This repository contains the LaTeX source for my PhD thesis. The final PDF version is here.
This thesis studies problems involving unreliable information. We look at how to aggregate conflicting reports from multiple unreliable sources, how to assess the trustworthiness and expertise of sources, and investigate the extent to which the truth can be found with imperfect information. We take a formal approach, developing mathematical frameworks in which these problems can be formulated precisely and their properties studied. The results are of a conceptual and technical nature, which aim to elucidate interesting properties of the problem at the core abstract level.
In the first half we adopt the axiomatic approach of social choice theory. We formulate truth discovery -- the problem of aggregating reports to estimate true information and reliability of the sources -- as a social choice problem. We apply the axiomatic method to investigate desirable properties of such aggregation methods, and analyse a specific truth discovery method from the literature.
We go on to study ranking methods for bipartite tournaments. This setting can be applied to rank sources according to their accuracy on a number of topics, and is also of independent interest.
In the second half we take a logic-based perspective. We use modal logic to formalise the notion of expertise, and explore connections with knowledge and truthfulness of information. We use this as the foundation for a belief change problem, in which reports must be aggregated to form beliefs about the true state of the world and the expertise of the sources. We again take an axiomatic approach -- this time in the tradition of belief revision -- where several postulates are proposed as rationality criteria.
Finally, we address truth-tracking: the problem of finding the truth given non-expert reports. Adapting recent work combining logic with formal learning theory, we investigate the extent to which truth-tracking is possible, and how truth-tracking interacts with rationality.
The thesis should be compiled with XeLaTeX, using biber and biblatex for
references. For convenience a Makefile is provided: use make pdf and make bib to build.