Page 265 -

264 O. Barreteau et al.

less formal, but the work of translation is less important. However, control of the
process still remains largely in the hand of the modeller but to a lesser degree than
in previous examples. This technique was further associated with semi-automatic
ontology building procedures by Dray and colleagues in order to generate collective
representations of water management in the atoll of Tarawa (Dray et al. 2006).
With inspiration coming similarly from the domain of ethnography, Bharwani
and colleagues have developed the KNeTS method to elicit knowledge. Apart from a
first stage with a focus group, this method is also based on individual interviews. As
in Becu’s work, interaction occurs in two phases: elicitation through questionnaires
and involvement in the model design at the validation stage, which is also considered
as a learning phase for stakeholders. These authors used an interactive decision
tree to check with stakeholders whether the output of simulation would fit their
points of view (Bharwani 2006). Control of this process is on the modeller’s side.
The stakeholders’ interaction is marginally deeper in the model than in previous
examples, since there is a direct interaction with the model as in management flight
simulator. On the other hand, the ontology which is manipulated seems to be poorer,
since the categories of choices open in the interaction are rather reduced. The format
of information is open in the first phase and very structured in the decision tree in
the second phase. The structuration process used in the modelling process occurs
outside of the field of interaction with the stakeholders.
On its side, group decision support system design domain is based on a collective
interaction with stakeholders as early as the design stage. These systems tend to
be used to address higher-level stakeholders. In the method he developed, ACKA,
Hamel organised a simulation exercise with the stakeholders of a poultry company.
In this exercise, the participants were requested to play their own roles in the
company. He constrained the exchanges taken place during the exercise through the
use of an electronic communication medium so that he could analyse them and keep
track of them later. All of the participants’ communication was transformed into
graphs and dynamic diagrams (Hamel and Pinson 2005). In this case, the format of
information was quite structured.

12.3.3 From Software Engineering

Close to the artificial intelligence trend, working like Hamel and Pinson on the
design of agent-based models, there is an emerging trend in computing science
based on agent-based participatory simulations (Guyot and Honiden 2006)or
participatory agent-based design (Ramanath and Gilbert 2004). This trend focuses
on the development of computer tools, multi-agent systems, which originate from
software engineering. Guyot proposes the implementation of hybrid agents, with
agents in the software controlled by real agents, as avatars (Guyot 2006). These
avatars help the players’ understanding the system (Guyot and Honiden 2006). They
can be thought as learning agents: they learn from choices of their associated player
and are progressively designed (Rouchier 2003). The approaches working on hybrid