The generation by an AI of texts or images can indeed revolutionize research on the Internet, with answers in natural language and no longer a list of documents, with the key to increased advertising revenues.

AI can also boost other aspects of an engine, such as mapping, meeting tools, forecasting techniques or image analysis.

Everything is accelerating in this new information battlefield.

In November, the Californian start-up OpenAI, supported in particular by Microsoft, launched its conversational robot ChatGPT, capable of answering any question - more or less precisely.

Offered free to the public, its success is phenomenal: 100 million users in two months.

At the end of January, Microsoft announced that it wanted to invest "several billion dollars" in OpenAI.

According to the American press, the group has already invested 3 billion there and could inject 10 billion more.

Hence its speed in deploying this technology in its products: Microsoft launched a more expensive version of its Teams communication software on Monday, with ChatGPT features, for example to generate meeting summaries.

And announced two weeks ago that it wanted to "add a layer of intelligence" from ChatGPT to all its other products, including its Bing search engine, so far crushed by Google.

It didn't take Google a month to retaliate.

Sundar Pichai, CEO of Alphabet, parent company of Google, in New Delhi, December 19, 2022 © Sajjad HUSSAIN / AFP/Archives

"Bard's ambition is to combine the breadth of knowledge of the world with the power, intelligence and creativity of our great language models. It draws on information from the web to provide up-to-date, high-quality answers" , Google said on Monday.

If it accelerated, it was because with AI, Microsoft risked, for the first time in 10 years, regaining control in internet research, a lucrative area where Google occupies more than 90% of the market.

The latter, seeing the success of ChatGPT, according to the American press, immediately lit a "red code", aware that this type of question-and-answer robot could kill research as we have known it for twenty years.

"No more need for motors"

"A search engine integrating AI will give a structured answer to a question asked. Not documents, as currently, but answers", explains to AFP Thierry Poibeau, director of research at the CNRS.

"With the risk that the Internet user is satisfied with it, despite the possible biases or control of this single response", warns Claude de Loupy, director of Syllabs, a French company specializing in automatic text generation.

"We will no longer need the engines. Better than finding a source, the AI ​​will be able to find a detailed answer. Like the illusion of an omniscient being who has read everything. But it's terrifying, because this answer is controllable. And it will be controlled”, warns the boss of Syllabs.

And for a site, "the only way to come out on top will be paid ads, which will therefore be even more numerous," he analyzes.

But conversational AIs like ChatGPT "also give wrong answers, which is embarrassing for a search engine. All the actors are therefore working on a tool that would mention the sources", specifies Thierry Poibeau.

The Meta group has also embarked on the AI ​​race © Lionel BONAVENTURE / AFP/Archives

Meta (Facebook), also very advanced in the race for AI, has also suffered setbacks.

Just before the release of ChatGPT, on November 15, the group released Galactica, a language model supposed to summarize scientific articles, or even help in their writing.

It had been trained on millions of articles, manuals, encyclopedias and courses, on the same principles as ChatGPT.

Except that Galactica generated too much nonsense and could write racist comments if asked about it.

Meta, who left first, had to withdraw its demo three days later.

© 2023 AFP