With the new Bing from Microsoft, the integration of artificial intelligence like ChatGPT into our search engines is gaining popularity. Google doesn’t offer this service yet, and for good reason: According to one of its leaders, it would cost 10 times more than today.
Conversational artificial intelligences are on the rise. Following the success of OpenAI’s ChatGPT, launched in November 2022, the American giant Microsoft has developed a version of its Bing search engine that is enhanced by this automated chat service.
Therefore, all eyes are on Google, the search engine giant with an 85% market share. But that shouldn’t be happening anytime soon, according to the president of its parent company, Alphabet. In an interview granted on Thursday, February 23 to the press agency ReutersJohn Hennessy estimates that an AI-powered Google search would cost ten times more than the current system.
A totally different method
As the manager told the British press agency, what is in doubt is the current operating model of AI such as ChatGPT. Today, Google search works through an indexing system: Google bots scan the web in an upward direction, store information according to the keywords they correspond to, and then return the most relevant ones according to the search. user demand. Repetition, definitely.
However, AIs like ChatGPT are algorithms that are trained to produce responses similar to those of a human. Therefore, your answer requires generating new responses each time. This is what experts call “inference,” he explains. Reuters.
The more words, the more expensive the search
An additional process that, therefore, would multiply by ten the cost of a single search, according to the president of Alphabet. In addition, the potential revenue from advertising integrated into this service would not be enough, he says. Reuters. The news agency reports a calculation made by analysts at the US bank Morgan Stanley: if half of the 3.3 billion searches sent to Google’s search engine last year were converted to search with AI, it would cost the company more than 6,000 million dollars by 2024.
And again, this applies if the responses generated by the AI are limited to 50 words. Because the unit cost of an AI search increases the longer the generated responses are. This same calculation reaches 9,000 million dollars for 75 words per response and up to 12,000 million dollars for 100 words.
A difficult bill to estimate
These huge costs of AI research are mainly linked to its mode of development: the creation of gigantic databases provided by thousands of graphics processors. In early February, two analysts questioned by the magazine Forbes he estimated that the cost of buying the GPUs needed to use AI in all of Google’s searches could even be as high as $100 billion.
However, according to Reuters, another analyst firm estimates that this cost could be limited to 3 billion dollars since the multinational is already developing its own chips dedicated to artificial intelligence calculations. But the energy cost of such a service is still difficult to reduce or ignore, especially if we add the ecological cost of such intensive consumption.
It’s probably too soon to make precise estimates of the exact cost of Google’s move to artificial intelligence. But as pointed out ReutersIt is probably these imprecise but frightening calculations that are currently preventing the tech giants from taking the plunge.
Do you use Google News (News in France)? You can follow your favorite media. Continue Frandroid on Google News (and Numerama).