Inspired by the notion of a “Darwin machine,” Wilson et al. aim to reconcile diametrically distinct evolutionary processes, such as innate versus adaptive and domain-general versus task specific, in a move toward a science of behavioral and cultural change. We applaud this step, though we think that the authors' rapprochement between Darwin machines and “multi-agent cooperative systems” requires some elaboration. What seems to be missing are the concepts, methods, and tools of self-organizing dynamical systems tailored specifically to the coordinated activities of living things – how they move, adapt, learn, develop, and so on (Beek et al. Reference Beek, Peper and Stegman1995; Calvin & Jirsa Reference Calvin, Jirsa, Huys and Jirsa2010; Haken et al. Reference Haken, Kelso and Bunz1985; Kelso Reference Kelso1995; Kelso & Haken Reference Kelso, Haken, Murphy and O'Neill1995; Schöner & Kelso Reference Schöner and Kelso1988; Turvey & Carello Reference Turvey and Carello2012; Warren Reference Warren2006; Zanone & Kelso Reference Zanone and Kelso1992). Among others, Coordination Dynamics (CD) has long been inspired by the works of Howard Pattee, who understood the significance of biological coordination, particularly the complementary nature of symbolic and dynamic descriptions (Kelso & Engstrøm, Reference Kelso and Engstrøm2006; Pattee & Raczaszek-Leonardi Reference Pattee and Raczaszek-Leonardi2012).

Instead of opposing genetically fixed and adaptive processes, Coordination Dynamics sees them as dual processes evolving on different time scales. Apparently “fixed” processes are not immutable; they are stable or slowly evolving. In complex systems, processes evolving on slower time scales have been shown to constrain faster ones (Haken Reference Haken1983). This opens the possibility to inquire under which conditions fast-evolving processes escape such slowly evolving (viz. inherited) constraints and reorganize the entire behavioral repertoire. Evidence shows that by scrutinizing how behavioral stability is lost or increases, it is possible to address the fundamental nature of change on several time scales: behavior (Schöner & Kelso Reference Schöner and Kelso1988), development (Sporns & Edelman Reference Sporns and Edelman1993; Thelen & Smith Reference Thelen and Smith1994; Thelen et al. Reference Thelen, Kelso and Fogel1987), learning (Zanone & Kelso Reference Zanone and Kelso1992), and adaptation (Warren Reference Warren2006).

Intentional change acts not on a blank slate but on an initial behavioral repertoire that favors or alters intentional action (Johnston Reference Johnston1981). A basic aspect of self-organizing CD is that elements of any nature tend to coordinate spontaneously when information is exchanged (usually bidirectionally) with the environment, creating “for free” intrinsic coordination tendencies. Such tendencies are not fixed but constitute a dynamic potential – the initial behavioral repertoire – that inhabits the same information space as intention (Kelso Reference Kelso1995; Reference Kelso2002). This is a reciprocal interaction: The repertoire shapes, while being shaped by, intentional forcing. Evidence on both behavioral and brain levels shows that an intentional change of behavior is determined by the relative stability of pre-existing patterns in the repertoire (DeLuca et al. Reference DeLuca, Jantzen, Comani, Bertollo and Kelso2010). Such tendencies specify the nature of change-driving parameters, the competitive or cooperative mechanisms involved, the gradual or abrupt pathways of behavioral evolution, and the transfer of acquired changes (Kostrubiec et al. Reference Kostrubiec, Zanone, Fuchs and Kelso2012; Zanone & Kelso Reference Zanone and Kelso1992; Reference Zanone and Kelso1997). In line with a science of intentional change, CD suggests that the paths and outcomes of behavioral evolution are predictable. Prediction is possible, however, on the condition that the relevant variables capturing coordination tendencies are identified and the initial behavioral repertoire assessed before applying change-driving factors.

If, like the authors, CD rejects the blank-slate tradition, the question arises of how the initial behavioral repertoire prior to learning specifies criteria for selection mechanisms to operate. Although useful, Skinner's (Reference Skinner1984) “selection by consequences” leaves open this issue (Timberlake Reference Timberlake and Dewsbury1990).

In our work on learning dynamics (e.g., Kostrubiec et al. Reference Kostrubiec, Zanone, Fuchs and Kelso2012), careful measurement of the initial repertoire allowed the discovery of “selection via instability” and “selection via matching” principles (Kelso Reference Kelso, Bergman, Cairns, Nilsson and Nystedt2000). The former predicts that when the initial repertoire contains only a few stable patterns, such that they and environmental requirements are far apart in pattern space, competition arises, leading to instability and sudden, abrupt phase transitions, at which a wide range of unstable, transient patterns are generated. Once one of them succeeds in dominating the system, it tends to persist as a new stable pattern in the repertoire. Conversely, when the required and initial patterns are close to each other in pattern space cooperation between them ensues. This entails a smooth, gradual shift of one of the initial patterns in the repertoire that matches the environmentally required behavior but does not add to the number of patterns in the repertoire itself. Hence, the contingent mechanism for change is “selection via matching.”

Empirical data indicate that selection via instability leads to long-term persistent behavior, whereas the outcomes of selection via matching are rapidly forgotten (Kostrubiec et al. Reference Kostrubiec, Tallet and Zanone2006).

In the last analysis, cooperation and competition dictate the path of change (viz. smooth or abrupt) and Darwinian-like manifestations (viz. generation, selection, retention) appear as observable a posteriori outcomes. A key aspect the authors may consider is that the initial behavioral repertoire and its essentially nonlinear dynamics influence not only how new behaviors are formed but also their sustainability.

Viewing selection as a competitive process, the authors argue that an account of coordination between distinct processes around a given task is required. CD may help in clarifying what cooperation means. Evidence suggests that due to the tremendous degeneracy of living systems, where the same outcome may be produced by different combinations of elements and via different pathways, the significant functional units are context-dependent coordinative structures (Kelso Reference Kelso and Meyers2009). Coordinative structures are softly assembled; all the parts are weakly interacting. Perturbing one part may produce a remote effect somewhere else without disrupting – indeed preserving – integrity of function. Coordinative structures are collective states whose spatiotemporal dynamics prove to be quite rich, including interesting transient regimes that are neither fully ordered nor disordered in space and time. For present purposes, a system may be termed “cooperative” if it is open to information exchange and hosts numerous context-dependent elements whose nonlinear interaction leads them to coordinate. Viewed in this light, a Darwin machine is a cooperative system in which the slowly and rapidly adapting parts remain separated while transiently interacting. Such dual coexisting tendencies would mean that the coordination dynamics of a Darwin machine are metastable, thereby providing a number of evolutionary advantages (Kelso Reference Kelso2012).