Kluver and colleagues (2003) point out the main difference in building theories between social sciences and natural sciences. The latter always proceed through increment, starting with rather simple models and enlarging them by successive steps, whenever the progress of research made it necessary. This normal science is guided by a certain paradigm that is transformed into an increasingly complex model as long as the paradigm makes it possible. By contrast, social sciences often adopt an all-at-once approach by which theorists try to capture from the beginning as much of social complexity as they can within their conceptual framework. Kluver and colleagues have no doubt about the fact that social sciences must embrace a normal posture in order to progress towards a better understanding of social complexity (2003: 3):

That is not possible in computational or mathematical sociology, respectively: the basic models must be simple in order to understand their behaviour in principle. The enlargement of the basic models that is always necessary in advancing research can only be done if this basic understanding has been achieved. Therefore the social sciences have to adopt this methodical procedure from the natural sciences if a formally precise theory of social complexity is to be achieved.

The SAI paradigm has supported, through normal and positivist procedures, the implementation of always more complex models of individual decision and social behaviour. If emotional decisions or the ‘embodiment’ of cognition are still on the shelves, the social rationality of agents, as opposed to their goal-satisfying rationality, has recently attracted much attention. Hogg and Jennings (2001: 382), recognising that individual and social concerns often conflict leading to the possibility of inefficient performances, proposed a framework for making socially acceptable decisions, based on social welfare functions, that combine social and individual perspectives:

To be socially rational, an individual maximizes his social welfare function over the different alternatives. This function represents how an individual agent may judge states of the world from a moral or social perspective by taking into consideration the benefit to others of its course of action and weighing it against its own benefits.

The dynamic balance between individual and social utility functions is controlled by a metalevel controller that considers available resources (adaptation) and past experiences (learning) to tune the amount of cognitive efforts put into social rationality. An equivalent mechanism is used by Castelfranchi (2001) in order to formalise social functions (or roles) assumed by individuals as non-intentional mental processes. In this case, a metalevel controller supersedes the intentional and rational system of the agent. A learning-without-understanding process reinforces mechanically some individual beliefs, whether they are beneficial for the agent or not.

In both cases, the cognitive architecture becomes increasingly complex and the normality paradigm appears to act as a ‘patching’ process applied on a system of formal inferences that was not designed for such a purpose in the first place. Furthermore, these architectures and organisations seldom rely on direct evidence for validation. Instead, virtual social experiments are used to generate results that are evaluated against plausible utility values at the system and individual levels.

In order to overcome the increasing complexity of SAI architectures, and to facilitate, to some extend, the direct validation of the building assumptions, Jager and Janssen (2003) propose an alternate formalism. Their consumat theory is to be considered one of these conceptual jumps characteristic of the evolution of normal science: a new paradigm supports the creation of simpler models compared with the previous ageing generation. The consumat approach considers basic human needs and environmental uncertainties as the driving factors behind decision making processes. Agents engage in different cognitive processes, including social imitation and comparison, according to their perceptions of individual needs and environmental threat (Figure 3.4). Hence, the consumat paradigm overrides two structural limitations of the symbolico-cognitivist paradigm: experience preempts cognition and social rationality is directly built into intentional processes. Being at an early stage of development, the new theory relies on relatively simple rules that can be validated against experimental economics settings.

Source: Jager and Jannsen 2003

Funtowicz and Ravetz (1993) also use environmental uncertainties, along with decision stakes, to analyse problem-solving strategies in the context of environmental and population risk policy issues. They argue that traditional scientific methodologies are ineffective when either attribute is high. Instead, they propose a new scientific posture they call post-normal science (op. cit.: 739):

In those circumstances, the quality assurance of scientific inputs to the policy process requires an ‘extended peer community’, consisting of all those with a stake in the dialogue on the issue. Post-normal science can provide a path to the democratization of science, and also a response to the current tendencies to post-modernity.

The dynamic of resolution of policy issues in post-normal science involves the inclusion of an adequate set of legitimate participants in the process of quality assurance of the scientific inputs. For example, persons directly affected by an environmental problem will have a keener awareness of its symptoms. Thus, they perform a function analogous to that of peer-reviewers in traditional science, which otherwise might not occur in these specific contexts. Closer to the concern of this chapter, Funtowicz and Ravetz (1993: 745) challenge the commonly admitted rationality of decision and action:

Until now, with the dominance of applied science, the rationality of reductionist natural-scientific research has been taken as a model for the rationality of intellectual and social activity in general. However successful it has been in the past, the recognition of the policy issues of risk and the environment shows that this ideal of rationality is no longer universally appropriate. The activity of science now encompasses the management of irreducible uncertainties in knowledge and in ethics, and the recognition of different legitimate perspectives and ways of knowing.

Now, let’s put this post-normal scientific posture into the context of our chapter. Among the majority of MABS used to explore human ecosystems, a significant number are meant to demonstrate how individuals and populations interact with their environment, as well as the environmental and social consequences of these interactions (Bousquet and LePage 2004). These situations are often characterised by:

We have to accept the fact that Multi Agent Models, even the more sophisticated ones, will always be pale copies of the original, subjective and partial representations of a dynamic reality. But recognising this very peculiar fact doesn’t mean that these models are useless, even Lissak and Richardson (2001: 105) in their criticism of computer-based social models admit that:

There is no need for the models in question to have predictive power, despite the strong desire of both consultants and their clients that such models ‘work’. The pedagogical value of exploring the interactions of complex relations through the manipulation of models is more than enough to justify the efforts that go into model development and proliferation. Clearly, it is easier to manipulate a computer model than a fully fledged ‘in reality’ laboratory experiment, but the limitations of such models must be remembered.

Chefs-d’Oeuvres are chiselled by talented craftsmen, not by their tools. Talent is all about sharing a vision and choosing the appropriate tools. In this regard, we must admit that the ultimate criteria of validation for this type of model is its actual appropriation by the final users (policy-makers, local communities, or else) and the consensual acceptation of the simulated outcomes. In this case, we can infer that:

Due to irreducible uncertainties and complex interactions within human ecosystems, social simulation designers should abide by a principle of subsidiary formalism. The principle acknowledges the fact that the most reliable source of knowledge about human decisions and behaviour are indeed the actors of the real drama themselves. Hence, the principle of subsidiarity stipulates that designer’s cognitive efforts should be directed towards engaging with real stakeholders and eliciting their own mental models in the first place. Only when this option is unrealistic, the designers shall make it clear to everyone that decisional rules and algorithms are derived from theoretical or empirical predicates.

The subsidiarity principle offers a non-threatening opportunity for ethnography to accept the challenge of a new and dynamic formalism, beside narratives (Lansing 2003). The principle also mitigates the accusation of mere indexicality uttered by System Thinkers against individual-based simulations (Lissak and Richardson 2001). Finally, a post-normal posture might help solve the problem of validation that any complex system model is faced with: partial scientific validation associated with social authentication legitimate the modelling process and its outcomes (Funtowicz and Ravetz 1993).

During the early 1990s, Granath (1991) introduced the concept of collective design in industry to define a process by which all actors involved in the production, diffusion, or consumption of a product are considered as equal experts and invited to participate to the design of the product. Each expert actively contributes to the collective process of transdisciplinary creation. Collective design usually faces two problems:

Hence, in order to implement a collective design process, it is important to initially elicit specific knowledge and practices among experts. Then, the process itself must be grounded into successive mediating objects. These artifacts (ideas, models, products) are meant to channel creativity and to structure communication among experts. Collective design is to be considered as a social construct, no longer functional or rational. The final product emerges from conflicts, alliances, and negotiations.

During the late 90s, collective design and other approaches like system thinking or action learning were appropriated and adapted by scientists working on natural resource management (Hagmann et al. 2002; Barreteau 2003; D’Aquino et al. 2003). Interestingly, most of the cases were characterised by:

The co-construction of these models with local stakeholders didn’t intend to provide normative models of reality, instead they were meant to enhance discussion and collective action through interactions around and about the mediating object (Lynam et al. 2002: 2):

It is important to emphasize that, in the contexts in which these case studies are presented, the models were used more as part of a process of developing and exploring a common understanding of problems and possible solutions. They were not designed to be highly validated, predictive models in the sense in which systemic models are usually developed and used. We are not aware of other examples in which local people, who have no history of computer-based modeling, have been involved, not only in the use of computer models, but also in their development.

In these models, agents are designed according to the consensual information provided by their real counter-parts and the people they interact with. Decisional or behavioural rules are as complex and rational as the creators wish them to be. Likewise, agent’s beliefs are tailored according to the phenomenological expression of the real stakeholder’s mental models (Dray et al. 2006). Hence, this constructivist and post-normal modelling of cognitive processes doesn’t intend to tell How does it work?, but rather What is there that is so important? But Becu and colleagues (2003) give evidence that the phenomenological expression of personal beliefs and intentions might not provide a consistent enough set of rules or assumptions to be directly encapsulated into the agent.

Recently, a group of scientists, Collectif ComMod, has decided to formalise their approach in order to establish deontic rules for developing companion models by direct interactions with local stakeholders (Collectif ComMod 2005). Companion modelling (ComMod) is an approach making use of simulation models in a participatory way to understand and facilitate the collective decision-making process of stakeholders sharing a common resource. The principle is to identify the various viewpoints and subjective referents used by the different stakeholders, and to integrate this knowledge into simulation models that serve as mediating objects. There is an iterative process of confrontation between factual evidence, model design, and scenario exploration (Figure 3.5). Other mediating objects or methods are usually used in conjunction with computer simulations, like role-playing games (Barreteau 2003; Dray et al. 2006). The different stakeholders, including researchers, aim at working out a common vision of the common resource management that highlights the diversity of interests. This approach has attracted growing attention from decision-makers and community-based organisations in order to engage more dynamically with stakeholders.