 2) simple model with just 5 different rectangular shapes; 3) more complicated model with 15 different shapes of varied form; 4) simple model with few shapes but many pieces.

From theory to model

Implementing a theory from social science in a computational model is by no means straightforward. Typically, theories leave many elements unspecified. Model designers have to fill the gaps. For instance, the Social Identity Approach (SIA) has been used in computational modelling. It models agents as enacting a particular social role or identity that is context (institutionally) dependent and emotionally meaningful. From reviewing such papers, we learned how difficult it is to model a “complete” social world. We failed to find a single model yet that models SIA to its richness, and can actually be replicated. To accommodate this, a toolbox approach is used by the network project SIAM (SIAM: Social Identity in Agent-based Models, https://www.siam-network.online/), offering a set of formalizations that can then be specified for specific purposes/aims. Still, this is challenging. We believe that interdisciplinary work yields substantial benefits here.

Which theory

In selecting theories to work with, a thousand flowers can blossom. In our case, for creating models with relational agents that have simple ontology but great range, we believe that Kemper’s work, and SIA mentioned before could provide the Lego blocks. Both place individuals (called “agents” in what follows) in a rich world consisting of many groups with salience mechanisms. SIA gives agents both an individual and social identities. Kemper has no self but only reference groups, that is, groups existing in the mind of an individual, not necessarily in the outside world. For Kemper, the organism with its needs and urges is one reference group. Crucially, Kemper additionally gives the agents status and power motives; we believe this to be crucial for social agents. In SIA, agents act upon motives too (such as the need for positive distinctness and self-esteem), while status is achieved through comparison with outgroups. Heise’s Affect Control Theory (ACT) is similar to Kemper here, and more articulate for describing verbal communication; but it works for single groups only. Efforts at broadening ACT to multiple, overlapping and interacting groups, are currently underway  (Hoey & Schröder, 2015).

In what follows we mainly lean on our interpretation of Kemper, as the most generic and simplest of these theories.

What, how and why

The “what” of our relational world consists of individuals and groups. A person can belong to several groups, and not everyone necessarily has the same shared belief about who belongs to which group. Furthermore, there will be an environment with certain affordances; we will come to this later.

Basic rules for the “why” are:

  • What individuals do, is determined by the groups to which they affiliate. Those groups will act as reference groups.
  • People’s choices depend on what they believe their reference groups want them to do.
  • These beliefs are about status and power; they can be about individuals, or about groups.
  • Status beliefs are about the status worthiness of actions, people, and groups; and about appropriate ways of claiming and conferring status.
  • Power beliefs are about the power of people and groups; and about appropriate ways of using power.
  • For obtaining what they want, people can choose between status tactics and power tactics.
  • Status tactics involve claiming and conferring status. As long as conferrals exceed claims, they tend to be pleasant, and create trust. If claims exceed conferrals, people will feel insulted, and power tactics will be used.
  • Power tactics involve coercion and deceit, and tend to lead to resentment and repercussions, except where power is perceived as legitimate.
  • In practice, power use is often couched as status conferral; misunderstandings can also occur.

A model with these primitives would qualify as a GRASP model. The fine print of all of these rules – what is considered appropriate for whom, and in what circumstances – depends upon culture (Gert Jan Hofstede, 2013). This implies that the actual status-power game is quite complex and varied, even though there are few primitives.

The “How” would depend on the context, because the primitives need to be bound to instantiations. Here, the four “elementary forms of sociality” of anthropologist Alan Page Fiske could be useful. This may require a bit of introduction. Fiske, having carried out field studies in various civilizations, came up with four “ elementary forms of sociality” (Fiske, 1992). These are: communal sharing, authority ranking, equality matching and market pricing. Fiske aims with these elementary types to bring unity to the myriad of psychological theories. He says people use these four structures when they “transfer things”, and interestingly, they correspond with four sales in which “things” can be compared: nominal, ordinal, interval and ratio. He comes up with a wide range of issues and situations where the four forms obtain. These are not mutually exclusive: we might use communal sharing in one setting, authority ranking in another, and market pricing in yet another. The balance will depend on the issue or group and on culture.

If we assume that the thing to exchange is social importance or, in Kemper’s sense, status, then the following obtains:

  • Under communal sharing, it is the group, not the individual, that is the unit of status accordance, claiming, and worthiness
  • Under authority ranking, there is a clear hierarchy in social importance, and status accords, as well as power exertion, are asymmetric based on ascription. “Quod licet Iovi not licet bovi” (“What the god Jupiter may do, a cow may not”).
  • Under equality matching, each individual or group is equally worthy, should claim and be accorded the same amount of status.
  • Under market pricing, there is no need for a moral stance, since the market decides.

The likelihood of these four forms is obviously culture-related. In particular, two of Hofstede’s dimensions seem relevant (see figure 4). These forms could directly be used as model mechanisms, or their emergence in agents could be studied based on Hofstede “software of the mind”.

Manif - Picture 4

Figure 4: Likelihood of Fiske’s elementary forms (quadrants) across Hofstede’s dimensions of culture (axes). Market pricing is indifferent to power distance.

Readers are invited to consider current events in their lives, or in the political arena, through a relational lens. Once one distinguishes the silent voice of reference groups, and the dynamics of mutual status and power use, one can also see historical continuity within the relational lives of people, groups, companies, and nations.

Proposed architecture

Figure 4 shows what we propose are key ingredients of our relational architectures for artificial sociality. There is a correspondence between the concepts in the four columns, with the left column reflecting the micro level of individual operation on the level of the organism to operationalise emotions and related individual-centred concepts. To our mind – and put forth in this paper –  the most universal Lego blocks of artificial sociality are relational. In figure 5 we use Tönnies’ term Gemeinschaft for this. Figure 5 shows Kemper’s concepts of status, power and reference groups; but alternatives with similar relational content could be chosen. This relational column is always required. Depending on the application, the concepts in one or more of the other columns are needed. If they are included, they have to be mapped onto them, making status, power and reference groups the basic operational concepts for driving the model’s dynamics. For instance, emphatic agents need to feel and communicate emotions. Social robots need proxemics, i.e. to know the emotional impact of closeness, motion and posture; models that explain phenomena such as tribalism require individual-level concepts in addition to relational conceptions. Speaking to scale, simulations that model social complexity at the societal level, and are concerned with effects of policies require Gesellschaft concepts such as norms and institutions.

Examples for such models include the reaction to imposed behavioural constraints as part of the Covid-19 countermeasures employed throughout nation states – with vastly varying responses based on social structure and influence (expressed in the relational column) and individual motivations of various kinds, including perceived challenges to liberty, economic well-being, etc.  Whatever the variable configuration of sociality elements, we require a conceptual mapping to the physical world, such as the operationalisation in status and power in currencies relevant to the society of concern (e.g., status symbols).

Figure 5 is organised into columns. The leftmost column is organismic on an objective sense, but subjectively perceived. The middle two are intersubjective, continually construed by people in interactions, although things in the Gesellschaft column tend to be perceived by many as objective (Searle, 1995).  The rightmost column is about the physical world, considered objective but often perceived from a subjective, or rather intersubjective, stance.

The impact of this position is that a direct mapping from the physical world to emotions, or from money to behaviour, will not yield versatile models. Data based models without a strong social model of sound theoretical basis using e.g. financial actions to predict future economic behaviour, or past voting to predict future voting, might accommodate specific application cases, but their range of application across cases and time will be limited. More importantly, such models lack the explanatory potential that conceptual models of sociality can offer.

Manif - Picture 5

Figure 5: building block concepts for artificial sociality.

Conclusion

This position paper argues for a biological, relational turn in artificial human sociality. Such a turn will lay a foundation that can reconcile case-specific or discipline-specific model ontologies.

Artificial sociality has the potential to greatly enhance all knowledge technologies that impinge on the social world, including e.g. social robotics and body-worn AI devices.

In this paper we mainly aim to increase the usefulness of computational models of socio- ecological, -economical and -technical systems by tackling their social aspects on a par with the other ones, in a foundational, thorough way.

Many theories, in a great many disciplines, could possibly be used in constructing ontologies for artificial sociality. We provide some pointers and examples. We also present ideas for a “relational world” that could inspire modellers.

There is a lot of work to do.

Appendix: contributions to sociality from various disciplines

The appendix is sorted, admittedly somewhat arbitrarily, according to whether a field of research focuses more on the “What”, the “Why” or the “How” of behaviour. Within those three, the order is alphabetic.

Mainly the “what”

Anthropology

Computational simulations have been made of historic civilizations. In these, simulated populations live in a simulated environment. This requires a mix of historical data and assumptions, in particular about resources and / or social drives. If the various hypotheses that are implemented in the models hold, then the simulations could throw light on historical contingencies, or even reproduce the actual history. A famous example is the “artificial Anasazi” model by Epstein that ”replays” the rise and fall of the Anasazi civilization (Epstein & Axtell, 1996). The agents in this model have no sociality, but are constrained by resources. A recent example is e.g. a model of island colonization based on the concept of gregariousness (Fajardo, Hofstede, Vries, Kramer, & Bernal, 2020).

Another contribution from Anthropology is to study typical patterns of human social organization. The work of Alan Page Fiske is interesting in this respect. Fiske’s, four “ elementary forms of sociality” were mentioned before, in the context of figure 4 (Fiske, 1992). To repeat: communal sharing, authority ranking, equality matching and market pricing.

Institutional Economics

A fundamental feature of humans is our ability to coordinate – at scale, that is. Humans can coordinate on group, societal and global level, both towards shared interests (e.g., emergence of economic and personal liberties in the French revolution; international treaties such as the Whaling convention), but, at times, they also contradict those (e.g., climate change, e.g., (Shivakoti, Janssen, & Chhetri, 2019)). In an attempt to identify the cause of prosperity or demise of societies, New Institutional Economics (North, 1990) integrate the many strands of human behaviour – including the ones outlined above. Rooted in our biology and manifested in our psychology, as humans we possess “minds as social institutions” (Castelfranchi, 2014) that continuously exercise coordination activities. Institutions, here understood as the “integrated systems of rules that structure social interactions” (Hodgson, 2015), or simply “rules of the game” (North, 1990) are the catalysts. They include sophisticated constructs such as written contracts and courts, enabling cooperation at scale (Milgrom, North, & Weingast, 1990); (North, Wallis, & Weingast, 2009), but also informal arrangements for resource governance  (Ostrom, 1990), pointing to opportunities to address social dilemmas, such as the Tragedy of the Commons (Hardin, 1968).

Neurobiology and endocrinology

A model of sociality is more valid to the extent that it fits the evidence about our bodies. This includes the brain of course, with e.g. its mirror neurons that are a vehicle for empathy, but also older physiological systems such as the sympathetic (fear and anger) and parasympathetic (well-being) nerve system and the digestive system (all kinds of impulses, e.g. mediated by our gut microbiome). The recent semantic pointer theory of emotions (Kajic, Schroeder, Stewart, & Thagard, 2019) capitalizes on the mathematical apparatus of Affect Control Theory discussed above to embed the sociality of affective experience into neurobiological mechanisms through a neurocomputational simulation model.

Tönnies’ Gemeinschaft and Gesellschaft

A fundamental sociological theme that structures the arena of social behaviour is the dialectic between different forms of social organisation that represent anchor points for an integrated artificial sociality, namely Gemeinschaft (community) and Gesellschaft (society), introduced by (Tönnies, 1963 [1887]), and subsequently popularised by Weber. This distinction was part of an extended debate in early sociology about the core concepts of societal structure, where Gemeinschaft captures the characterisation of social ties observable in a social setting as primarily based on personal relationships, enacted roles and associated values as present in prototypical peasant societies prevalent at the time. Any interaction in those societies was based on what Tönnies referred to as natural will (“Wesenwille”) exhibited by members. Gesellschaft, in contrast, reflects the depersonalised counterpart in which individuals act in indirect form based on assigned roles, formal rules, processes and values, stereotypical structures associated with urban societies. Fittingly, Tönnies characterised motivations for any such interaction driven by rational will (“Kürwille”) encoded in the role individuals exhibit.

Likened to Durkheim’s differentiation between mechanical vs. organic solidarity (Durkheim, 1984), the concepts are stereotypical for the themes and worldviews that structured debate at the time. Instead of drawing on the particularities of either variant of this duality[1], they bear essentials that still apply to group dynamics found in modern societies.

Where behaviour is structured and planned, leading agents to create rules, react to imposed policy or enforce such, the representation of socio-institutional dynamics are of concern. While building and relying on concepts such as status and roles identified in the Gemeinschaft conception, concepts such as rules and governance structures extend beyond neurobiological and psychological bases of group formation, but are the mechanisms that lead to depersonalised coordination structures characteristic for the Gesellschaft interpretation of society. Doing so, models of artificial sociality can resemble the characteristics of real-world societies, including “growing” the complexity arising from systemic interdependencies of actors, roles and resources, and reflect the non-linearity of behavioural outcomes we can observe at scale.

Mainly the “why”

Behavioural biology

Behavioural biology has studied social behaviour of all kinds of animal, including those that resemble us very much. Frans de Waal stands out for his extensive studies about dominance, politics, reconciliation and pro-sociality among primate (Waal, 2009). Chimpanzees and bonobos in particular can teach us a lot about the sociality of Homo sapiens. Like chimps, we have bands of males fighting one another and dominating females. Like bonobos, we have female solidarity, social sexuality, and male reluctance to use their physical superiority.

Evolutionary biology

Our stress on the deep historic continuity of life in an unbroken chain of reproduction under variation implies that we see evolutionary biology as the mother of the social sciences. Our perspective owes to the work of authors such as De Waal, who concluded his discussion of morality in all kinds of animals, particularly primates, as follows: “We seem to be reaching a point at which science can wrest morality from the hands of philosophers” (Waal, 1996).

Evolutionary psychologist Turner argued that emotions have become much more important in humans than in other species, because we do not limit our contacts to either one predictable set of others, or an anonymous mass  (J. E. Turner, 2007). We needed to find a relational compass. Our expressive faces and gestures, and our open faces, developed for that purpose.

Clinical psychology

Clinical psychologist Abraham Maslow gave us the famous model of human needs, by observing his patients and seeing an overarching pattern (Maslow, 1970). This model is antithetical to Homo economicus. The problem with it is that it is hard to operationalise. A more proximate concept in human drives is emotions (Frijda, 1986). Emotions have been used quite a bit in computational social simulation, e.g. the cognitive synthesis of emotions in the OCC model (Ortony, Clore, & Collins, 1998). This has been used as underpinning of empathic computational agents (Dias, Mascarenhas, & Paiva, 2016).

Leadership psychology

The psychology of leadership naturally touches upon sociality. For instance, Van Vugt et al assert “leadership has been a powerful force in the biological and cultural evolution of human sociality” (Van Vugt & von Rueden, 2020). Human groups faced with problems of coordination and collective action turn to leadership for achieving collective agency. Different contexts have led to different leadership styles.  Leaders can base their role on dominance (coercion), or on prestige (voluntary deference), and people still turn to more dominant leaders in times of stress.

Cultural psychology

Cultural psychology adds a comparative perspective to leadership psychology, showing that leadership styles and follower styles are co-dependent and have historical continuity across generations (G. Hofstede et al., 2010). It is also a discipline in its own right, and it shows how all of social psychology is in fact culture-dependent (Smith, Bond, & Kagitcibasi, 2006).

Social Psychology: Social Identity approach

A set of theories useful for modelling group behaviour and intergroup relations are presented in the Social Identity approach (SIA). SIA refers to the combination of Social Identity Theory (H Tajfel, 1982; H Tajfel & Turner, 1986) and Self-Categorization Theory (Reicher, Spears, & Haslam, 2010; J. C. Turner, Hogg, Oakes, Reicher, & Wetherell, 1987).

SIA proposes that social identification is a fundamental basis for collective behaviour, as people derive a significant part of their concept of self from the social groups they belong to (H. Tajfel, 1978; J. C. Turner et al., 1987). When a person’s identity as a group member becomes salient in a particular context, this affects who is seen as being an ingroup member versus someone outside of the group. When a social identity is salient, group membership becomes an important factor for individual beliefs and behaviour – what is important for the group becomes important for the individual. Moreover, groups have their own social norms and expected behaviours. For instance, thinking as members of collectives changes helping behaviour, as we are more likely to provide help to ingroup members (Levine, Prosser, Evans, & Reicher, 2005).

We deem SIA particularly well suited to model sociality, as it spans from the why (motives) to the how (e.g., saliency of social identities that impact on behavior, dynamics between groups), and connects the micro level of individuals with the macro level of groups, groups in groups, all the way up to societies. SIA has been used in social simulation to address diverse research questions from Sociology, opinion dynamics, Environmental Sciences and more (for two qualitative reviews see (Kopecky, Bos, & Greenberg, 2010; Scholz, Eberhard, Ostrowski, & Wijermans, 2021 (in press)). However, up to now there is no standard formalization, and formalizations found vary widely.

Mainly the “how”

Computational biology

Simulations include work of emergent patterns occurring in swarms and fish schools, based on simple positioning rules that fish and birds use while moving. A seminal contribution in the field of behavioural biology was made by the DomWorld model that showed, among other things, how spatial configurations in primate groups could emerge from dominance interactions (Charlotte K. Hemelrijk, 2000; Charlotte K Hemelrijk, 2011). Here, the contribution of a behavioural theory involving dominance and fear was crucial. The swarm and Domworld models also are instances of agent-based models. i.e. computational simulation models in which individuals live in a spatial world. These models have heterogeneity and path dependence, just like real historical developments.

Computational sociology

Sociologists have been at the origin of artificial sociality – avant la lettre. In 1971, mathematical sociologists Sakoda and Schelling published models showing self-organization in societies resulting in unintended, but robust collective patterns. The history of these models was recently traced by (Hegselmann, 2017). Computational sociologists have followed in their tracks, helped by the advent of simulation software (Hegselmann & Flache, 1998) (Deffuant, Carletti, & Huet, 2013). Recent computational models of this kind include emotions and their spread (Schweitzer & Garcia, 2010).

Development psychology

Developmental psychologists show how, during infancy, childhood and puberty, people acquire a more varied concept of the social world. For instance, rough-and-tumble play peaks in boys at the onset of adolescence (G.J. Hofstede, Dignum, Prada, Student, & Vanhée, 2015); among Dutch adolescents a nested set of reference groups develops, and girls are more prosocial overall than boys in a dictator game (Groep, Zanolie, & Crone, 2019; Güroglu, Bos, & Crone, 2014).

Economics

Economics came up with the concept of the profit-maximizing Homo economicus, useful as a standard with which to compare actual human behaviour, in contexts where “profit” can be defined. Not all contexts are like that, which is why behavioural economist Richard Thaler predicted that “Homo economicus will become more emotional” (Thaler, 2000). Experiments in behavioural economics and game theory have now shown that people have relational motives that moderate their actions, and often lead to “non-rational” behaviour that may be heavily culturally biased (Henrich et al., 2005). This is an important finding, because if the pleasantly simple Homo economicus model does not hold in reality, then what is the alternative?

Human motivation: Heise’s Affect Control Theory

Sociologists have also studied universals of human social motivations, either in small groups (Heise, 2013) or more generically (Theodore D. Kemper, 2017).

Heise posited Affect Control Theory, a relational theory on how people in small groups maintain relations. According to Affect Control Theory, every concept has not only a denotative meaning but also an affective meaning, or connotation, that varies along three dimensions:[1] evaluation – goodness versus badness, potency – powerfulness versus powerlessness, and activity – liveliness versus torpidity. His work has recently been elaborated upon in social simulation (Heise, 2013) and combined with decision theoretic (rational) reasoning models (Hoey et al., 2018).

Human motivation: Kemper’s relational world

Kemper, who worked with Heise sometimes, developed a model of human drives that is similar but less operationalized, and wider in scope. It distinguishes two major dimensions, derived empirically, having to do with coerced versus voluntary compliance: power, and status. Kemper’s word “status” is thus not a measure of power, but in a sense the opposite: it is a measure of not needing power. It has been dubbed “social importance” which captures the meaning but is lengthy (Mascarenhas et al., 2013). Readers will recognize these dimensions as the leadership styles named dominance and prestige in the above, and the connotations of goodness and powerfulness in Heise’s theory. Kemper used these two concepts to underpin a generic theory of emotions, to be discussed further down. He extended his idea into a “status-power theory of relations” involving also group life (Theodore D Kemper, 2011). Recently, he wrote a concise version of his theories that is amenable to computational modelling (Theodore D. Kemper, 2017). In a nutshell, his theory posits that all people live in a status-power relational world. Status comes in many currencies. It implies love, respect, attention, applause, financial rewards, sexual favours, or a thousand other things large and small. People strive to attain these things by “claiming status”, through actions, nonverbal behaviours, clothes, appearance, hobbies, exploits, or vested in formal roles. This position paper, for instance, constitutes a status claim by its authors, in the currency of scientific credibility.

People thus strive for status. Yet they are not just selfish, but also motivated by love and affection to “confer status” upon others they deem worthy, or even upon heroes, symbols, deities, or groups. One person’s status worthiness is another one’s motive for conferring status. Status is thought to be a key driving factor in sustainable/durable inequality (Ridgeway, 2019).

When status claims fail, or when love is unrequited, people could respond by sadness, or by anger. In the latter case they might try to obtain the denied items by coercion, “power”. How to play the status-power game in life is something that people learn in their childhood, in a conjunction of “nature and nurture”. The fine print of the status-power game is cultural. For instance, some societies put a lot of value on power as a source of status, others do not; some societies divide status worthiness equally across people, others do not.

Two scientists who took their work and linked it to other disciplines could form an important source of inspiration for advances in sociality. They are Theodore Kemper and Antonio Damasio.

Socio-psycho-neurology: Kemper

Sociologist Theodore D. Kemper was mentioned above. He proposed a “Social interactional theory of emotions” that explicitly integrates socio-physiology of emotions, including work on the fit between neurophysiology and his own status-power model of relations (Theodore D Kemper, 1978). This is known in the literature as the “autonomic specificity hypothesis”, and Kemper’s theory supported it strongly, by linking neurotransmitters of the sympathetic nervous system with unpleasant events involving status loss (noradrenaline) and subjection to power (epinephrine). Acetylcholine, released by the parasympathetic nervous system, was associated with fulfilled status and power needs.

Kemper’s work was reviewed by sociologists with awe and admiration, but also with disbelief (Fine, 1981). It went largely forgotten. Recent work lends support to the specificity hypothesis once more, but without using Kemper’s theory, or integrating the findings across disciplines (McGinley & Friedman, 2017). Obviously, Kemper was ahead of his time. We believe his work is still innovative and important for the way in which it links neurobiology and sociology. According to Kemper, emotions tell their bearer whether survival is being facilitated (well-being signifies adequate status and power) or threatened (depression and fear signify reduced status or threat of others’ power) by events. Emotions are felt by individuals, carried by hormones, but induced by social situations involving relations between people. This is not to say that artificial sociality should include neurobiology. The importance of Kemper’s work is that it links disciplines operating at different levels of analysis, and shows the neurological roots of status and power motives.

Neuroscience: Damasio

Neuroscientist Damasio (2018) covers similar ground as Kemper does, but approaching from the opposite direction. Having noticed in his career that people are driven by more than their brains, he investigates the role of “feelings” in human cultural activity. Feelings, for Damasio, include avoidance of pain and suffering, and the pursuit of well-being and pleasure. They are more bodily, and less articulate, than emotions. For instance, “ache” is a feeling, “shame” is an emotion; feelings and emotions often co-occur. Damasio finds that feelings are not a new invention of evolutionary history, but are manifest in any single-cellular organism. He argues that any organism must maintain homeostasis of its inner environment in order to stay alive. “Feelings are the mental expressions of homeostasis” (ibid., p.6). Since all of our ancestors in the billion-years evolutionary history have had to maintain homeostasis in order to reproduce, “homeostasis, acting under the cover of feeling, is the functional thread that links early life-forms to the extraordinary partnership of bodies and nervous systems [of ourselves]”. Feelings are a primitive, powerful mechanism: we feel with our skins and our guts. Brains are just the latest addition to the organismic arsenal for maintaining homeostasis.

Damasio then turns to the social world: “In their need to cope with the human heart in conflict, in their desire to reconcile the contradictions posed by suffering, fear, anger, and the pursuit of well-being, humans turned to wonder and awe and discovered music making, painting, dancing and literature. They continued their efforts by creating the often beautiful and sometimes frayed epics that go by such names as religious belief, philosophical enquiry, and political governance.” (ibid., p. 8).

The impact of Damasio’s work is to downplay the role of intellect and mind in the shaping of collective behaviours, in favour of feelings. Damasio legitimizes gut feelings as motivators. It does not take much imagination to summarize his picture of feelings as a status-power world in the sense found by Kemper. Having adequate status causes well-being; being confronted with power causes fear. Since the world of feelings and emotions is less complex than the world of ideas, primacy of the former reduces the number of primitives required to model sociality, compared with a “brainy” world.

Damasio and Kemper together lay a strong foundation of consilience to the work of artificial sociality. Both give a central role to the organism, but not to the “self”. Kemper considers the “self” a superfluous notion; he considers the organism, with its feelings, as only one of the many reference groups that influence a person’s actions. Damasio shows that our organism has a life of its own, only some of which reaches our consciousness.

Acknowledgements

We thank the 150 attendants to the Artificial Sociality track at SocSimFest 2021, many of whom made valuable remarks that helped us.

[1] Durkheim puts a stronger emphasis on the stereotypical micro-level mechanisms in both forms of solidarity, such as enforcement mechanisms.

References

Axelrod, R. (1997). Advancing the Art of Simulation in the Social Sciences. 21-40.

Axelrod, R. (1997). The Dissemination of Culture; A model with local convergence and global polarization. Journal of Conflict Resolution, 41(2), 203-226. doi:10.1177/0022002797041002001

Balke, T., & Gilbert, N. (2014). How Do Agents Make Decisions? A Survey. Journal of Artificial Societies and Social Simulation, 17(4:13). doi:10.18564/jasss.2687

Beugelsdijk, S., & Welzel, C. (2018). Dimensions and Dynamics of National culture: Synthesizing Hofstede with Inglehart. Journal of Cross-Cultural Psychology, 49(10), 1469-1505. doi:10.1177/0022022118798505

Carley, K., & Newell, A. (1994). The nature of the social agent. Journal of Mathematical Sociology, 19(4), 221-262.

Castelfranchi, C. (2014). Minds as social institutions. Phenomenology and the Cognitive Sciences. doi:10.1007/s11097-013-9324-0

Damasio, A. R. (2018). The Strange Order of things: Life, Feeeling, and the Making of Cultures: Pantheon.

Deffuant, G., Carletti, T., & Huet, S. (2013). The Leviathan Model: Absolute Dominance, Generalised Distrust, Small Worlds and Other Patterns Emerging from Combining Vanity with Opinion Propagation. JASSS, 16(1), 5. http://jasss.soc.surrey.ac.uk/16/1/5.html

Diallo, S. Y., Shults, F. L. R., & Wildman, W. J. (2020). Minding morality: ethical artificial societies for public policy modeling. AI and Society. doi:10.1007/s00146-020-01028-5

Diamond, J. (1999). Guns, Germs and Steel: The fates of human societies: Norton & co.

Dias, J., Mascarenhas, S., & Paiva, A. (2016). FAtiMA Modular: Towards an Agent Architecture with a Generic Appraisal Framework. In Emotion Modeling (Vol. LNCS 8750, pp. 44-56): Springer.

Dignum, F., Hofstede, G. J., & Prada, R. (2014, 2014). From autistic to social agents. Paper presented at the 13th International Conference on Autonomous Agents and Multiagent Systems, AAMAS 2014.

Durkheim, E. (1984). The Division of Labour in Society.

Edmonds, B., Le Page, C., Bithell, M., Chattoe-Brown, E., Grimm, V., Meyer, R., . . . Squazzoni, F. (2019). Different Modelling Purposes. Journal of Artificial Societies and Social Simulation, 22(3), 6. doi:10.18564/jasss.3993

Edmonds, B., & Moss, S. (2005). From KISS to KIDS — An `Anti-simplistic’ Modelling Approach. Lecture Notes in Computer Science, 3415, 130-144. doi:10.1007/978-3-540-32243-6_11

Eliot, L. (2009). brain, blue brain: How small differences grow into troublesome gaps – and what we can do about it. Boston: Mariner Books.

Elsenbroich, C., & Verhagen, H. (2016). The simplicity of complex agents: a Contextual Action Framework for Computational Agents. Mind & Society, 15(1), 131-143.

Epstein, J. M. (2006). Generative Social Science: Studies in Agent-Based Computational Modeling: Princeton University Press.

Epstein, J. M., & Axtell, R. (1996). Growing artificial societies: social science from the bottom up. Washington D.C.: The Brookings Institution.

Fajardo, S., Hofstede, G. J., Vries, M. d., Kramer, M. R., & Bernal, A. (2020). Gregarious Behavior, Human Colonization and Social Differentiation: An Agent-based Model. JASSS, 23(4), 11. http://jasss.soc.surrey.ac.uk/23/4/11.html

Fine, G. A. (1981). Book review: A Social Interactional Theory of Emotions. Social Forces, 59(4), 1332-1333.

Fishbein, M., & Ajzen, I. (2010). Predicting and Changing Behavior: the Reasoned Action Approach: Psychology Press, Taylor & Francis.

Fiske, A. P. (1992). The four elementary forms of sociality: Framework for a unified theory of social relations. Psychological Review, 99(4), 689-723. doi:10.1037%2F0033-295X.99.4.689

Flache, A., Michael Mäs, Feliciani, T., Chattoe-Brown, E., Deffuant, G., Huet, S., & Lorenz, J. (2017). Models of Social Influence: Towards the Next Frontiers. Journal of Artificial Societies and Social Simulation, 20(4), 2. http://jasss.soc.surrey.ac.uk/20/4/2.html

Frijda, N. H. (1986). The Emotions. Cambridge: Cambridge University Press.

Gilbert, N., & Conte, R. (1995). Artificial Societies: The Computer Simulation of Social Life University College London Press.

Glöckner, A., & Witteman, C. (2010). Beyond dual-process models: A categorisation of processes underlying intuitive judgement and decision making. Thinking & Reasoning, 16(1), 1-25.

Groep, S. v. d., Zanolie, K., & Crone, E. A. (2019). Giving to Friends, classmates, and strangers in Adolescence. Journal of Research on Adolescence(online first), 1-8. doi:10.1111/jora.12491

Güroglu, B., Bos, W. v. d., & Crone, E. A. (2014). Sharing and giving across adolescence: an experimental study examining the development of prosocial behavior. Frontiers in Psychology, 5(291), 1-13. doi:10.3389/fpsyg.2014.00291

Haidt, J. (2012). The Righteous Mind: Why Good People are Divided by Politics and Religion: Penguin.

Hardin, G. (1968). The Tragedy of the Commons. Science, 162(3859), 1243-1248.

Hegselmann, R. (2017). Thomas C. Schelling and James M. Sakoda: The Intellectual, Technical, and Social History of a Model. JASSS, 20(3), 15. http://jasss.soc.surrey.ac.uk/20/3/15.html

Hegselmann, R., & Flache, A. (1998). Understanding Complex Social Dynamics: A Plea For CellularAutomata Based Modelling. JASSS, 1(3), 1. http://jasss.soc.surrey.ac.uk/1/3/1.html

Heise, D. R. (2013). Modeling Interactions in Small Groups. Social Psychology Quarterly, 76(1), 52-72. doi:10.1177/0190272512467654

Hemelrijk, C. K. (2000). Towards the integration of social dominance and spatial structure. Animal Behaviour, 59(5), 1035-1048. doi:http://dx.doi.org/10.1006/anbe.2000.1400

Hemelrijk, C. K. (2011). Simple Reactions to Nearby Neighbors and Complex social Behavior in Primates. In R. F. J. Menzel (Ed.), Animal Thinking: Comparative Issues in Comparative Cognition (pp. 223-238): MIT Press.

Henrich, J. (2016). The Secret of Our Success: How Culture Is Driving Human Evolution, Domesticating Our Species, and Making Us Smarter: Princeton University Press.

Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., . . . Tracer, D. (2005). “Economic man” in cross-cultural perspective: Behavioral experiments in 15 small-scale societies. Behavioral and brain sciences, 28, 795-855.

Hodgson, G. M. (2015). On defining institutions: Rules versus equilibria. Journal of Institutional Economics. doi:10.1017/S1744137415000028

Hoey, J., & Schröder, T. (2015). Bayesian affect control theory of self. Paper presented at the Proceedings of the AAAI Conference on Artificial Intelligence.

Hoey, J., Schröder, T., & Alhothali, A. (2016). Affect control processes: Intelligent affective interaction using a partially observable Markov decision process. Artificial Intelligence, 230, 134-172.

Hoey, J., Schröder, T., Morgan, J., Rogers, K. B., Rishi, D., & Nagappan, M. (2018). Artificial intelligence and social simulation: Studying group dynamics on a massive scale. Small Group Research, 49(6), 647-683.

Hofstede, G., Hofstede, G. J., & Minkov, M. (2010). Cultures and Organizations, software of the mind: McGraw-Hill.

Hofstede, G. J. (2013). Theory in social simulation: Status-Power theory, national culture and emergence of the glass ceiling. Paper presented at the Social Coordination: Principles, Artefacts and Theories, Exeter. http://www.scopus.com/inward/record.url?eid=2-s2.0-84894173186&partnerID=MN8TOARS

Hofstede, G. J. (2018). Social Simulation as a meeting place: report of SSC 2018 Stockholm. Review of Artificial Societies and social Simulation, 2018. https://rofasss.org/2018/09/19/gh/

Hofstede, G. J. (2019). GRASP agents: social first, intelligent later. Ai & Society, 34(3), 535-543. doi:https://doi.org/10.1007/s00146-017-0783-7

Hofstede, G. J., Dignum, F., Prada, R., Student, J., & Vanhée, L. (2015). Gender differences: The role of nature, nurture, social identity and self-organization (Vol. 9002).

Hofstede, G. J., Jonker, C. M., Verwaart, T., & Yorke-Smith, N. (2019). The Lemon Car Game Across Cultures: Evidence of Relational Rationality. Group Decision and Negotiation, 28(5), 849-877. doi:10.1007/s10726-019-09630-9

Hofstede, G. J., & Liu, C. (2019). GRASP world. https://www.comses.net OpenABM Computational library. Retrieved from https://www.comses.net/codebases/f1089671-38da-4cca-88c5-711e936b2ada/releases/1.0.0/.

Hofstede, G. J., & Liu, C. (2020). To Stay or Not to Stay? Artificial sociality in GRASP world. In H. Verhagen, M. Borit, G. Bravo, & N. Wijermans (Eds.), Advances in Social Simulation (pp. 217-231). Cham: Springer.

Jager, W. (2017). Enhancing the Realism of Simulation (EROS): On Implementing and Developing Psychological theory in Social Simulation. JASSS, 20(3). doi:10.18564/jasss.3522

Jager, W., & Janssen, M. A. (2012). An updated conceptual framework for integrated modelling of human decision making: The consumat II. Paper presented at the ECCS, Brussel.

Jager, W., Janssen, M. A., & Vlek, C. A. J. (1999). Consumats in a commons dilemma: Testing the behavioural rules of simulated consumers [Press release]. Retrieved from http://www.ppsw.rug.nl/cov/staff/jager/simpaper.pdf

Kahneman, D. (2011). Thinking, fast and slow: Macmillan.

Kajic, I., Schroeder, T., Stewart, T. C., & Thagard, P. (2019). The semantic pointer theory of emotion: Integrating physiology, appraisal, and construction. Cognitive Systems Research, 58(December), 35-53. doi:https://doi.org/10.1016/j.cogsys.2019.04.007

Kemper, T. D. (1978). A social interactional theory of emotions: John Wiley & Sons.

Kemper, T. D. (2011). Status, Power and Ritual Interaction: a Relational Reading of Durkheim, Goffman and Collins: Ashgate.

Kemper, T. D. (2017). Elementary Forms of social Relations: Status, power and reference groups Routledge.

Kopecky, J., Bos, N., & Greenberg, A. (2010). Social identity modeling: Past work and relevant issues for socio-cultural modeling. Paper presented at the 19th Conference on Behavior Representation in Modeling and Simulation, Charleston, SC.

Levine, M., Prosser, A., Evans, D., & Reicher, S. (2005). Identity and emergency intervention: How social group membership and inclusiveness of group boundaries shape helping behaviour. Personality and Social Psychology Bulletin,, 34(4), 443-453. doi:http://dx.10.1177/0146167204271651

Mascarenhas, S., Guimarães, M., Santos, P. A., Dias, J., Prada, R., & Paiva, A. (2021). FAtiMA Toolkit–Toward an effective and accessible tool for the development of intelligent virtual agents and social robots. arXiv preprint arXiv:2103.03020. https://arxiv.org/abs/2103.03020

Mascarenhas, S., Prada, R., Paiva, A., & Hofstede, G. J. (2013). Social Importance Dynamics: A Model for Culturally-Adaptive Agents. In R. Aylett, B. Krenn, C. Pelachaud, & H. Shimodaira (Eds.), Intelligent Virtual Agents (Vol. 8108, pp. 325-338): Springer.

Maslow, A. H. (1970). Motivation and Personality (2nd ed.). New York: Harper & Row.

McGinley, J. J., & Friedman, B. H. (2017). Autonomic specificity in emotion: The induction method matters. International Journal of Psychophysiology, 118, 48-57. doi:https://doi.org/10.1016/j.ijpsycho.2017.06.002

Mercier, H., & Sperber, D. (2017). The enigma of reason: Harvard University Press.

Milgrom, P. R., North, D. C., & Weingast, B. R. (1990). The Role of Institutions in the Revival of the Trade: The Law Merchant, Private Judges, and the Champagne Fairs. Economics and Politics, 2(1), 1–23.

Montesquieu, C. L. d. (1979 [1742]). De l’esprit des lois (Vol. 1). Paris: GF-Flammarion.

Moutoussis, M., Trujillo-Barreto, N. J., El-Deredy, W., Dolan, R., & Friston, K. (2014). A formal model of interpersonal inference. Frontiers in human neuroscience, 8, 160.

North, D. C. (1990). Institutions, Institutional Change, and Economic Performance: Cambridge University Press.

North, D. C., Wallis, J. J., & Weingast, B. R. (2009). Violence and Social Orders: A Conceptual Framework for Interpreting Recorded Human History: Cambridge University Press.

Ortony, A., Clore, G., & Collins, A. (1998). The Cognitive Structure of Emotions. Cambridge, UK: Cambridge University Press.

Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action: Cambridge University Press.

Reicher, S. D., Spears, R., & Haslam, S. A. (2010). The Social Identity Approach in Social Psychology. In M. S. Wetherlell & C. T. Mohanty (Eds.), The SAGE handbook of identities (pp. 45-62). London: Sage.

Ridgeway, C. L. (2019). Status: Why Is It Everywhere? Why Does It Matter? : Russell Sage Foundation.

Schlüter, M., Baeza, A., Dressler, G., Frank, K., Groeneveld, J., Jager, W., . . . NandaWijermans. (2017). A framework for mapping and comparing behavioural theories in models of social-ecological systems. Ecological Economics, 131, 21-35. doi:10.1016/j.ecolecon.2016.08.008

Scholz, G., Eberhard, T., Ostrowski, R., & Wijermans, N. (2021 (in press)). Social Identity in Agent-Based Models—Exploring the State of the Art.

Schröder, T., Hoey, J., & Rogers, K. B. (2016). Modeling dynamic identities and uncertainty in social interactions: Bayesian affect control theory. American Sociological Review, 81(4), 828-855.

Schweitzer, F., & Garcia, D. (2010). An agent-based model of collective emotions in onlinecommunities. The European Physical Journal B, 77(4), 533-545. doi:10.1140/epjb/e2010-00292-1

Searle, J. R. (1995). The Construction of Social Reality. London: Penguin.

Shivakoti, G. P., Janssen, M. A., & Chhetri, N. B. (2019). Agricultural and natural resources adaptations to climate change: Governance challenges in Asia. International Journal of the Commons. doi:10.5334/ijc.999

Shults, F. L., Gore, R., Wildman, W. J., Lynch, C. J., Lane, J. E., & Toft, M. D. (2018). A generative model of the mutual escalation of anxiety between religious groups. Journal of Artificial Societies and Social Simulation, 21(4), 7. http://jasss.soc.surrey.ac.uk/21/4/7.html

Shults, F. L., Lane, J. E., Wildman, W. J., Diallo, S., Lynch, C. J., & Gore, R. (2018). Modelling terror management theory: computer simulations of the impact of mortality salience on religiosity. Religion, Brain & Behavior, 8(1), 77-100.

Smith, P. B., Bond, M. H., & Kagitcibasi, C. (2006). Understanding Social Psychology Across Cultures Sage.

Sun, Z., Lorscheid, I., Millington, J. D., Lauf, S., Magliocca, N. R., Groeneveld, J., . . . Schulze, J. (2016). Simple or complicated agent-based models? A complicated issue. Environmental Modelling & Software, 86, 56-67.

Swaab, D. F., & Hedley-Prole, J. (2014). We Are Our Brains: A Neurobiography of the Brain, from the Womb to Alzheimer’s: Random House.

Tajfel, H. (1978). Intergroup behaviour. In Introducing social psychology (pp. 401-422). Harmondsworth: Penguin.

Tajfel, H. (1982). Social identity and intergroup relations: Cambridge University Press.

Tajfel, H., & Turner, J. C. (1986). The social identity theory of intergroup behaviour. In S. Worchel & W. G. Austin (Eds.), Psychology of intergroup relations (2nd ed., pp. 7-24). Chicago: Nelson-Hall.

Thaler, R. H. (2000). From homo economicus to homo sapiens. Journal of economic perspectives, 14(1), 133-141.

Tolk, A. (2015). Learning Something Right from Models That Are Wrong: Epistemology of Simulation. In L. Yilmaz (Ed.), Concepts and Methodologies for Modeling and Simulation: A Tribute to Tuncer Ören (pp. 87-106). Cham: Springer International Publishing.

Tönnies, F. (1963 [1887]). Community \& society : (Gemeinschaft und Gesellschaft).

Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, M. (1987). Rediscovering the social group: A self-categorization theory. Oxford, UK: Basil Blackwell.

Turner, J. E. (2007). Human Emotions: A Sociological Theory: Routledge.

Van Vugt, M., & von Rueden, C. R. (2020). From genes to minds to cultures: Evolutionary approaches to leadership. The Leadership Quarterly, 101404.

Veissière, S. P., Constant, A., Ramstead, M. J., Friston, K. J., & Kirmayer, L. J. (2020). Thinking through other minds: A variational approach to cognition and culture. Behavioral and brain sciences, 43.

Waal, F. d. (1996). Good natured: the origins of right and wrong in humans and other animals: Harvard University Press.

Waal, F. d. (2009). The Age of Empathy. New York: Random House.

Wilson, E. O. (1999). Consilience: The unity of knowledge (Vol. 31): Vintage.

Wilson, E. O., & Hölldobler, B. (2005). Eusociality: Origin and consequences. Proceedings of the National Academy of Sciences of the United States of America, 102(38), 13367-13371. doi:10.1073/pnas.0505858102

Zhu, J., & Thagard, P. (2002). Emotion and action. Philosophical psychology, 15(1), 19-36.


Hofstede, G.J, Frantz, C., Hoey, J., Scholz, G. and Schröder, T. (2021) Artificial Sociality Manifesto. Review of Artificial Societies and Social Simulation, 8th Apr 2021. https://rofasss.org/2021/04/08/artsocmanif/


 

How Can ABM Models Become Part of the Policy-Making Process in Times of Emergencies – The S.I.S.A.R. Epidemic Model

By Gianpiero Pescarmona1, Pietro Terna2,*, Alberto Acquadro1, Paolo Pescarmona3, Giuseppe Russo4, and Stefano Terna5

*Corresponding author, 1University of Torino, IT, 2University of Torino, IT, retired & Collegio Carlo Alberto, IT, 3University of Groningen, NL, 4Centro Einaudi, Torino, IT, 5tomorrowdata.io

(A contribution to the: JASSS-Covid19-Thread)

We propose an agent-based model to simulate the Covid-19 epidemic diffusion, with Susceptible, Infected, symptomatic, asymptomatic, and Recovered people: hence the name S.I.s.a.R. The scheme comes from S.I.R. models, with (i) infected agents categorized as symptomatic and asymptomatic and (ii) the places of contagion specified in a detailed way, thanks to agent-based modeling capabilities. The infection transmission is related to three factors: the infected person’s characteristics and the susceptible one, plus those of the space in which contact occurs. The asset of the model is the development of a tool that allows analyzing the contagions’ sequences in simulated epidemics and identifying the places where they occur.

The characteristics of the S.I.s.a.R. model

S.I.s.a.R. can be found at https://terna.to.it/simul/SIsaR.html with information on model construction, the draft of a paper also reporting results, and an online executable version of the simulation program, built using NetLogo. The model includes the structural data of Piedmont, an Italian region, but it can be readily calibrated for other areas. The model reproduces a realistic calendar (e.g., national or local government decisions), via a dedicated script interpreter.

Why another model? The starting point has been the need to model the pandemic problem in a multi-scale way. This was initiated a few months before the publication of new frontier articles, such as Bellomo et al. (2020), so when equation-based S.I.R. models, with their different versions, were predominating.

As any model, also this one is based on some assumptions: time will tell whether these were reasonable hypotheses. Modeling the Covid-19 pandemic requires a scenario and the actors. As in a theatre play, the author defines the roles of the actors and the environment. The characters are not real, they are pre-built by the author, and they act according to their peculiar constraints. If the play is successful, it will run for a long time, even centuries. If not, we will rapidly forget it. Shakespeare’s Hamlet is still playing after centuries, even if the characters and the plot are entirely imaginary. The same holds for our simulations: we are the authors, we arbitrarily define the characters, we force them to act again and again in different scenarios. However, in our model, the micro-micro assumptions are not arbitrary but based on scientific hypotheses at the molecular level, the micro agents’ behaviors are modeled in an explicit and realistic way. In both plays and simulations, we compress the time: a whole life to 2 or 3 hours on the stage. In a few seconds, we run the Covid-19 pandemic spread in a given regional area.

With our model, we move from a macro compartmental vision to a meso and microanalysis capability. Its main characteristics are:

  • scalability: we take in account the interactions between virus and molecules inside the host, the interactions between individuals in more or less restricted contexts, the movement between different environments (home, school, workplace, open spaces, shops, in a second version, we will add transportations and long trips between regions/countries; discotheques; other social aggregation events, as football matches); the movements occur in different parts of the daily life, as in Ghorbani et al. (2020);

the scales are:

    • micro, with the internal biochemical mechanism involved in reacting to the virus, as in Silvagno et al. (2020), from where we derive the critical importance assigned to an individual intrinsic susceptibility related to the age and previous morbidity episodes; the model incorporates the medical insights of one of its co-authors, former full professor of clinical biochemistry, signing also the quoted article; a comment on Lancet (Horton, 2020) consistently signals the syndemic character of the current event: «Two categories of disease are interacting within specific populations—infection with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and an array of non-communicable diseases (NCDs)»;
    • meso, with the open and closed contexts where the agents behave, as reported above;
    • macro, with the emergent effects of the actions of the agents; this final analysis is a premise to evaluate the costs and benefits of the different intervention policies;
  • granularity: at any level, the interactions are partially random and therefore the final results always reflect the sum of the randomness at the different levels; changing the constraints at different levels and running multiple simulations should allow the identification of the most critical points, i.e., those on which the intervention should be focused.

Contagion sequences as a source of suggestions for intervention policies

All the previous considerations are not exhaustive. The critical point that makes helpful the production of a new model is creating a tool that allows analyzing the contagions’ sequences in simulated epidemics and identifying the places where they occur. We represent each infecting agent as a horizontal segment with a vertical connection to another agent receiving the infection. We represent the second agent via a further segment at an upper layer. With colors, line thickness, and styles, we display multiple data.

As an example, look at Fig.4: we start with two agents coming from the outside, with black as color code (external place), the first one–regular, as reported by the thickness of the segment, starting at day 0 and finishing at day 22–is asymptomatic (dashed line) and infects five agents; the second one–robust, as reported by the thickness of the segment, starting at day 0 and finishing at day 15–is asymptomatic (dashed line) and infects no one; the first of the five infected agents received the infection at home (cyan color) and turns to be asymptomatic after a few days of incubation (dotted line), and so on. Solid lines identify symptomatic agents; brown color refers to workplaces, orange to nursing homes; yellow to schools; pink to hospitals; gray to open spaces. Thick or extra-thick lines refer to fragile or extra-fragile agents, respectively.

This technique enables understanding at a glance how an epidemic episode is developing. In this way, it is easier to reason about countermeasures and, thus, to develop intervention policies. In Figs. 1-4, we can look both at the places where contagions occur and at the dynamics emerging with different levels of intervention. In Fig. 1 we find evidence of the role of the workplaces in diffusing the infection, with a relevant number of infected fragile workers. In Fig. 2, by isolating fragile workers at home, the epidemics seems to finish, but in Fig. 3, we see a thin event (a single case of contagion) that creates a bridge toward a second wave. Finally, in Fig. 4, we see that the epidemic is under control by isolating the workers and any kind of fragile agents. (Please enlarge the on-screen images to see more details).

A scatter graph showing an epidemic with regular containment measures, showing a highly significant effect of workplaces (brown)Figure 1 – An epidemic with regular containment measures, showing a highly significant effect of workplaces (brown)

A scatter graph showing The effects of stopping fragile workers at day 20, with a positive result, but home contagions (cyan) keep alive the pandemic, exploding again in workplaces (brown)

Figure 2 – The effects of stopping fragile workers at day 20, with a positive result, but home contagions (cyan) keep alive the pandemic, exploding again in workplaces (brown)

A scatter graph showing The effects of stopping fragile workers at day 20, with a positive result, but home contagions (cyan) keep alive the pandemic, exploding again in workplaces (brown)

Figure 3 – Same, analyzing the first 200 infections with evidence of the event around day 110 with the new phase due to a unique asymptomatic worker

A scatter graph showing the impoaact of Stopping fragile workers plus any case of fragility at day 15, also isolating nursing homes

Figure 4 – Stopping fragile workers plus any case of fragility at day 15, also isolating nursing homes

Batches of simulation runs

The sequence in the steps described by the four figures is only a snapshot, a suggestion. We need to explore systematically the introduction of factual, counterfactual, and prospective interventions to control the spread of the contagions. Each simulation run–whose length coincides with the disappearance of symptomatic or asymptomatic contagion cases–is a datum in a wide scenario of variability in time and effects. Consequently, we need to represent compactly the results emerging from batches of repetitions, to compare the consequences of each batch’s basic assumptions.

For this purpose, we used blocks of one thousand repetitions. Besides summarizing the results with the usual statistical indicators, we adopted the technique of the heat-maps. In this perspective, with Steinmann et al. (2020), we developed a tool for comparative analyses, not for forecasting. This consideration is consistent with the enormous standard deviation values that are intrinsic to the problem.

Figs. 5-6 provide two heat-maps reporting the duration of each simulated epidemic in the x axis and the number of the symptomatic, asymptomatic, and deceased agents in the y axis. 1,000 runs in both cases.

The actual data for Piedmont, where the curve of the contagions flattened with the end of May, with around 30 thousand subjects, is included in the cell in the first row, immediately to the right of the mode in Fig. 6. In the Fall, a second wave seems possible, jumping into one of the events of the range of events on the right side of the same figure.

Figure 5 – 1000 Epidemics without containment measures (2D histogram of (Symptomatic+Asymptomatic+Deceased against days)

Figure 6 – 1000 Epidemics with basic non-pharmaceutical containment measures, no school in September 2020 (2D histogram of (Symptomatic+Asymptomatic+Deceased against days)

In Table 1 we have a set of statistical indicators related to 1,000 runs of the simulation with the different initial conditions. Cases 1 and 2 are those of Fig. 5 and 6. Then we introduce Case 4, excluding from the workplace workers with health fragilities, so highly susceptible to contagion, with smart work when possible or sick pay conditions. The gain in the reduction of affected people and duration is relevant and increases – in Case 5 – if we leave at home all kinds of fragile people.

Scenarios Total symptomatic Total symptomatic, asymptomatic, deceased Days
1. no control 851.12
(288.52)
2253.48
(767.58)
340.10
(110.21)
2. basic controls, no school in Sep 2020 158.55
(174.10)
416.98
(462.94)
196.97
(131.18)
4. basic controls, stop fragile workers, no schools in Sep 2020 120.17
(149.10)
334.68
(413.90)
181.10
(125.46)
5. basic controls, stop fragile workers & fragile people, nursing-homes isolation, no schools in Sep 2020 105.63
(134.80)
302.62
(382.14)
174.39
(12.82)
7. basic controls, stop f. workers & fragile people, nursing-homes isolation, open factories, schools in Sep 2020 116.55
(130.91)
374.68
(394.66)
195.28
(119.33)

Table 1 – Statistical indicators, limited to the mean and to the standard deviation, reported in parentheses, for a set of experiments; the row numbers are consistent with the paper at https://terna.to.it/simul/SIsaR.html where we report a larger number of simulation experiments

In Case 7, we show that keeping the conditions of Case 5, while opening schools and factories (work places in general), increases in a limited way the adverse events.

A second version

A second version of the model is under development, using https://terna.github.io/SLAPP/, a Python shell for ABM prepared by one of the authors of this note, referring to the pioneering proposal http://www.swarm.org of the Santa Fe Institute.

References

Bellomo, N., Bingham, R., Chaplain, M. A. J., Dosi, G., Forni, G., Knopoff, D. A., Lowengrub,, J., Twarock, R., and Virgillito, M. E. (2020). A multi-scale model of virus pandemic: Heterogeneous interactive entities in a globally connected world. arXiv e-prints, art. arXiv:2006.03915, June.

Ghorbani, A., Lorig, F., de Bruin, B., Davidsson, P., Dignum, F., Dignum, V., van der Hurk, M., Jensen, M., Kammler, C., Kreulen, K., et al. (2020). The ASSOCC Simulation Model: A Response to the Community Call for the COVID-19 Pandemic. Review of Artificial Societies and Social Simulation. URL https://rofasss.org/2020/04/25/the-assocc-simulation-model/.

Horton, R. (2020). Offline: Covid-19 is not a pandemic. Lancet (London, England), 396(10255):874. URL https://www.thelancet.com/action/showPdf?pii=S0140-6736%2820%2932000-6.

Silvagno, F., Vernone, A. and Pescarmona, G. P. (2020). The Role of Glutathione in Protecting against the Severe Inflammatory Response Triggered by COVID-19. In «Antioxidants», vol. 9(7), p. 624. http://dx.doi.org/10.3390/antiox9070624.

Steinmann P., Wang J. R., van Voorn G. A., and Kwakkel J. H. (2020). Don’t try to predict covid-19. if you must, use deep uncertainty methods. Review of Artificial Societies and Social Simulation, 17. https://rofasss.org/2020/04/17/deep-uncertainty/.


Pescarmona, G., Terna, P., Acquadro, A., Pescarmona, P., Russo, G., and Terna, S. (2020) How Can ABM Models Become Part of the Policy-Making Process in Times of Emergencies - The S.I.S.A.R. Epidemic Model. Review of Artificial Societies and Social Simulation, 20th Oct 2020. https://rofasss.org/2020/10/20/sisar/


 

Flu and Coronavirus Simulator – A geospatial agent-based simulator for analyzing COVID-19 spread and public health measures on local regions

By Imran Mahmood

A Summary of: Mahmood et al. (2020)

In this paper we reviewed the lessons learned during the development of the ‘Flu and Coronavirus Simulator’ (FACS) and compare our chosen Agent-based Simulation approach with the conventional disease modelling approaches.

FACS provides an open-ended platform for the specification and implementation of the primary components of Agent-Based Simulation (ABS): (i) Agents; (ii) Virtual environment and (iii) Rule-set using a systematic Simulation Development Approach. FACS inherits features of a comprehensive simulation framework from its ancestors: (i) FLEE (Groen & Arabnejad, 2015) and (ii) FabSim3 (Groen & Arabnejad 2014). Where, FLEE mainly specializes in ABS complex dynamics e.g., agent movements; FabSim3 provides the ability to simulate a large population of agents with microscopic details using remote supercomputers. The combination of this legacy code offers numerous benefits including high performance, high scalability, and greater re-usability through a model coupling. Hence it provides an open-ended API for modellers and programmers to use it for further scientific research and development. FACS generalizes the process of disease modelling and provides a template to model any infectious disease. Thus allowing: (i) non-programmers (e.g., epidemiologists and healthcare data scientists) to use the framework as a disease modelling suite; and (ii) providing an open-ended API for modellers and programmers to use it for further scientific research and development. FACS offers a built-in location graph construction tool that allows the import of large spatial data-sets (e.g., Open Street Map), automated parsing and pre-processing of the spatial data, and generating buildings of various types, thus allowing ease in the synthesis of the virtual environment for the region under consideration. FACS provides a realistic disease transmission algorithm with the ability to capture population interactions and demographic patterns e.g., age diversity, daily life activities, mobility patterns, exposure at the street-level or in public transportation, use, or no use of face mask, assumptions of exposure within closed quarters.

We believe our approach has proven to be quite productive in modelling complex systems like epidemic spread in large regions due to ever-changing model requirements, multi-resolution abstraction, non-linear system dynamics, rule-based heuristics, and above all large-scale computing requirements. During the development of this framework, we learned that the real-world abstraction changes more rapidly than in other circumstances. For instance, the concept of social distancing and lockdown scenarios have evolved significantly since early March. Therefore, rapid changes in the ABS model were necessary. Model building in these cases benefits more from using a bottom-up approach like ABS, as opposed to any centralized analytical solution.

References

Groen, D., & Arabnejad, H. (2014). Fabsim3. GitHub. https://github.com/djgroen/FabSim3

Groen, D., & Arabnejad, H. (2015). Flee. GitHub. https:// github.com/djgroen/flee

Mahmood, I.,  Arabnejad, H., Suleimenova, D., Sassoon, I.,  Marshan, A.,  Serrano-Rico, A.,  Louvieris, P., Anagnostou, A., Taylor, S.J.E., Bell, D. & Groen, D. (2020) FACS: a geospatial agent-based simulator for analysing COVID-19 spread and public health measures on local regions, Journal of Simulation, DOI: 10.1080/17477778.2020.1800422


Mahmood, I. (2020) Flu and Coronavirus Simulator - A geospatial agent-based simulator for analyzing COVID-19 spread and public health measures on local regions. Review of Artificial Societies and Social Simulation, 10th Sept 2020. https://rofasss.org/2020/09/10/facs/