A few posts ago, I’ve already quoted Peter Fingar is arguing that the Cognitive Computing Era is Upon Us. In particular, I’ve argued that, at the end of the day, Softwarization of Telecommunications will be more and more oriented on implementing pervasive “Cognition”. See also my keynote at ONDM2015.
In fact, Softwarized Networks – just like a nervous system - will be able to sense and collect massive data (e.g., by pervasive sensors, smart terminals, things, machines, robots); these big sets of data will be exchanged and moved very quickly (transported by optical and mobile networks with high bandwidth and low latency); also they will be elaborated (with Cloud/Edge and Fog Computing) in order to make decisions and then to actuate local actions (by any pervasive actuators embedded into the reality around us).
In view of that capillarity and dynamism, Softwarized Networks will require a deep change of paradigm in the design and operations. In fact, such a continuum of virtualized resources is more similar to a Complex Adaptive System, as a real nervous system is. No centralized management, but a plethora of interacting processes determining emergent properties and characteristics.
I’ve already mentioned that in the future we’ll be using more and more cognitive methods, heuristics and algorithms, machine learning, knowledge representation-reasoning and, eventually, massively parallel computation to crunch, and make use of, the Big Data. But we could even go beyond that, by leveraging directly on Neuroscience analysis and models.
As a matter of fact, Neuroscience is arguing the nervous system networks represent the best low-cost structural basis for coexistence of informational processing (both segregation and integration) and transfer. Striking analogies between nervous system networks and future Softwarized Networks are stimulating the idea that Neuroscience theories, models and methods could provide valuable interdisciplinary elements for designing management and control of future networks.
As an example, I’m fascinated by the Global Workspace Theory (GWT): it argues that human cognition is implemented by a multitude of relatively simple, special purpose processes (processors). Interactions between them are based on cooperation-competition, thus allowing, at the end, coalitions of processes find their way into a Global Workspace (GW).
GW is a shared virtual space which acts to broadcast messages of certain coalitions to all processors, in order to recruit others to join. In summary, GW serves to integrate many competing and cooperating networks of processes. Global behavior will be driven by a myriad of local micro-behaviors rather than what’s happening in current networks, where a former built “knowledge representation” is used to manage the networks behavior. Practically the approach permits to rehearse global behaviors prior enacting local processes; said behaviors are evaluated, and the relative salience of a set of concurrently executable actions can be modulated as a result: those behaviors whose outcome is associated to a gain (or reward) become more and more salient and, at the end, selected and executed (e.g. with winner-take-all-strategy).
Examples like GWT could be easily implementable as distributed software frameworks (do you remember Linda and its evolutions ?) where processes could be seen as software service components or virtual network functions. A very exciting area of research and innovation...