The experiment was short-lived for the founders of the conversational robot called Lucie. Launched this Thursday, January 23 by the Lingora company, supported by the State, the AI was disconnected three days later, targeted by numerous mockeries on social networks. The cause: numerous calculation errors or absurd answers.
“Cow's eggs”, “weight of a hole of Gruyère cheese”…There were many people testing the new tool called Lucie on Thursday, January 23. Just as many people showing the shortcomings that AI still has on social networks. Alerted by the mockery, the project’s promoters decided after three days to close its access to everyone online, Michel-Marie Maudet, the CEO of Linagora, the company behind Lucie, explained to AFP on Sunday.
“A research project” not yet “ready for use”
Designed to offer a French alternative, based on open sources, to the artificial intelligence (AI) of the tech giants, Lucie had to be unplugged after delivering a cascade of absurd results to Internet users who questioned her. “We made the mistake of making (Lucie) available as is”, without specifying sufficiently that it was not a ready-to-use robot, but a “research project”, he acknowledged.
On the other hand, “all artificial intelligence systems have safeguards”, which prevent them from producing hateful or insulting texts, he stressed. “Our second mistake was to make Lucie available without these safeguards”.
A hasty exit ?
The manager explains that he wanted to put Lucie online before the international summit in Paris on AI on February 10 and 11 and that he had “not anticipated this excitement at all”, because Linagora “works in free software where communities are generally kind and encouraging”.
This initial failure does not prevent Lucie's promoters from hoping to put Lucie back online soon, after having made it evolve, to offer everyone “a language model of general interest”.
A project without “ambition economic”
The platform, winner of the France 2030 program launched by the State, aims to provide a “transparent and reliable” alternative to AIs like ChatGPT, and could for example be used for education or research.
Unlike tech giants like ChatGPT, Lucie, which “has no economic ambition”, promises transparency on the data used to train it, explained Mr. Maudet. This will make it possible, for example, to ensure that the answers come from verified scientific data. The model is intended to be used for applications in the world of education, but “to date, no specific work has been carried out with the National Education to personalize or adapt the model for educational use”, its promoters have emphasized.