Epistemic reasoning in AI
Abstract
In multi-agent systems, intelligent agents should be able to give explanation of their decisions. Indeed, in case of failure, agents need to justify their decisions in an understandable way, in particular to comply with recent laws (e.g. GDPR in Europe). Also, they need to make meaningful decisions to cooperate with other agents such as humans. To achieve this goal, the agents should model the humans' mental states. For instance, a robot may inform the human of the position of an object if it believes that the human needs that object and does not know its position. In this tutorial, we will present the latest advances in reasoning about knowledge/beliefs.
This tutorial aims at being accessible to a broad audience. It is illustrated by the pedagogical tool
Hintikka’s world. The tool depicts mental states by means of comic strips and features a lot of simple multi-agent systems such as games.
We will discuss several formal tools for modeling the following:
- knowledge obtained by seeing/perceiving the environment/other agents;
- temporal aspects of knowledge
- Complex epistemic actions (public/private announcements, etc.)
- Strategic reasoning under imperfect information
- Knowledge-based programs
Bio
Tristan Charrier has defended his PhD in December 2018
on epistemic reasoning. He contributed to the field in several aspects:
symbolic models, epistemic planning, languages for specifying epistemic
situations, demonstration of epistemic reasoning.
Tristan teaches formal logic including temporal logics and model checking, algorithmics, programming.
Tristan Charrier
tristan.charrier[at]irisa.fr
François Schwarzentruber is associate professor at ENS Rennes
(France). His current research interests are mainly focused on theory
and applications of logic to artificial intelligence, agency and
multi-agent systems and computer science. He has been a PC member of
some editions in that topics such as AAMAS and IJCAI. He was reviewer
for journals such as Synthese, Studia Logica, Theoretical Computer
Science. Since 2011, his research mainly focuses on studying dynamic
epistemic logic.
François Schwarzentruber
Bâtiment Alfred Sauvy
École normale supérieure de Rennes
Campus de Ker Lann
35170 Bruz
France
Tél : (+33) 2 99 05 93 23
Fax : (+33) 2 99 05 93 29
francois.schwarzentruber[at]ens-rennes.fr
Prerequisities
- Basic knowledge in logic
- The tutorial will be for a large public and we will use the pedagogical software called Hintikka’s world.
Tutorial materials
Slides: