Reasoning About Priming Effects
Albert Emil Mølgaard Thayssen

Abstract:
This thesis develops a formal-logical framework for representing and reasoning about content priming effects and their influence on belief-formation, where the relevant contents are properties or concepts. Content priming effects are cases in which exposure to a stimulus increases the accessibility of contents stored in memory, thus influencing a response to a subsequent stimulus or task. This response can manifest as observable behavior that can provide new information about the acting agent to any observing agent. In turn, this newly acquired information can lead to the formation of new beliefs. In developing our framework, we start by explaining the conceptual spaces framework, which we will use to (i) represent properties and concepts, and (ii) specify a measure of similarity between such contents. We then embed (i) and (ii) into a model of how content-processing can increase content accessibility, which constitutes the first original contribution of this thesis. In doing so, we first develop a general means for representing processing-induced accessibility increases. Then, we present a similarity-based account, which is used throughout the rest of the thesis. The second original contribution of this thesis is a formal model of how content accessibility can influence behavior. In this model, the key notions are those of selection score and executability condition. Selection scores measure how strongly an action is selected for based on content accessibility, while executability conditions tell us which action is selected based on this measure. We then use this model of action selection, together with our similarity-based account of processing-induced accessibility increases, to analyze two famous examples of content priming effects popularized by Daniel Kahneman. Finally, we cast the newly developed machinery in a formal-logical framework. This is done in two steps. First, we introduce a logic that can capture the effects of content processing on content accessibility, and express selection scores and executability conditions. Then, we revisit one of the earlier analyzed scenarios and show that our logic delivers the correct results. Secondly, we extend this logic to capture the beliefs of agents, as well as how agents update their beliefs with new information gained from observed behavior, where the behavior in question is caused by underlying degrees of content accessibility. This extension allows us to formally capture how content priming can affect belief-formation, which is the third original contribution of this thesis. Lastly, we put the machinery to use, presenting and analyzing a multi-agent scenario that show how content priming can induce a false belief.