Facing Openness with Socio Cognitive Trust and Categories
Matteo Venanzi, Michele Piunti, Rino Falcone, Cristiano Castelfranchi
A crucial issue for agents interacting in open MAS is the ability to filter out information in order to assess a subjective evaluation of trust for possibly unknown counterparts. While typical solutions scan information sources by relying on the circulation of reputational images, or on statistical analysis of past interactions, this work presents an alternative approach based on the cognitive ability to: ({1}) estimate trustworthiness towards unknown agents based on reasoning over abstract classes or categories, in order to delegate tasks and sub-tasks; ({2}) learn a series of emergent relationships between observable properties and the effective agent abilities to fulfill tasks in situated conditions. In this view, categorization is provided to recognize those explicitly readable signals (Manifesta) through which it is possible to infer hidden properties and capabilities (Kripta), which finally determine agents' behavior and performances in concrete work environments. Learning is provided in order to refine reasoning attitudes needed to ascribe tasks to categories of agents. On such a basis, a series of computational architectures are evaluated allowing agents to dynamically assess trust combining categorization abilities, personal experiences and context awareness. Experiment results are finally described in order to compare and evaluate the proposed solutions.