SG1: Defeasibility of agent facets


The topic of SG1 is defeasible: Most agent facets, including goals, beliefs and norms, are defeasible, that is, it can be modified or revised in the presence of more important, contradictory information. They are thus defaults rather than firm knowledge. Existing formalisms to deal with revision include: non-monotonic logics, belief revision, some conditional logics based on revision, logics of preference, logics of utility, etc. The links existing between these techniques and their scope of applicability is currently unclear: one goal of SG1 is to clarify their links, and the applicability of each technique to each of the agents' facets (e.g., is the revision of belief tractable with the same techniques than the revision of norms, as studied in deontic logic?) While these logics generally consider a single agent for simplicity, our group aims at understand how group of agents can be considered as having collectively revisable goals, beliefs, and norms.

Revision is also important for the description of agents' dynamics: the state of an agent is usually modified by incoming information, but only in a parsimonious way. Conversely, the temporal dimension affects some types of revisions.


  1. Write up a state of the art of non-monotonic logics, and their applicability to agent facets: norms, beliefs, goals;
  2. Set up a list of examples/case studies revealing potential problems in such combinations; specially, can all facets be treated by a similar technical mechanism or are they behaving intrinsically differently?
  3. Propose a specific combination as the main working focus of the group. The chosen basis is a multi-modal logic based on dynamic logic, with defeasibility modelled by preferential logics.