Google chief scientist Jeff Dean: AI needs 'algorithmic breakthroughs', AI not responsible for most of data center emissions growth

AI News


Google added fuel to the climate change debate this month when it said in its annual environmental report that emissions from its data centers will increase by 13% in 2023, citing the “AI transition.” But Google's chief scientist, Jeff Dean, said the report doesn't tell the whole story and unfairly places too much of the blame on AI.

Dean, the chief scientist at Google DeepMind and Google Research, said Google won't back down from its commitment to run on 100% clean energy by the end of 2030. But he said progress “will not necessarily be linear,” because some of Google's efforts with clean energy providers won't be operational until several years later.

“These efforts will significantly increase the share of carbon-free energy, but we also want to focus on making the system as efficient as possible,” Dean said in an onstage interview with Fortune's AI editor Jeremy Kahn at Fortune's Brainstorm Tech conference on Tuesday.

Dean also made the important point that AI is not causing as much growth in data center usage, and therefore carbon emissions, as critics say.

“There's been a lot of attention on the growth in AI energy usage, and from a very small base, that usage is definitely growing,” Dean says, “but people tend to confuse that with overall data center usage. AI is a small part of it right now, but it's growing rapidly, and then they attribute the growth rate of AI-based computing to overall data center usage.”

Dean said it was important to look at “all the data” and “the true underlying trends,” but did not elaborate on what those trends might be.

One of Google's earliest employees, Dean joined the company in 1999 and is credited with being one of the key people who transformed an early internet search engine into a powerful system that indexes the internet and reliably serves billions of users. Dean co-founded the Google Brain project in 2011 and spearheaded the company's efforts to become a leader in AI. Last year, Alphabet merged Google Brain with DeepMind, the AI ​​company Google acquired in 2014, and named Dean chief scientist reporting to CEO Sundar Pichai.

Dean said combining the two teams will allow the company to “build a foundation of better ideas” and “pool computing power to focus on training one large effort like Gemini, rather than multiple fragmented efforts.”

Algorithmic breakthrough needed

Dean also answered questions about the status of Google's Project Astra, a research project that DeepMind head Demis Hassabis announced in May at Google I/O, the company's annual developer conference. Hassabis described Astra as a “universal AI agent” that can understand the context of the user's environment, and in a video demo, he showed how a user can point their phone's camera at nearby objects and ask the AI ​​agent appropriate questions, such as “What neighborhood am I in?” or “Did you see where I put my glasses?”

At the time, the company said Astra's technology would be coming to the Gemini app later this year, but Dean was more modest, saying, “We hope to have it available in some form to test users by the end of the year.”

“The ability to combine the Gemini model with a model that can actually have agency and perceive the world around you in a multimodal way is going to be incredibly powerful,” Dean said. “We're obviously approaching this responsibly, so we want to make sure the technology is ready and there aren't any unintended consequences, so we're going to roll it out to a small number of early test users first.”

When it comes to the continued evolution of AI models, Dean noted that additional data and computing power alone won't be enough: A couple more generations of scaling could make significant progress, but ultimately “further breakthroughs in the algorithms” will be needed, he said.

Dean said his team has long been focused on how to combine scaling and algorithmic approaches to improve factuality and inference capabilities “so that models can imagine plausible outputs and then reason about which ones make the most sense.”

Dean said such advances will be important “to make these models even more robust and reliable than they are now.”

Read more articles from Brainstorm Tech 2024:

Wiz CEO says “consolidation in the security market is really needed” as reports of $23 billion Google acquisition swirl

Grindr's CEO says 'synthetic employees' will create a fierce talent war for tech startups

Experts worry that the US-China cold war could turn hot: 'Everyone is waiting for something to happen in Asia'

Recommended Newsletters:

CEO Daily provides essential context for the news business leaders need to know. More than 125,000 readers trust CEO Daily every weekday morning for insight on and from the C-level. Subscribe now.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *