English version below

Des machines à parler : conversation et création avec l'Intelligence artificielle

Le GPT-3, un modèle de langage artificiel de pointe, a largement alimenté le débat scientifique concernant ses promesses et ses limites. Alors que l’on parle déjà de GPT-4, de nombreuses questions demeurent concernant non seulement les aspects techniques mais aussi les applications possibles des “machines parlantes” dans leur troublante capacité d’imiter et de simuler le langage humain.

Cette journée d’étude vise à faire dialoguer les spécialistes sur trois types de questions que soulèvent ces langages artificiels :

1. les principes, perspectives et limites de l’intelligence artificielle appliquée au langage ;

2. le problème des conditions économiques et matérielles de leurs conception et création, dont la place de la recherche publique dans le développement, très coûteux, de modèles de langage tel le GPT-3;

3.  les usages, notamment dans la conversation et la création, ainsi que dans les humanités numériques.

Plus prometteurs sont ces modèles, mieux les agents conversationnels simulent l’interaction dialogique avec l’humain, plus graves sont les questions qu’ils posent concernant les dynamiques sociales, les réactions émotives et affectives et les statuts identitaires attachés à ces intelligences artificielles.

Talking machines: conversation and creation with Artificial intelligence

GPT-3, a cutting-edge artificial language model, has largely fueled the scientific debate about its promises and limitations. However, many questions remain open concerning not only the technical aspects but also the possible applications of "talking machines" in their uncanny ability to imitate and simulate human language.

This workshop aims to bring together experts to discuss three types of questions raised by these artificial languages:

(1) the principles, perspectives and limits of artificial intelligence applied to language;

(2) the problem of the economic and material conditions of their conception and creation, including the role of public research in the development of very costly language models such as GPT-3;

(3) the uses of AI in conversation and creation, as well as in digital humanities.

The more promising these models are and the better the conversational agents simulate dialogical interaction with humans, the more serious are the questions they raise concerning social dynamics, emotional and affective reactions, and the identity status attached to these artificial intelligences.

Speakers :

Justine Cassel, professor and former Associate Dean, School of Computer Science, Carnegie Mellon University (2010-). Since January 2021 a member of CNNUM (Conseil National du Numérique)

Emmanuel Dupoux, director of studies at EHESS, member of the Laboratory of Cognitive Sciences and Psycholinguistics (LSCP)

Jacqueline Feldman, writer and journalist, worked as a conversation designer for multiple US firms, scripting chatbot dialogue.

Aldo Gangemi, full professor at University of Bologna, and director of the Institute for Cognitive Sciences and Technologies of the Italian National Research Council

Raphaël Millière, adjunct professor and Robert A. Burt Presidential Scholar in Society and Neuroscience at Columbia University

Valentina Presutti, researcher at the Semantic Technology Laboratory of the National Research Council (CNR) in Rome

---

This symposium is organized by TESaCo a project from the Academy of moral and political sciences (France), ALMA IA-Research Institute for Human Centered Artificial Intelligence (Bologne) and ISTC-CNR (Rome).