Modelling of a cognitive agent: processes and logic.
20 juin, 13h-15h
Salle F0.13, Maison de la Recherche (attention, salle différente)
This talk will be two-fold. First I will introduce the problem of providing a logical and neural model for a cognitive agent, and ask what can be reasonable goals, and what requisites this entails on the logic to use. More precisely, I will distinguish between two kinds of reasoning processes, one I call deliberative reasoning, and the other automatic inferences. I will explain why it is reasonable to hope to provide a neural model for automatic inferences, whereas it is not for deliberative reasoning. Thus restricting myself to automatic inferences, I will suggest a plausible neural process for these, and ask what logic could model it. This will bring me to dismiss frst any logical modelling relying on possible worlds, and second, any logical modelling involving defeating
The second half of the talk will essentially take up again the content of my recently submitted paper. I will present what I think is a suitable logic for automatic inferences, namely the one based on partial worlds models (which I introduced already in my previous talk). Then I will tackle again the problem of providing a complete set of rules for inference relations induced by these models. In my previous talk, I proposed to do this by the means of a compatibility relation. This time I will analyse the question in terms of defnable sets of partial worlds, and propose instead to enrich the language with an additional connective. Within this renewed framework, I will provide two representation theorems: one for inference relations induced by admissible smooth models, and another for relations induced by admissible ranked models.