Modeling the dynamics of innovation cascades introduces some fundamental and important questions, which have so far been largely glossed over in the social scientific literature. Agents in these cascades act intentionally, and as the case studies will certainly make clear, this intentionality assumes a fundamental role in how innovation dynamics play out. Yet what happens never simply follows particular agents’ intentions, in part because of ontological uncertainty. How can agent intentionality in the face of ontological uncertainty be represented in models, so that its effects can be assessed? Answering this question has many implications beyond the context of innovation and sustainability studies (not least in ICT intended to assist agent “decision-making”, which tends to start with situations structured precisely to rule out ontological uncertainty and the cognitive constraints it imposes!), but it is impossible to ignore in this context.
Nor can the problem of how to represent invention, in particular new attributions of functionality, be ignored: most existing models of innovation just rely on random search of some pre-defined space of possibilities, which completely misses the point from our constructivist perspective. What about the emergence of unintended consequences, and the processes through which agents identify them and engage in interactions to deal with them? Is it even possible to consider the social aspects of sustainability through modeling without some approach to this question? Finally, what role can modeling have in innovation and sustainability studies, given that the contexts in which these studies matter necessarily present agents with limited foresight horizons, and so the Newtonian ideal of “good models mean predictability” cannot possibly be achieved? Nor can the Wignerian precondition for “the unreasonable effectiveness of mathematics” in describing the physical world – that is, the existence of symmetries and invariances – be realized in innovation cascades that keep churning out novelties.
In INSITE we initiated a series of discussions on each of these issues, and we periodically confront and report on where we currently stand with respect to them, as well as to assemble examples of current “best-practice” in dealing with each of the modeling challenges described in this paragraph.
We are currently working with two classes of models (in accordance as well with MD research project Emergence by Design): the first is simple and abstract, which we will use for refining and making precise the ontology for innovation cascades, as well as carrying out experiments to understand, in the artificial world of the model, how, in what conditions, and to what extent emergent phenomena might be steered; while the second is more complicated agent-based models based upon the representations we generate for the relevant entities in our case studies, to test the adequacy of the theory as formalized in the model, by observing whether the model can generate stylized versions of the key emergent phenomena observed in those studies.
Obviously, the agent-based models we develop to simulate the dynamics observed in our case studies will be much more “realistic” than the other class of models. But their lack of realism of these models does not imply that they are not useful. First, translating the qualitative ontological theory into the simple but rigorously-defined world that such toy models represent can help locate ambiguities, omissions and unnecessarily complicated formulations of concepts and categories in the theory – perhaps more effectively than can be achieved with the necessarily more complicated “realistic” simulation models. Second, we are much more likely to discover “universal” phenomena associated with innovation processes in the simplest models that instantiate it (as in the work of Eigen and Schuster on hypercycles or Kauffman’s on the NK model), and it is through such phenomena that we can develop the simple stories that can help to “narrativize mass dynamics,” as described in the breakthrough discussion. Indeed, this is our main motivation for investing two years in constructing and experimenting such models.