Complex systems science is largely based upon the use of formal models, which may be very useful in addressing different problems. Yet, models must be “handled with care” as they are not recipes for the future, but rather tools to try to imagine possible futures, and to identify their features.

Dynamical models for decision support have been around for a long time, since the days of Forrester's system dynamics. While they have proven effective to describe some specific subsystems (e.g. the financial one), their usefulness for capturing the web of feedbacks in a complex organization like a company has been much more debated. Interestingly enough, one of the major areas where modelling has been applied is the educational one. Building a model of one's organization, or market, may be an illuminating experience, moreover playing with the model allows one to develop an intuition for possible behaviours which are infrequent in everyday experience.

Many of the early modelling approaches emphasized feedback mechanisms, which may be responsible for unexpected behaviour. However, in these models the "agents" which represented humans or organizations had fairly stereotyped and limited behaviours, lacking sophisticated information-processing capabilities. Moreover, the behavioural rule of the agents were defined in the beginning, and there was no adaptation of the rules themselves in response to interaction with the environment and with other agents.

Agent-based models, which are one of the workhorses of Complex systems science, are nowadays endowed both with sophisticated information-processing skills and adaptive rules.

The models of complex systems, provided that an appropriate use is made, allow powerful analyses to be carried on. Relevant Complex systems science tools include:

• agent-based models
• cellular automata
• adaptive systems, e.g. genetic algorithms and genetic programming
• network analysis

What may be very important for high level managers and trainers is also the fact that studying and using these tools makes one aware of their usefulness but also of their limitations. One must be very careful in interpreting the model results, they very often are not simple "predictions of the future", but rather ways to unfold the consequences, at a global level, of our hypotheses about the way in which our system actually works. So their output must not be considered like an oracle's statement, and must be "handled with care". Nonetheless, they contain useful information to foresee possible futures.