By akademiotoelektronik, 19/01/2023

We don't have to seek to humanize a conversational robot

Decryption#DEEPTECH Reading Time: 02'42''1 November 10, 2021 Laurence Devillers, professor in artificial intelligence at Sorbonne University.The growing performance of conversational robots require being vigilant in the face of drift risks, explains Laurence Devillers, professor of artificial intelligence, and co-author of a report by the National Committee for Digital Ethics (CNPEN) on the ethical issues ofChatbots, or conversational robots, given Tuesday at the Aupremier Minister.Sharing a wandering recording pdf / export

Why worry ethical issues around conversational robots?For the moment, their performance is still often limited ...

Chatbots that used somewhat reduced, which seemed without imagination, are changing.They are able to have much more realistic dialogues that vontengigate the user more efficiently.Designers of conversational robots are now looking to create personalized, emotional systems, kinds of "virtual friends".

The risk is that users, especially vulnerable people- children, elderly ...- project human capacities on these machines, and in the long term see their influenced beliefs and opinions.Technology allows for example to build "Deadbot": machines that re -speak a deceased person, after having assimilated the mountains of data on them, their conversations, the content they have posted on social networks.What are the challenges of this digital posterity?What impacts on mourning management?

You highlight the new technical possibilities of these robots, which now generate new sentences.

New generations of chatbots are increasingly efficient thanks to the evolution of machine learning techniques (Machine Learning)."Transformers", these statistical treatment models of natural language based on billions of data such as the GPT-3 of Open AI (Elon Musk) or the LAMDA (Language Model for Dialog Applications) of Google, allow you to create machinescapable of engaging a free conversation on a potentially infinite number of themes.The machines become capable of producing an unprecedented word, which surprises humans, and to manifest a kind of language creativity.How do you do not say anything?Who is responsible for the words they say and their effects?

À lire aussiPeut-on penser l’homme augmenté sans diminuer le vivant ?

We are not for strong regulation, but we recommend pushing scientific research on the functioning of these models, and to engage in a long path of reflection and implementation of standards or good practices, with committeesaudit in particular.The technology is extremely positive if we master it, but we will need collective intelligence to understand how to coax it and how not to be guided by GAFAM ...

Can you give examples of the report recommendation?

We ask for example to reduce the projection of moral qualities on a conversational agent.When you design it, you don't have to try to humanize it so that the user projects human qualities on it.The machine must say that it is a machine, and that it communicates on its purpose.We are also interested in the question of biases which can be perpetuated by these conversational agents, and in particular the question of gender biases.

If we systematically give servile agents names and female voices, what does that mean representations of women in society?We must worry about the discreet power of these choices which are made for us ... The mission recommends that the choice of names and voices (male or feminine) of the robot be made in "fair", and that the user can "modify this choice ”as he wishes.

À lire aussiIntelligence artificielle : 2,2 milliards d’euros pour la formation et l’innovationArticle écrit par Maddyness avec AFPSharing a wandering recording pdf / exportNewslettersS'abonnerRelated articlesDécryptagePar Heloïse Pons28 janvier 2022Après son rachat, Leka rebondit et parvient à lancer son robot éducatifDécryptagePar Arthur Le Denn et Géraldine Russell28 janvier 2022Michel Combes : « Le moment est venu pour SoftBank d’investir en France »DécryptagePar Heloïse Pons28 janvier 2022Index égalité professionnelle : les nouvelles obligations des entreprisesERRORSignaler une erreurErreur factuelleErreur factuelleSyntaxeMerci,NewsletterPas de spam, croix de bois, croix de ferMerci,Vous êtes bien inscrit•e, surveillez votre boite mail pour les prochaines news !Soumettre une infoMerci,