early february, First Google, then Microsoft, announced a major overhaul of its search engine. Both tech giants are investing heavily in building or buying generative artificial intelligence tools that use large language models to understand and answer complex questions. Now they’re trying to integrate them into search, in the hope they’ll give users a richer and more accurate experience. Chinese search company Baidu has announced it will follow suit.
But the excitement about these new tools may be hiding a dirty secret. The race to build a high-performance, AI-powered search engine is likely to require a massive increase in computing power, and with it, a massive increase in the amount of energy and carbon emissions tech companies need.
“There are already vast resources involved in indexing and searching internet content, but the integration of AI requires a different type of firepower,” said Alan Woodward, a professor of cybersecurity at the University of Surrey in the United Kingdom. “It requires processing power as well as storage and efficient search. Every time we see a step change in online processing, we see a significant increase in the power and cooling resources required for large processing centers. I think this may be such a step. “
Training large language models (LLMs), such as the one that underpins OpenAI’s ChatGPT, which will power Microsoft’s enhanced Bing search engine and Google’s equivalent, Bard, means parsing and computing connections in vast amounts of data, which is why they tend to Developed by companies with significant resources.
“Training these models requires a lot of computing power,” says Carlos Gómez-Rodríguez, a computer scientist at the University of Coruña in Spain. “Currently, only big tech companies can train them.”
While neither OpenAI nor Google have said how much their products cost computationally, the researchers’ third-party analysis estimated that training GPT-3, based in part on ChatGPT, consumed 1,287 megawatt hours and resulted in more than 550 tons of CO2 equivalent — equivalent to The amount of carbon dioxide consumed by a person traveling between New York and San Francisco 550 times.
“It’s not that bad, but you have to take into account [the fact that] Not only do you have to train it, you have to execute it and serve it to millions of users,” Gómez-Rodríguez said.
Using ChatGPT (investment bank UBS estimates 13 million daily users) as a standalone product is also very different from integrating it into Bing, which handles 500 million searches per day.
Martin Bouchard, co-founder of Canadian data center company QScale, believes that adding generative AI to the process will require at least “at least a four- to five-fold increase in compute per search,” based on what he’s learned about Microsoft’s and Google’s search plans. He noted that ChatGPT will currently stop making sense of the world at the end of 2021 as part of an attempt to reduce its computational demands.
This will have to change in order to satisfy the demands of search engine users. “If they’re going to retrain the model every so often and add more parameters and stuff, that’s a whole different story,” he said.
