WP4: Confidence in Dependability Cases
The objective of the work package is to deliver a method for structuring confidence-based dependability cases, based on a probabilistic interpretation of confidence.
1. Confidence based dependability cases
We will define a rigorous approach to the propagation of confidence in cases where claims are structured on the basis of architecture, functionality and attributes. The initial semantic approach will be to use PVS to model the architecture and attribute decomposition.The propagation of confidence will be based on the novel analytic Bayesian framework we have developed recently, which allows manipulation of a complete representation of all uncertainty (in contrast to the more familiar numeric approach involving tools like Hugin). Drawing on a large corpus of empirical work available to us on how real cases are currently structured, we will define and evaluate structuring mechanisms that support confidence propagation. Initial work will focus on how to propagate confidence through the attribute and functionality types of claim expansion and decomposition.
2. Structure – timebands and diversity
This task will investigate how to apply two important structuring mechanisms: timebands and diversity in dependability cases. It will formalise the role of time in the structuring of cases, in association with WP1. There will be experimentation with York tools and a configuration of a commercial dependability case tool. The case studies and safety case corpus will be used to evaluate the structuring approach. Semantic models and analysis techniques will be developed to support the use of diverse multi-legged arguments in dependability cases. As in T4.1, this will build on our recent work on analytic Bayesian inference: an important challenge will be to address the issue of scaling in these complex algebraic representations of uncertainty and confidence propagation. Success will lead to procedures and guidelines for the construction of more effective cases exploiting diversity: an important issue for regulators.
3. Trust and communication
Building on work on trust models from WP3, we will investigate issues of trust in argumentation, in particular differences between a notion of “good” (trustworthy) dependability cases and “convincing” (trusted) ones. We will identify the sociological and psychological issues involved in communicating sound cases convincingly to different individuals and groups (who may be quite disparate, e.g. vendor, regulator, utility, manager, public) and will investigate how this may be applied in dependability case presentation. The PIs and Is will work closely with the regulators and utilities involved with INDEED. Involvement of multi-site PIs and Is will ensure that DIRC’s interdisciplinary focus is exploited within this task. It will be supported with a (part time) senior RA with a multi-disciplinary psychological background.