Google executives claim to be unaware of how their AI works and how it is capable of learning languages

19 Apr. (Portaltic/EP) –

Google managers They have assured that their artificial intelligence (AI) is still in the process of experimentation and that they do not know how it is capable of learning languages, following an episode in which it knew how to respond to a language for which Bard was not trained.

The company presented at the beginning of February Bard, a ‘chatbot’ based on conversation technology Model for Dialogue Application (LaMDA) language with which he hoped to enhance the search experience to offer “current and high-quality answers.”

Google released this news as Microsoft presented the new version of its search engine, Bing, and its browser, Edge, driven by the supported by the ‘chatbot’ Chatgpt.

Due to the success of this launch, Google is completely redesigning the operation of its search engine in order to achieve a more personalized user experience. This change is now in the hands of a multidisciplinary team of the company, called Mag project.

Thus, Google is currently preparing what will be the new Google Search, with which it hopes to remain the most used engine on the Internet and in most ‘smartphones’ after learning that Samsung was considering replacing Google with Bing as the default system.

Now, company executives have given more details about Bard, an artificial intelligence that, “for several months he read almost everything on the internet and created a model of what language looks like,” according to senior vice president James Manyika.

This, together with the vice president of Google, Sissie Hsiao, have recognized that the first thing they learned was “that Bard does not look for answers on the internet as Google search does” and that the answers come from the AI ​​”of a self-contained program that was mostly self-taught.

Manyika, for her part, has assured that Bard is not “sensitive” or “self-aware”, but that they can exhibit human-like behaviors because they have learned from individuals “who have feelings, emotions, ideas, thoughts and perspectives”.

The CEO of Google, Sundar Pichai, for his part, has commented that he sees this AI “as an experiment in a limited way” and that it still presents certain problems that must be solved, such as the ability of these systems to learn “by themselves abilities that they were not expected to have“, as indicated in this interview, published by CBSN.

“We discovered that with very few indications in Bengali, the AI ​​can translate everything,” said Manyika, who insisted that Google is now making an effort to investigate how this has been possible and hopes to achieve that be done with “a thousand languages”.

While Pichai has referred to this as a ‘black box’, that is, events in AI that developers are unable to explain. “You don’t fully understand and you can’t tell why he says this or why he is wrong.”