Risk assessment of emerging technologies using Bayesian networks

The benefits of emerging technologies are often recognized prior to the acknowledgement of potential barriers such as legal restrictions, regulatory scrutiny or liability conundrums. Consequently, supervisory bodies and insurance companies often lag behind in the identification, assessment, response and control of new risks conceived by these innovations. Lack of empirical data on the likelihood or consequence of adverse events inhibits traditional risk analysis techniques. This reactive process poses a considerable threat to the safety of end-users, to insurance companies’ sustainability and to the public’s trust in the regulatory institutions responsible for the oversight of new technologies. This thesis contributes a novel methodology posited to overcome these limitations. Bayesian networks (BNs) are proposed as an effective technique for the proactive risk assessment of emerging technologies. BNs are probabilistic models of causes and effects, graphically expressing causal relationships (i.e. conditional probabilities) between different variables (Fenton and Neil, 2012). The method is suited to evaluate the risk of emerging technologies in its ability to dynamically update its belief as new information becomes available. Furthermore, BNs facilitate the use of expert judgement to bridge the informational gap where evidence is sparse, inconsistent or missing. This thesis examines the efficacy of the BN methodology applied to the risk assessment paradigms of both nanomaterials and autonomous vehicles. Chapter 2 and Chapter 3 extends the state-of-the-art risk assessment approaches of nanomaterials (NMs). Large inconsistencies in characterisation data, toxicological measurements and exposure scenarios make it difficult to map and compare the risk associated with NMs based on physicochemical data, concentration and exposure route. Chapter 2 conducts a novel investigation into the influence of nanomaterial characterization, type, and exposure on the level of risk posed in an occupational setting using a BN model. This research contributes to the literature by being the first to map NM occupational risk probabilities derived from the BN model onto a control banding illustration. The third chapter is a comparative analysis of BNs ability and capacity to rank the hazard of different nanomaterials against more established methods. This comparative study investigates the efficacy of the quantitative weight of evidence and BNs in ranking the potential hazard of TiO2, Ag, and ZnO. This research finds that hazard ranking is consistent for both risk assessment approaches. Furthermore, this research adds to the academic literature by demonstrating that the BN exhibits more stability when the models are perturbed with new data. Chapters 4 and 5 contribute novel insights into the probabilistic reasoning and forecasting ability of BNs for the assessment of the mutable and emerging risks inherent to the state-of-the-art in connected and autonomous vehicle innovations. The first application demonstrates a BN statistical risk estimation approach that can accommodate changing risk levels and the emergence of new risk structures. This method is applied to a Level 3 conditionally-autonomous vehicle for two scenarios, one where the driver is in control and one where the vehicle is in control. This approach is evaluated from the perspective of the insurer and it is recommended that a greater degree of collaboration is required between insurance companies and car manufacturers in order to develop a greater understanding of the risks underlying semi-autonomous and fully autonomous vehicles. Chapter 5 specifies a unique proactive connected and autonomous vehicle (CAV) cyber-risk classification model. This method uses a BN model, premised on the variables and causal relationships derived from the known software vulnerabilities within the National Vulnerability Database (NVD), to represent the probabilistic structure and parameterisation of CAV cyber-risk. The resulting model demonstrates nearly 100% prediction accuracy of the quantitative cyber-risk score and qualitative cyber-risk level. The model is then applied to the use-case of GPS systems of a CAV with and without cryptographic authentication. This demonstrates how the model can be used to predict the effect of risk reduction measures on the overall cyber-risk level. The resulting applications contribute state-of-the-art risk assessment frameworks for nanomaterial occupational hazard, semi-autonomous vehicle accident risk, and cyber-risk for connected and autonomous vehicles. Each chapter represents a peer-reviewed published/accepted journal article with a mean 2016 journal impact factor of 3.12.