Many of today's accident reports suffer from a number of limitations. In particular, it can be difficult to extract critical events from the mass of background detail in natural language accounts of human 'error' and system 'failure'. This, in turn, makes it difficult for readers to identify the evidence that supports the recommendations and findings in an accident report. Logic provides a means of avoiding these limitations. Formal proofs can be constructed to demonstrate that particular findings are actually consistent with the reported evidence. If such a proof cannot be derived then additional evidence must be found to support the analysis that is presented in an accident report. Unfortunately, conventional logics only provide limit support for this approach. In particular, they cannot be used to document the reasons why human 'error' jeopardises the safety of an application. This talk, therefore, demonstrates that epistemic logics can be used to formalise the implicit judgements that accident reports make about operator knowledge and motivation.
I will also discuss recent attempts to recruit computational models to support our analysis.
Finally, I will point out ways in which cognitive science might actively be recruited to support this approach and derive a pragmatic tool for accident analysis.