A first layer will be an artificial world (AW), a model of a real complex system. The system to be used as an example will be a model of a region operating in the both the socio-economic and environmental dimensions, which can be modeled using agent-based approach. This artificial world will serve as a source of data.

At the second layer, an artificial intelligence (AI) will use the data to forecast the key features of the future of the artificial world. In the beginning very rough characterizations of the future (for example, survival or collapse) will be considered. Later on, the resolution scale will be refined. The AI will be trained using a set of training runs of the AW (in contrast to real systems, the AW can be run many times, providing a reliable test-bed for training). Simplified mathematical models, empirical stochastic models, pattern recognition methods and other techniques will act as the artificial intelligence.

At the third layer, an artificial statistician (AS) will assess the reliability of the forecasts given by the artificial intelligence, using a series of test runs of the AW.

DV is an agent-based model of a regional development covering several features, such as heterogeneity and bounded rationality of agents, stochasticity in implementing the decisions, interactions of agents and their adaptation and learning. Agents are individuals, companies and the government acting in the physical environment. As in the real world, the agents produce and consume goods, use natural resources, make decisions, respond to incentives and regulations and experience feel feedback from the environment. Current model structure and interactions in it can be clearly described by a UML Classes Diagram.

Each class is shown as a rectangle with state variables in its upper part and its actions in the bottom part. The arrows represent the possible interactions between classes' members.

Workshop #1

On August 11, about 30 IIASA staff members, associates and YSSP students met for a First ASA Forum workshop to brainstorm on the structure of a DV agent-based simulation model. The model will capture the key socio-ecological interactions and economic-environmental feedbacks in an artificial Dream Valley World. Later on, the DV model will be used to test theoretic models and data-based forecasting techniques.

As a result of the workshop, the participants proposed the following groups of agents that should act in the model: Individuals; Households; Social Services; Economic Services; Education; Entertainment; Farms; Industry; Government; Non- governmental Organizations; and Environmentas well as a simple version of a network structure for the DV model emerged on the blackboard (see Figure).

While it is common for such participatory workshops to last several days, we accomplished our two primary objectives within the 2.5 hour time constraints. 1) Those present experienced first-hand a participatory modeling session, and 2) our DV model was informed and enhanced by the collective knowledge of the participants. This first approximation model generated during the workshop will then be further developed by the DV team, and others who expressed desire for continued involvement in the project.

Workshop #2

After a year, on August 13 the second meeting of project developers, experts and YSSP students took place. At this workshop, following an overview of Agent Based Modeling and Cultural Theory, participants constructed scenarios for government decision-making and predicted some outcomes within the artificial Dream Valley environment. Then the proposed scenarios were run on the version 3.0 of the DV Simulalor and it gave some output data. An expert panel assessed the validity of the models simulations in context of realistic system expectations. As a result, it was clear that after a year of work on the project, the simulation software became much more complicated and sensitive but still the calibration process of mechanisms in the Simulator is one of the most demanding tasks during development of the project. The experts also suggested plenty of improvements for the further development of the model.