Room 206 (2nd floor, badged access)
4 mai 2023 - 14h00
Artificial Intelligence and autonomous systems
par Joseph Sifakis de VERIMAG
Abstract: We discuss the adequacy of the criteria for assessing the intelligent behavior of agents. We argue that the Turing test fails to capture the many facets of human intelligence and propose a replacement test that relativizes the concept of intelligence based on the ability of an agent to replace successfully another agent in performing a task.
Autonomous systems incorporate the replacement test because they are intended to replace humans in complex organizations. We review the key features of autonomous systems and the challenges in developing them. We argue that realizing the vision of autonomy raises very difficult systems engineering issues not related to the fact that agents are intelligent.
A comparison between autonomous agents and human agents shows important differences and complementarity in their ability to develop and apply knowledge. We argue that empirical validation through testing is the only viable approach for such a comparison and discuss the possibility of extending current validation techniques to intelligent systems.
We conclude by showing that the proposed intelligence test offers a space of possible intelligences that deserves to be explored further, in particular by deepening the different aspects of human symbolic intelligence.
Séminaire décalé à cause du congrès de la SIF à Grenoble, les 6 et 7 avril 2023:
https://congres.societe-informatique-de-france.fr/
