Can Human Evidence Accumulation Be Modeled Using the Set-Theoretic Nature of Dempster-Shafer Theory?

Belief updating and the integrating of streams of evidence are critical tasks for artificially intelligent systems. Traditional AI systems tend to employ Bayesian models of conditioning or approaches based on Dempster-Shafer. These approaches have assumptions that run counter to the cognitive models developed for human belief updating. In addition, human cognition has an added limitation of finite working memory, which requires the human cognitive system to develop and manage representations of data to manage capacity. While the evidence updating approach of Dempster-Shafer has issues with behaving as humans, the approaches use of the concept of the power set and the management of the beliefs through subsets of varying cardinality as evidence unfolds can serve as a promising model for human management of working memory. This paper uses a Clue® game approach to testing human subjects. The results show that the use of sets of information at varying cardinalities may be an effective model for describing how human subjects develop and manage beliefs.