Incorporating complex adaptive systems concepts in ontology driven Bayesian network models : towards resolving wicked problems

Abstract

Wicked problems are complex ill defined problems very difficult to solve tractably using analytical methods and interventions. They include problems like pandemics, climate change effects, traffic jams, and financial market crashes. Attempts at solving such problems using analytical methods tend to produce counter-intuitive, unpredictable pathological outcomes. Wicked problems emerge, in part from the character of complex adaptive systems, and from stakeholder disagreements on their definition and resolution. We argue that baseline Bayesian models do not have adequate constructs to provide compact, and tractable modelling support for wicked problems. Applying an iterative and rigorous abductive design science research methodology, an ontology driven Bayesian modelling framework is applied to design the Granular Niche probabilistic Bayesian model, a formal, ontologically sound, and explainable artificial intelligence model, incorporating complex adaptive systems theory concepts: context; granularity; and perspective, as constructs. Using evaluation metrics from applicable kernel theories comparative evaluation of the model is carried against baseline Bayesian models. The results indicate that the novel model out-performs baseline Bayesian models against the following evaluation criteria: i) complex adaptive systems’ representation accuracy and precision; ii) structure learning; iii) parameter estimation; iv) knowledge discovery; and v) explicitly modelling and reconciling divergent multiple stakeholder perspectives of a given wicked problem.