Self-organization has been shown to exist for geometrical structures; that is, complex global spatiotemporal structures have been produced in nontrivial ways by relatively simple constraints, deterministic or stochastic, at the level of the individual parts. The large number of interacting parts and some kind of underlying force field produce long-range correlations that allow the geometrical structures to emerge. However, producing instances of self-organization for functional features (such as intentional actions) is much more difficult, where meaningful functions, not explicitly programmed by the designer of the model, are emerging properties of such networks of interactions. Louzoun and Atlan have been able to simulate some features of an intentional action without resorting to the existence of an ethereal consciousness. In this case, “intentional” means that the system has a goal to reach, even though the goal may not necessarily be evident to the running program at the beginning.

For goals to emerge, it was found that at least four elements are required. First is an indirect dynamic link between the initial and final states. Second is memory, for in the absence of memory, the network would not be able to retrieve an initial state from a final state. Third is a learning algorithm from past events (memory is involved, and both memory and the learning algorithm are merged in this work). Fourth is an evolving set of goals allowing for both stability and newness. For more real-world applications, this may involve genetic algorithms, thus avoiding the traditional Cartesian controversy of free will. Louzoun and Atlan substitute some of its implications with an alternative monist theory of action, so this work simply allows (or does not forbid) the existence of free will.

I think this work is an important departing point for more exhaustive research into several cognitive areas related to intentional action. Maybe future, deeper research into the subject of free will is in order.