International Encyclopedia of Systems and Cybernetics

The progressive self organizing mode of a system or model composed of numerous elements each of which endowed with limited elementary capabilities and able to interact in a not absolutely determined way.

This notion has been described as follows by J. ERCEAU and J. FERBER in their study on Distributed Artificial Intelligence:

"Agents endowed with an elementary behavior potential, of the type stimulus-action (for example: to be attracted by a signal) are able to interact. From these local interactions, structures arise, which appear as globally organized. "Intelligence" thus emerges from the interactions of a great number of agents, individually devoid of "intelligence". An analogy sometimes used here is the anthill, the termitehill or the beehive" (1991 ,p.755)

Furthermore: "The elementary units have thus specific behaviors, which do not necessarily coincide with subfunctions of the wanted global function. This will emerge from the interactions among elementary units.

"…In the bottom-up approach, the system organizes itself to react in an adequate fashion in every case, from which its flexibility results. Furthermore, a breakdown in some unit does not impair the global working of the whole, thus confering its robustness to the system" (Ibid).

This description seems to fit even very complex social systems (i.e. systems including a very great number of elements capable of undergoing a very considerable number of interactions). As well as to societies of insects, it can probably be extended to societies of automata, to human societies and, possibly to societies of neurons in the brain.

As to models, for instance in systems dynamics, the bottom up procedure insists on starting with complete models of all the separate subsectors. D. MEADOWS et al state that these "should be carefully estimated, tested, elaborated and explored, so that one understands and has confidence in each part of the model before they are fitted together" (1982, p.274)

Still, in the case of the Systems Dynamics model, the process of fitting together, far from spontaneous, is decided by the modelizer and thus more or less depends on her/his subjective views.

The "bottom up" approach should not be exclusive of the "top-down" one. Both are in fact complementary, as expressed by Cl. PAHLWOSTL (1993):"…the dichotomy between top-down and bottom-up control converges to a mutual and inseparable dependence on both factors. Neither a purely reductionist approach nor a merely holistic perspective is sufficient to encompass the intrinsic nature of the systems behavior"