This universal AI is capable of understanding a thousand languages

By Courtesy of Google Google Googleplex Mountain View

Google is currently working on a new intelligent language model, capable of understanding the world's 1000 most-spoken languages. This will make it a universal model of artificial intelligence, accessible to the greatest number of people.

Google is currently working on a new intelligent language model, capable of understanding the world's 1000 most-spoken languages. This will make it a universal model of artificial intelligence, accessible to the greatest number of people.

The much talked-about ChatGPT has impressed professionals and ordinary folk alike, and has opened up a world of amazing possibilities for this form of artificial intelligence. To compete with such technology, all the most important actors of the tech industry are jumping on the AI bandwagon and launching a dizzying array of new projects. Google recently unveiled a preview of Bard, its own conversational robot project. With this AI, Google aims to optimize how information gets sorted and organized and, above all, to make information more accessible and useful to everyone. Bard, of course, will draw on information available on the web in order to provide up-to-date and contextualized answers.

But the Mountain View-based company has no intention of stopping there and has another project in the pipeline, an artificial intelligence model compatible with no less than 1000 languages. Google has already announced that it has trained its language model on 12 million hours of speech and some 28 billion sentences covering more than 300 languages, the goal being to soon reach 1000. It can already automatically recognize, in audio form, common languages such as English or Mandarin, but also rarer ones such as Azerbaijani. The learning process of the model developed by Google could in theory be adapted to any other language.

In addition to being a translation tool, this artificial intelligence has the potential to open up a whole range of lofty new prospects. It could even, in the long run, be integrated into connected objects, such as headsets or glasses. The first glimpse of its possibilities could be revealed at the next Google I/O conference, scheduled for May 2023, which is likely to focus on AI in general.

© Agence France-Presse