© Copyright JASSS

  JASSS logo ----

Olivier Barreteau, François Bousquet, Jean-Marie Attonaty (2001)

Role-playing games for opening the black box of multi-agent systems: method and lessons of its application to Senegal River Valley irrigated systems

Journal of Artificial Societies and Social Simulation vol. 4, no. 2,

To cite articles published in the Journal of Artificial Societies and Social Simulation, please reference the above information and include paragraph numbers if necessary

Received: 01-Nov-00      Accepted: 01-Feb-01      Published: 31-Mar-01

* Abstract

Multi-agent systems and role playing games have both been developed separately and offer promising potential for synergetic joint use in the field of renewable resource management, for research, training and negotiation support. While multi-agent systems may give more control over the processes involved in role playing games, role playing games are good at explaining the content of multi-agent systems. The conversion of one tool to another is quite easy but organisation of game sessions is more difficult.

Both these tools have been used jointly in a fully described experiment in the Senegal river valley for issues of co-ordination among farmers. Role-playing games first enabled us to work on the validation of the MAS. Subsequently, the combination of both tools has proved to be an effective discussion support tool.

multi-agent systems, role-playing games, validation, negotiation support tool, legitimisation, irrigated systems

* Introduction

The use of modelling based on multi-agent systems (MAS) for tackling natural resources and environment management issues is growing steadily (Bousquet 1999). Publications relating to its use have appeared in scientific journals such as JASSS as well as in workshops and seminars, MABS (Sichman 1998) and SMAGET (Ferrand 1999). Just beneath this field of research, Artificial Life (AL) studies may lead to the use of artificial ecosystems to explore management strategies.

These models, based on MAS or AL techniques, are generally confined to the laboratory. To take them out into the field, where natural resource management really takes place, we must be able to explain their content. Thanks to the synergy between the techniques of multi-agent systems and role playing games (Barreteau 1998), we have developed a methodology for opening the MAS black box through role playing games (RPG). Before describing this methodology, we will emphasise the importance of explaining the contents of the MAS and we will finish this paper with an example of application to irrigation systems in the Senegal River Valley.

* Objectives of explaining the content of a MAS

Several uses of a MAS related to natural resource management

Even in the field of natural resource management, several uses of MASs exist. In this paper, we focus on three of them: simulation of ecosystem evolution scenarios for research, training and discussion support purposes. Anything related to problem solving, evaluation or prediction is beyond the scope of this paper.

In the three uses considered above, the MAS is seen as a representation of a real ecosystem, which is itself defined as the relevant scale of evolution in space of the natural resource under study. This is actually a kind of virtual ecosystem on which it is possible to conduct experiments according to scenarios defined by its user(s). These experiments are simulations of the evolution of the real ecosystem under a given hypothesis, just as could be done on a small-scale physical model. What distinguishes the three uses is the purpose of these simulations: research, training, or negotiation support.

Simulation models as research tools
Ecosystems are usually complex systems as far as natural resource management is concerned. Their dynamics may be quite slow. Moreover management questions imply that people are included in the system under study. For these three reasons, it is difficult to conduct experiments on real ecosystems with a view to analysing their management. The need for repeatability and for control of parameters, and the time required to perform experiments explain the failure of such approaches for research purposes.

MAS models make it possible to conduct experiments with fully repeatable and controllable scenarios, of reasonable duration (depending on the quality of the modelling and on the performance of the machine used) and with no potentially harmful consequences for the people living in the ecosystem concerned. Analysis of the results provides a means to develop theories on the behaviour of the virtual ecosystem, which can then be compared to other observations. Some difficulties arise when returning to the real ecosystem to extend these theories to the management of real natural resources. A certain amount of validation may therefore be necessary.
Simulation models as training tools
Simulations may also be used for teaching and training purposes (Hatchuel 1993), either for learning the art of natural resources systems management or for developing new skills in this area.

Many schools of thought exist when dealing with management of natural resources, even for a single given resource such as water for example. These schools often rely upon sets of collective rules intended to ensure the viability of the system. Training in this field, whoever is concerned, has to include a presentation of these sets of rules and of their various potential consequences. The MAS provides a tool for playing with these rules and for exploring the consequent behaviour of the ecosystem, starting out from different initial contexts. They thus offer a better understanding of the complex behaviour of ecosystems. They also give the opportunity to test the sensitivity of the consequences of a given set of collective rules with respect to a set of assumptions on individual behaviours.

All these schools of thought have their weaknesses. New solutions regularly emerge for solving problems encountered by pre-existing methods. This progress in the art of managing natural resources has, until now, followed the method of learning by doing. With this method, people depending on the exploitation of the resource sometimes pay a heavy price for the errors committed. The use of models, such as MASs, is a way to limit the cost of trial and error methods. It provides a means to shift partly from learning by doing towards learning by simulating. We are assuming here that learning in a virtual world may be useful way to learn about the real world, if the trainee is able to see the link between the two.

To be considered as relevant by trainees, any MAS used in a training process must be legitimate. In order to be useful, it must also be at least partly validated, especially to limit the remaining share of learning by doing.

Simulation models as decision support tools
Simulation models are increasingly used as decision support tools. In the case of natural resource management, a decision is seldom the result of one hypothetical decision-maker, it is rather a matter of interactions between several stakeholders (Weber 1995). What are identified as "decisions" emerge from these interactions via a small number stakeholders. Some stakeholders play a special role (managers for example). They are more "decision-inducers" than "decision-makers". Our aim is therefore primarily to support the process which leads to these decisions: to bring it alive, to ensure the suitability and the legitimacy of its results. This raises issues of stakeholder participation in decision making. We thus envisage the use of the MAS as a group decision support tool through better negotiation.

For this purpose, the MAS is seen as a mediating object in the negotiation process (Bousquet, Barreteau et al. 1999). Interactions between stakeholders regarding natural resource management may take several channels: indirect through the perception of the consequences of others' actions on the resource, or direct, either on a one-to-one basis or in institutionalised collective frameworks (Rouchier 1998). All these direct interactions are based on mediating objects. A MAS constitutes such a mediating object, inducing stakeholders to agree on a common representation of their joint natural resource system and facilitating communication in relation to it.

It actually serves as a discussion support tool, by providing a representation and a simulation of proposed scenarios of collective rules for common natural resource management. It offers an additional channel of interaction between stakeholders which is more controllable and more "open". It may thus prevent dead-end conflicts due to misunderstandings.
By using simulation models as discussion support tools, be they MASs or other simulation systems, the problems encountered by and known to every stakeholder, though diluted in time and space, are brought to common knowledge, thereby preventing certain misrepresentations or bad faith behaviours. The use of a simulation tool leads to the concentration of selected processes in time and space. This causes each stakeholder to modify his/her representations, leading to a better knowledge of the representations of others. Shared representations mediate this evolution towards an improvement in stakeholders' cognitive capacities (Teulier-Bourgine 1997). The legitimacy of the decision process will then be enhanced through a better knowledge of its grounds and construction rather than its supposed optimality (Funtowicz 1999).

Misrepresentation or bad faith behaviours may nevertheless reappear in the discussion of the model's validity in representing reality. An agreement between stakeholders on the model and its capacity to represent reality is necessary as a first step, with a view to legitimising its use in the negotiation process. For this use, this once again raises the question of explaining the content of the model to stakeholders in order to partly validate and legitimise it.

Validation and legitimacy of the MAS for three uses: research, training and decision support

 The three uses of a MAS as a model for natural resource management require at least partial validation and/or legitimisation for stakeholders. For all of them, an explanation of its content is an interesting way to proceed.
Validation for research purposes
 The use of MASs for research purposes leads to the production and validation of theories. The validation of these theories relies upon finding a match between observed and simulated results as well as between modelled and real processes (Barreteau 1999). As far as natural resource management questions are concerned, relevant results to be validated are mainly dynamic and therefore difficult to observe. One way to go about this is to compare them with other simulated results which are easier to compare to real dynamics. To validate the representation upon which the MAS is based, the capacity of the set of selected processes to display the same patterns of interactions must be tested. Through the use of this model, have we grasped the right level and the right angle of complexity for the question in hand?
 In both cases, the role-playing game appears, a priori, to offer an attractive solution. It is possible to record and partly control its progress in order to compare it with simulations on MASs and if players and/or observers are some of the stakeholders involved in the natural resource management, they provide a means to go back to real natural resource management dynamics. They also constitute a way to open the black box of assumptions underlying the model, to present it to represented stakeholders so that they can understand and discuss it (Herimandimby 1999). Their discussion of this set of assumptions partly fulfils our validation objectives.
Validation for training purposes
 For training purposes, validation is more important since it may concern people who do not know the real system and who may not trust a model proposed by the instructor. The advantage of using a MAS in a training process is its capacity to explore virtually the consequences of different scenarios, including collective rules and individual behaviours, but also to provide an insight into the processes involved in system dynamics. The MAS used should therefore have been partly validated relative to the set of assumptions as well as to the simulation results, as is the case in the validation for research purposes described above.
Validation is not necessarily so important if we address the patterns rather than the structure of the system. Here, in fact, we assume that the closer the MAS to the real system from the structural point of view, the stronger the validation must be. We should stick to the heuristic value of the model: it is the trainees' job to make their way from the model to real systems. This is facilitated if the model is not a black box for them.

As, in this case, validation is of similar importance to validation for research purposes, we again propose to use RPGs to explain the model content. Moreover, they may also address the problem of ensuring that the model is considered legitimate by trainees in order for the training to be fruitful, since RPGs can also be used with them as players.
Validation for negotiation support purposes
For negotiation support purposes, the main aspect of the content requiring explanation is its legitimacy in the negotiation process. This legitimacy is required as soon as the first, and perhaps most important, part of negotiation begins: the building of a joint representation of the system (Boltanski 1991). Cases where the building of the model used in the negotiation was not made public to all stakeholders are reported to have failed (Reitsma 1996).

One way to achieve this legitimacy is, once again, to open the black box of the model, to reveal the contents and to ask stakeholders involved in the negotiation whether the model's assumptions match their own representation of the system dynamics and provide agents with a large enough range of possible actions. The model alone may be of little interest in many contexts when stakeholders do not have the time or the education to understand and to appropriate the often badly-written code of the model. Opening the black box with RPGs, with stakeholders as players, may therefore help to reach the objective of making realities of assumptions and of inducing discussions upon a jointly accepted representation of the system.

The RPG thus appears, a priori, to be an interesting tool for these three different uses of MASs for natural resource management. But it raises another question: are MASs still necessary for these three uses or could RPGs be sufficient per se ? Actually they constitute models in themselves and have already been used without the support of a MAS in all of these uses. RPGs are indeed a quite common tool for training and teaching purposes. For example, several games have been devised for the management of irrigated systems, (Burton 1989). Some collaborative methods use RPGs in negotiation (Heathcote 1998). And experimental economics provide us with examples of uses of RPGs to deal with theories, for example, again in the field of water management, of water management as a common-pool resource (Ostrom 1994). In another experimental economic study, dealing with reciprocity, instead of optimisation, in human economic behaviours, RPGs provide some features that MASs lack: they may profit by the players' own personal history (Gintis 2000), an interesting feature for dynamic modelling.

But as the sole modelling tool of a method, they are reported to be limited as they are cumbersome and slow to develop and analysis of their results is still difficult (Piveteau 1995), especially in the case of complex systems. Moreover comparison between different experiments is difficult since many parameters in a game are not controlled. We therefore chose to explore the potentialities of a method based on MASs, for their control and representation of interactions and dynamics capacities, and RPGs, for their openness and capacity to initiate discussion. This coupling of both tools in field conditions is the counterpart of an ongoing process in experimental conditions. In a recent research project on monetary theory, a game is used in association with an Agent-Based Model. In a first stage, the model helps to design the game and the game supports the legitimacy of the model. The model is then used to explore the scenarios more rapidly and more systematically (Duffy 2001). The results of experiments with the game and with the model are comparable.

* From MAS to RPG

 Such a method relying on MASs and RPGs to deal with natural resource management questions depends on several elements: joint field study with design of MASs (Bousquet, Barreteau et al. 1999), design of the game content and organisation, feedback from game experiments to the MAS and the field. We will focus mainly on the second element and partly on the third: how to convert a MAS into a game, how to organise the progress of the game, how to combine both tools in the method.

Conversion of MAS components to design a RPG

 A first step in the design of the game is the conversion of objects, agents and rules included in the MAS into the game support, players and role cards, with a constraint of obtaining a "playable size" of game. The initial idea is to consider the RPG as a living MAS in which players are the agents and the set of roles is the rule base. But in some cases, including the one described below, the number of agents and the number of rules available to them is too high. So certain size parameters of the MAS have to be changed, with the constraint of keeping what is interesting in the complexity of the MAS for the question in hand. As is the case for model design through interaction with field study, this simplification may be achieved in several steps of interaction, profiting from feedback from players and simulating the consequences of simplifications on the MAS. In this case, the conceptual model sustaining the RPG is a simplified version of the one sustaining the initial MAS.

 To this end, we used the concepts taken from studies on good RPG practice to perform this conversion of the MAS down to a reasonably sized RPG. These studies consider game design as the art of reaching the right balance between several opposing elements rather than as a science. Nevertheless a few recommendations are formulated: imitate what is recognised as amusement in other RPGs, insist on mechanisms of cooperation and competition, create the illusion through a small number of carefully chosen symbols, include elements of uncertainty, be careful about involving all players and pay attention to time management (Mauriras-Bousquet 1984).

 Since dynamic processes are involved, which represent one of the advantages of using MASs and one of the reasons for resorting to RPGs as a validation tool, the question of the conversion of time management parameters must be dealt with carefully. In MASs, the speed of computer calculations makes it possible represent the system with relatively short time steps, and simulations with a large number of time steps can be produced. In RPGs however, actions last longer and too much repetition of a given action may lead to boredom among players.

For time representation, like for game complexity, our goal in converting MASs into RPGs has been to stick with certain principles of a playable RPG: it must be feasible to organise half-day sessions including a game learning stage.

Organisation of the game

The second set of questions to be solved when converting MASs into RPGs concerns the organisation of game sessions. Who should take part?, How should the game be arranged? How long should the game sessions last?

Given the necessarily limited number of players in a given RPG session and the risk of boredom, the set of participants may be adjusted according to the purpose of each session: validation of a MAS, presentation of a model of a particular natural resource management case, enhancement of discussions about collective rules. Most players may be stakeholders of the real system if the game is played for validation or negotiation support purposes, observers or experts of the real systems such as administrative officers, researchers, developers, if it is played for validation purposes, trainees if it is played for teaching purposes, or "neutral" players if it is played for validation and negotiation support purposes.

Participants of a RPG session are not only players, but also observers of the session. For validation and negotiation support purposes, players may be neutral and observers may be stakeholders in order to prevent strategic behaviours which could modify the progress of play. The game must win the trust of stakeholders. The role of another participant, the game master, is very important at this stage: his role is to manage the group, keeping them amused while preventing a too large gap from forming between the on-going processes and what the game is designed for. If players are rather neutral, the RPG tends to resemble a stage play. When participants are stakeholders, the question of their homogeneity must also be taken in account since the mixing in a given RPG of superiors and subordinates is reported to be problematic (Mauriras-Bousquet 1984).

The place where the game is played is another element which can also be adjusted to represent certain characteristics of a real system viewed as particularly important. For example, should a given game be held in a single place or should the game areas be physically separated? The case study below provides an example in which two areas are required because the presence of stakeholders in one place or another is very important in real processes which are represented in the MAS.

There is also a symbolic component to the place where the session is held, especially when stakeholders are involved. For validation and negotiation support purposes, the links which may habitually exist between the players and the place must be taken into account.

The materials used should be as simple as possible to ensure that players become rapidly involved in the game, especially if we expect them to become game managers.

Regarding the duration of the session, if the RPG is somewhat complicated, it may be useful to separate the learning stage from the game process stage, though this makes the session even longer. For the experiment described below, the aim was to achieve a balance between simplicity and experimental depth.

Combining both kinds of tools

Both tools, MASs and RPGs are linked in a method. Beyond a merely linear process implied in the term "conversion", we see it as a cyclic process, based on the cycle used in the modelling process (Bousquet, Barreteau et al. 1999) as represented in Figure 1.

Figure 1. association of MAS, RPG and field observations in a two-cycle method

These two cycles are designed to allow feedback to the real world and stakeholders and to promote communication. The use of simulation results is a first means, more dedicated to testing consistency of assumptions and to the identification of missing knowledge. The RPG is an additional means to communicate about the model with stakeholders. This communication may result in more than corrective modifications of the current MAS or RPG model. According to the stage and the purpose of these tools, several successive versions may exist and be used.

* Application to irrigated systems in the Senegal River Valley

This method is currently under test with an application to Senegal River Valley irrigated systems. These systems, built less than thirty years ago, have produced disappointing results and present a variety of problems. To analyse the reasons behind poor system viability, we started by looking for a tool to test the hypothesis of a link between stakeholder co-ordination and this lack of viability. Since the question was highly dynamic and focused on interactions, we chose to use a representation of systems with a MAS. The first use in need of explanation to stakeholders is thus the use of the MSA for research purposes, though it has also been used as a teaching tool and is currently being considered for use, coupled with a RPG, as a discussion support tool, actually a first step towards its use as a negotiation support tool.

The models, MAS and RPG

Description of the MAS model
A first model, called SHADOC [1]and fully described elsewhere (Barreteau 2000), was then built in interaction with a field study in the Senegal River Valley. Based on the assumption that irrigated systems are a place for acquisition and allocation of water and credit, it represents an archetypal irrigated system in the Senegal River Valley.

The simulated scheme is represented in the object diagram (Rumbaugh 1991) [2] in Figure 2. It is composed of a pumping station which supplies water to a main canal which delivers water to a collection of 5 watercourses along which a number of plots are situated. Water allocation among the watercourses depends on their gate position and on the amount of water available from upstream to downstream. Water allocation among open plots along the same watercourse is represented so that upstream plots receive more water than downstream ones; each open plot receives twice as much water as the following one downstream along the same watercourse.

Figure 2. scheme representation structure (from Barreteau & Bousquet, 2000)


The simulated society is also represented by the object diagram in Figure 3. Each farmer is assumed to belong to three groups related to the irrigated system, one for water allocation among plots along the same watercourse, one for credit matters and one for pumping management. Each farmer has his own representation of the system, including the supposed state of his plot, the supposed water availability and a list of other farmers with whom he is ready to cooperate. This last component of each farmer's representation of the system consists in a friendship group which is specific to each individual, within which he may exchange information and services [3]. Each farmer has also a goal, among four categories identified in observed fields, which specifies the relative importance he gives to his plot: amount of inputs to invest and priority of irrigation of this particular plot in relation to his other activities. Beside these individual behaviours, individual characteristics also include a social status in a strictly hierarchical society.

Each group has its own autonomy as an agent and therefore may act and communicate with other agents, groups and farmers. It may be described with its grounds and its list of members. It also has a representation of the system, in which it keeps a memory of the state of its relations with other agents with whom it is in relation, such as potential debts and debtors within a credit group. Finally, each agent, farmer or group, has a set of rules which describes how it shall behave. The purpose of each rule in the set depends on the agent concerned: individual rules for farmers or collective rules related to their grounds for groups. Among these rules, each agent has a rule which evaluates his own criterion regarding the course of a cropping season and another to change the rules if the criterion is not satisfied.

Figure 3. object representation of a simulated society (from Barreteau & Bousquet, 2000)

Dynamically, simulations are based on several time steps in which agents make their different choices:
  1. the day for choice of activity
  2. the cropping season stage for the group of rules to activate
  3. the whole cropping season for changing the set of rules
One simulation may link several cropping seasons and stops on the user's request or when the irrigated system is not used anymore. Typically, a simulated cropping season begins with a credit research stage during which farmers ask for credit either to their group or to other farmers, according to their representation of their chances of success. When a farmer thinks he is able to begin the cropping season, he sends a message to the group which manages the pumping station. If its corresponding rule is satisfied, the group allows him in return to get water during the season. This group also has a rule to start the pumping station: pumping is activated when, according to the group's representation of the system, the starting conditions appear to be fulfilled. The authorised farmers then begin to work on their plot: sowing then irrigating according to their presence on their plot, their right to take water and its availability. By the end of the season, each farmer who has been able to sow a plot of rice gets its yield [4]. Everyone who obtained credit may reimburse it if he is able to and if this course of action is among his preferences. Every agent then assesses the course of the season and, if not satisfied, changes his rules for the following season, imitating another agent who seems to have obtained good results. This imitation consists in switching the whole set of rules to the set of the imitated agent, the choice of whom depends on a rule of changing rules: the best one or the nearest good one (in the space of rules). A new cropping season may then begin.
Conversion of SHADOC in a RPG
To convert SHADOC into a RPG, a simplification stage was necessary, to avoid the need for play sessions involving some sixty people, repeating similar choices more than a hundred times with a large number of possible rules. We thus chose to: The conceptual model presented is thus simpler than the one supporting SHADOC. We then built a second MAS, initially named "ShadocLight", based on the same conceptual model as the RPG.

Players take the roles of farmers, each cultivating a plot in the same irrigated scheme in the Senegal River Valley. They are situated in a first room which represents the space of villages. This room contains two sets of tables, representing named villages in the game and equivalent to friendship groups in the model. In this same room, two other single tables represent the two groups in the game, who implement the collective rules. At the beginning, each player carries a random set of three cards on the basis of which he or she behaves throughout the session. These cards reflect the attributes considered important by the simulation process and the feedback of a first batch of play sessions. The attributes concerned are:

These cards are trilingual: pulaar, wolof, French. These are the commonest languages spoken in areas where the RPG has been played. For teaching purposes, an English version is being developed. The goal cards are shown in Figure 4 below.

Figure 4. set of goal cards among which players have to choose randomly

The scheme is drawn in a second room on a board according to the number of players. Plots are separated along two watercourses supplied by one pumping station. Each plot is initially in an open state. During the season, farmers' choices make them open or close. A chart is used to determine water allocation among plots when the pumping station is in operation. It is based on the same pattern as in the MAS and depends on the number of plots in the open state along the water course and the relative place of each given plot. Each plot is used to grow rice and is characterised during the game session by a water level and a state of cultivation including a date of sowing, a choice of variety from among two possibilities and a level of inputs utilised according to the farmer's goal.

Figure 5. representation of the scheme on a blackboard during a game session in Senegal

This board represented in Figure 5 is located in a different place from where the players are mainly based, representing the villages area. The two places may be near one another but out of view. There are two distinct villages separated from one another and with relations of conflict. Players belong to one of them randomly and independently of the location of their plot in the scheme.

Some collective rules are adopted by certain players chosen by the assembly: one person in charge of regulating water allocation along each watercourse, one person in charge of pumping station management and water fee collection, one person in charge of credit accessing and allocation. These players take over this collective role in addition to their own individual role, and they randomly choose a card specifying that role also.

A game comprises several cropping seasons and each season is divided into three stages: credit access, irrigation management, season closure and assessment. The average of eight turns constituting the representation of the cropping season arises from a compromise between the fact that players need water at least twice and the need to avoid making the game boring with too many identical turns. The player in charge of the pumping station may slightly change this number of turns, and may or may not inform other players. At each turn, each player draws an "occasion card" which may or may not allow him/her to go to the scheme area according to the goal card. Those who are allowed to, go to the board and choose: It is possible to work for others and to lend money according to each individual's social status.

For research purposes

Simulation results obtained with SHADOC are consistent and can be used to classify scenarios of collective rules and hypotheses of individual behaviours into classes of viability (Barreteau 1998). It was nevertheless important to work on the validation of the MAS to gain the trust on the methodology and on the researchers and thus to strengthen the theories which may be elaborated.
This validation was made through the use of the RPG described above in 5 short half-day sessions with stakeholders of the irrigated systems [5] upon which the MAS was based: 4 within each studied irrigated system and one within a federation of village development associations of the area with, among others, a goal of supporting the development and the management of irrigated systems. The latter took place first and also had a goal of adapting the game: it led, for example, to the elimination of a few rules initially kept in the game, considered by the association as not so important.
These sessions, which also had a goal of simply presenting the results of a three-year study which had involved several people of these systems through interviews, concluded in a satisfactory representation for stakeholders. They considered the game seriously and were sometimes quite concerned to make sure that our portable board was cleaned up before we left, so that people from outside their real system (people from other villages for example) could not see how they behaved. The dynamic validity of the MAS could not however be tested because the sessions were too short. They were devoted to discovering and learning about the RPG and never went beyond the second cropping season.

In order to go further with this validation, a three-day workshop was organised at an inter-professional agricultural training centre [6] in October 1999, with 3 participants as observers, two from the training centre with knowledge of irrigation in the valley and one from a French research institute with expertise in decision support systems and 10 participants as players from different places in the valley and different roles: farmer, persons in charge of irrigation association, rural activists, agricultural advisors. This wider range of participant origins enabled us to test the genericity of the representation at the scale of the valley. The length of this session allowed us to organise a game learning stage and a full play stage with a larger number of simulated cropping seasons .
The validity of the representation, i.e. of the assumptions in the RPG, was more strongly challenged by this group of participants, especially the importance attached to social status. According to the two local observers, this may be due more to politically correct behaviours among people who had never met each other before, hence giving greater weight to their social and political position [7], than to a more fundamental disagreement with the model. There was more agreement among participants, on the basis of their own experience of irrigation systems, regarding the validity of the dynamic aspect of the simulation. Nevertheless it was somewhat distorted by other politically correct behaviours which induced players to select mostly intensive cultivator and good payer profiles, which is not always the case. So we did not make a homogeneous exploration of the whole set of scenario.

For training purposes

Up to now, the MAS and RPG versions of SHADOC have been used for training purposes in two kinds of situations always with a student audience: in training sessions on renewable resource management and in training sessions on MASs with renewable resource management. These training sessions have taken place mainly in France, but also in South Africa, Thailand and the Philippines, and have always received goodfeed back from trainees who took part in the game.
In both cases, the main aim of sessions was to present the complexity of managing a renewable resource and the absence of a single answer to the question of good management. They also aimed to present possible tools available to address this question. But they did not go as far as the goal of learning by simulating.

For discussion support purposes

When organising the sessions with stakeholders or experts of the irrigated systems in the Senegal River valley, we received feedback on several occasions, even in the first batch of short sessions, indicating that it was an interesting tool for initiating discussion among stakeholders regarding their collective behaviour in the system. The most striking effect of all these sessions was their stimulation of discussion among participants about their own experience of these systems. They discussed certain assumptions in depth and came to an agreement on a representation of irrigated systems.

By the end of the 1999 workshop, they agreed to give both MASs and RPGs a local name, "njoobaari ilnoowo", which means: "the minimum package an irrigator brings with him when he travels", which can be roughly translated as the "irrigator's overnight bag". This indicates a perception of both tools as a support tool for irrigation system management, as well as an appropriation of the tool. This appropriation is confirmed by the request of several players to keep the game with them to use it in their own villages to support discussions about their irrigated systems.

There has not yet been any real use as a discussion support but advances are being made in this direction. Moreover, during the 1999 workshop, the simultaneous presentation of the RPG and the lighter version of the MAS corresponding exactly to the RPG show that, even if the use of the computer-based tool to explore scenarios in this context was not yet ready, there was a good understanding of the MAS and its possibilities after playing. People who had played the game, were at ease with choosing scenarios and discussing simulations results.

* Conclusion and perspectives

An interesting methodology to emphasise

The association of an RPG with a MAS, through this experiment in Senegal River valley, seems to provide a good way to explain the content of a model in order to validate it and to communicate upon its basis. Further work must be done to develop in-depth joint use of the two for teaching and negotiation support purposes. We must therefore tackle the issue of the possible polymorphism of these tools in terms of objectives and the question of game session recording techniques.
The use of the RPG itself in this context and with these objectives has produced interesting results. Players have understood the RPG and the corresponding MAS reasonably well, have found ways to initiate discussion of their real systems through its use, have envisaged the possibility of using it themselves... and have had fun, which is a precondition of its playability.
Nevertheless a few lessons may already be drawn up from this ongoing experience. The most important criticisms received during validation use of the RPG concerned socially sensitive characteristics of the model, such as castes or the existence of land-keeping objectives among farmers. A partial solution to this method was given during the 1999 workshop about the question of castes. There was absolute refusal to take them into account as a reality in relation to irrigation systems. But it was acknowledged that other kinds of social hierarchies have appeared and have the same supposed effect on the dynamics of the system for exchange of money and labour. This has led us to use the archetypal representation of social relations (hierarchy, equivalence networks) rather than over-particular ones which could raise sensitive issues (such as castes in Senegal). Particular instances of the archetype of social relations may simply be mentioned as examples in order to explain what they represent, the choice of the right example being made by the game master according to the participants involved.
Our experience up to now has shown that the main aspect of RPGs which enhances discussion among game session participants is the way in which, thanks to the unity of time and space in the RPG, the problems encountered in the field and known by each individual separately are turned into common knowledge.
This synergy explored in this experiment from MAS to RPG offers additional potential by suggesting alternative ways of combining both tools. Other ongoing experiments are focusing on the value of RGPs as a collective interview method. The RPG is also an useful tool for informing a MAS (Lynam 2000).

Looking towards experimental economics

The conclusion above on the movement towards RPGs with archetypal social relations goes in the same direction as recent work in experimental economics. For a game to test the "ratchet effect", several protocols have been tested with, among others, managers already aware of this effect or students ignoring it and context-explicit or generic language. When players are aware of the context, this context, if explicit, has a strong effect on the process of the game, while generic language makes the game more independent of the players' experience (Cooper 1999). For the organisation of the session, between the extremes of a stage play and too much uncontrolled drift with respect to the initial design, the role of the game master is very important. Thus, up to now, the design of RPG from a MAS remains mainly empirical and follows the viewpoint that RPG design is an art rather than a science, in the same way as teaching games (Mauriras-Bousquet 1984).
However in experimental economics, whose objectives are the same as ours, except for the identification of population attributes, the most highly criticised aspect in any case, considerable expertise is being developed in the design of experimental games (Friedman 1994). The success in combining games with agent-based modelling shows that the avenues explored in this paper are promising, since the use of the model is a good support for designing experiments (Duffy 2001). It goes one step beyond the design processes based on a theory to be tested: games are thus well controlled but rather simple. Simulations make it possible to go further in the exploration of these theories and their consequences.
Games with experimental economics still have their limits: many authors report difficulties in taking communication into account (Verstegen 1998;Duffy 2001) while MASs and other agent-based models do so easily. RPGs in field studies, as the one described in this paper, achieve more communication among players, but it is not always controlled, and it may become difficult to analyse.
Another currently mentioned limit to experimental economics is the motivation of players who are usually paid an amount for their participation which depends on their gains in the game. Besides the fact that it makes the implicit assumption that players all aim to maximise their gains, it does not really guarantee their motivation to play under strictly controlled conditions. In a field study context, it is less difficult to motivate players who are participating freely, at least when they are concerned by the processes at stake in the model. But it is then necessary to cope with strategic behaviours of players who might be interested in a specific end to the game. Thus both controlled experiments and field games should enrich one another, learning from the parallel use of simulations with MASs or other agent-based models.

* Acknowledgements

The authors wish to thank Sophie Thoyer and one anonymous referee for useful comments.

* Notes

1 French acronym for hydro-agricultural simulator describing organization and coordination modes.
2 This diagram represents most important classes with some of their attributes, the relations between them and the multiplicity of these relations
3 This friendship group is actually a projection in the irrigated system of groups or partitions of population which exist in the local society: age groups, villages, political leanings...
4 This yield depends on several factors. When the plot is sown, the yield is initiated to a potential value depending on the level of inputs invested and on the choice of a variety (among three: the longest the most productive). Then each event of water stress during the season, as well as any too long delay for harvesting, induces a drop in this potential yield according to FAO equations (Doorenbos 1979 ). By the end of the season, the resultant potential yield is multiplied by a random factor to represent other events not simulated but with effects on real yields.
5 Except one in which the organisation of a game session has not been possible for reasons of calendar.
6CIFA, St Louis, Senegal
7 This workshop took place in the context of preparation of presidential elections in Senegal, and all that concerns system of casts is officially supposed not to exist any more.

* References

BARRETEAU O (1998). Un système multi-agent pour explorer la viabilité des systèmes irrigués : dynamique des interactions et modes d'organisation. PhD thesis from Ecole Nationale du Génie Rural, des Eaux et des Forêts. 260 p.

BARRETEAU O and Bousquet F (1999). Jeux de rôles et validation de systèmes multi-agents. Gleizes M-P and Marcenac P (Eds.) Ingénierie des systèmes multi-agents, actes des 7èmes JFIADSMA: Hermès : 67-80.

BARRETEAU O and Bousquet F (2000) SHADOC: a Multi-Agent Model to tackle viability of irrigated systems. Annals of Operations Research 94. pp. 139-162.

BOLTANSKI L and Thévenot L (1991) De la justification, les économies de la grandeur. Gallimard.

BOUSQUET F, Barreteau O, Le Page C, Mullon C and Weber J (1999). An environmental modelling approach. The use of multi-agent simulations. Blasco F and Weill A (Eds.) Advances in environmental and ecological modelling: Elsevier : 113-122.

BURTON M A (1989) Experiences with the irrigation management game. Irrigation and drainage systems 3. pp. 217-228.

COOPER D J, Kagel J H, Lo W and Gu Q L (1999) Gaming against managers in incentive systems: experimental results with chinese students and chinese managers. The American Economic Review 89. pp. 781-804.

DOORENBOS J and Kassam A H (1979). Réponse des rendements à l'eau. Rome, FAO.

DUFFY J (2001) Learning to speculate: experiments with artificial and real agents. Journal of Economic Dynamics and Control 25. pp. 295-319.

FERRAND N (Ed.) (1999). Modèles et systèmes multi-agents pour la gestion de l'environnement et des territoires, Cemagref éditions.

FRIEDMAN D and Sunder S (1994) Experimental methods, a primer for economists. Cambridge University Press.

FUNTOWICZ S O, Martinez-Alier J, Mundo G and Ravetz J R (1999). Information tools for environmental policy under conditions of complexity, European Environment Agency, environmental issues, n°9.

GINTIS H (2000) Beyond homo economicus: evidence from experimental economics. Ecological economics 35. pp. 311-322.

HATCHUEL A and Moisdon J-C (1993) Modèles et apprentissage organisationnel. Cahiers d'économie et sociologie rurale pp. 17-32.

HEATHCOTE I W (1998) Integrated watershed management, principles and practice. John Wiley & sons.

HERIMANDIMBY V, Randriarijaona E, Barreteau O and Bousquet F (1999). Systèmes Multi-Agents et démarche patrimoniale : utilisation de jeux de rôles. Ferrand N (Ed.) Modèles et systèmes multi-agents pour la gestion de l'environnement et des territoires: Cemagref éditions : 261-278.

LYNAM T, Bousquet F, d'Aquino P, Barreteau O, Chinembiri F and Mombeshora B (2000) Adapting science to adaptive managers - spidergrams, belief models and multi-agent systems modelling. submitted to Conservation ecology.

MAURIRAS-BOUSQUET M (1984) Théorie et pratique ludiques. Economica.

OSTROM E, Gardner R and Walker J (1994) Rules, games and common-pool resources. The University of Michigan Press.

PIVETEAU V (1995) Prospective et territoire : apports d'une réflexion sur le jeu. Cemagref éditions.

REITSMA R, Zigurs I, Lewis C, Wilson V and Sloane A (1996) Experiment with simulation models in water-resources negotiations. Journal of water resources planning and management 122. pp. 64-70.

ROUCHIER J, Barreteau O, Bousquet F and Proton H (1998). Evolution and co-evolution of individuals and groups in environment. ICMAS'98, Paris, IEEE Computer Society.

RUMBAUGH J, Blaha M, Premerlani W, Eddy F and Lorensen W (1991) Object oriented modelling and design. Prentice Hall.

SICHMAN J S, Conte R and Gilbert N (1998) Multi-agent systems and agent based modelling. Springer.

TEULIER-BOURGINE R (1997). Les représentations: médiation de l'action stratégique. Avenier M-J (Ed.) La stratégie chemin-faisant: Economica : 95-135.

VERSTEGEN J A A M, Sonnemans J, Huirne R B M, Dijkhuizen A M and Cox J C (1998) Quantifying the effects of sow-herd management information systems on farmers' decision making using experimental economics. American journal of agricultural economics 80. pp. 821-829.

WEBER J (1995). Gestion des ressources renouvelables : fondements théoriques. Montpellier.


ButtonReturn to Contents of this issue

© Copyright Journal of Artificial Societies and Social Simulation, 2001