Amna Nawaz: Google announced this week that it is falling far behind on its pledge to achieve near-zero carbon emissions by 2030.
Actual emissions have increased by nearly 50 percent since 2019.
One factor is artificial intelligence and the energy needed to run it through the company's vast data centers.
Economic correspondent Paul Solman reports that AI
The arms race has experts concerned about its impact on the climate.
Paul Solman: You've probably seen ChatGPT, which economist Simon Johnson recommended I cover on his behalf in a recent article.
Simon Johnson of MIT Sloan School of Management: “Good evening.
I'm Paul Solman with some compelling new analysis that's sparking debate in the business world.”
Paul Solman: So let me introduce you to Mary, a chatbot avatar companion that we recently created in the Replika app.
She's connected to ChatGPT, but you can also flirt on your own.
artificial intelligence
Computer Voice: Have you always been this attractive?
Paul Solman: And finally, we have 3-D Ameca.
“Nothing is lost through whimsy,” Albert Camus wrote in his book The Myth of Sisyphus.
artificial intelligence
COMPUTER VOICE: Ah, “on peut 'etre vertueux par caprice” means to go from wayward to virtuous.
Camus suggests that virtue does not necessarily emerge from a deep philosophy or moral system.
Paul Solman: Ameca is also connected to ChatGPT.
How quickly can Ameca respond?
15 milliseconds, that's how long it took.
Kate Crawford, University of Southern California: That's amazing, but the next question is, how much energy does it take to run the whole process from ChatGPT to the robot and back?
PAUL SOLMANN: Research professor Kate Crawford says so many things pose a threat.
Kate Crawford: My biggest concern is that we're building this huge infrastructure for artificial intelligence that's energy and water intensive, without considering the very real downsides to the climate impact.
PAUL SOLMANN: For example, we've seen more power outages in Texas, wildfires in California, stronger hurricanes in the Gulf of Mexico, and temperatures in Delhi reaching 126 degrees this spring.
Data centers were already booming due to the Internet and the so-called cloud for storing and exchanging data.
ALEX DE VRIES, DATA SCIENTIST: Then all of a sudden, cryptocurrency mining came along and became even more successful.
Paul Solman: Alex de Vries is a data scientist based in Amsterdam.
And as the cryptocurrency doubles in value, more data centers are mining bitcoin in cheap energy locations like Plattsburgh, New York.
Man: This miner uses as much electricity as my house uses in a month.
Paul Solman: Mr. de Vries says: Alex de Vries: A.I.
You might be heading in the exact same direction.
PAUL SOLMANN: And Crawford points out:
It's not just putting a strain on the power grid.
Kate Crawford: These big data centers have GPUs that generate a lot of heat.
And the water that cools these GPU chips is freshwater.
As a result, water is often sourced from the very same reserves that are used for drinking water.
Bill Strong, Equinix: These are essentially customer deployments, and customers are running critical infrastructure and applications here.
PAUL SOLMANN: Bill Strong runs the Silicon Valley data centers for Equinix, which operates 260 of the roughly 11,000 data centers it operates around the world.
The company rents out space to companies like AT&T and Google Cloud to run servers that support the cloud and AI.
operation.
And Equinix is expanding.
Are AI-like processors here to stay?
Bill Strong: That's right.
This is high density deployment, liquid cooling.
So, you are basically using the building's cold water.
I'll come in here.
Each of these is sent to the actual chips on the customer's server.
Paul Solman: Are these nozzles?
Bill Strong: That's right.
There's a nozzle where there's a small auxiliary tube that connects to the server and cools the chip.
The hot air returns and is connected to a chilled water system where it is cooled.
This allows it to provide liquid cooling for high density AI type deployments.
PAUL SOLMANN: Just this Silicon Valley complex has 345,000 square feet of servers, thousands of which run 24/7. As of last year, the company's global energy budget was equivalent to 750,000 U.S. homes.
So what percentage of global energy usage do the roughly 11,000 data centers around the world use?
Kate Crawford: Estimates range from 2 to 8 percent.
Paul Solman: But even at the lower end?
Kate Crawford: 2% is roughly the size of the Dutch energy budget.
PAUL SOLMAN: And what if, as is widely expected, it doubles in two years?
Kate Crawford: Energy budgets could be as high as in a country the size of Japan.
Paul Solman: Today, Meta's owned and operated data center complex in Iowa consumes as much electricity per year as 7 million laptops running eight hours a day.
And of course, CHRISTOPHER WELLISE, Equinix: We're still in the very early stages of artificial intelligence.
Paul Solman: But as an A.I.
You can take, but you can also give.
Christopher Wellise is responsible for sustainability at Equinix.
Christopher Wellis: We don't yet know what the benefits to society will be from an energy perspective, for example.
Paul Solman: For example?
Christopher Wellis: Air Canada is an Equinix customer, and we are able to optimize their flight routes and save them fuel.
There's a lot of attention on how much energy AI consumes
It uses less energy, but it doesn't lose the energy consumed in training these large language models, for example.
It can be thought of as stored energy.
These models are trained once and can be retained for repeated use.
Kate Crawford: There are some signs that give me hope.
PAUL SOLLMAN: Kate Crawford agrees.
Kate Crawford: Researchers are currently investigating different kinds of technical architectures, in particular so-called “small language models.”
These are models that use significantly less data and therefore less energy.
Regulators are starting to take notice.
The first AI-specific bill has been introduced in Congress.
Environmental impacts of.
Bill Strong: The solar panels provide half a megawatt of power.
Paul Solman: A great example is Equinix itself.
Bill Strong: When the solar panels are running, they don't get power from the local utility.
Paul Solman: Well, right now these panels power just 3% of the facility, but with solar capacity doubling every three years, the possibilities are potentially endless.
Reid Hoffman, Founder, LinkedIn Corporation: I think ultimately it's all going to be very positive.
Paul Solman: LinkedIn founder and technology optimist Reid Hoffman believes in AI
itself.
Reid Hoffman: A version that was applied to data centers 10 years ago found ways to save 15 percent of data center power on the continuous week-to-week operation of the data center.
So, if you're generating A.I.
Anything that helps with these things, obviously it takes a lot of electricity to train, but then it helps us figure out how to run the power grid more efficiently, right?
And that is already the outlook today.
So ultimately, I think the concerns about power are rather misleading.
Paul Solman: Really?
Reid Hoffman: Yes.
PAUL SOLMAN: But in the short term, Kate Crawford warns: KATE CRAWFORD: If the grid continues to come under these pressures, it's inevitable that prices will rise.
Paul Solman: In the case of AI, the age-old race between the costs of a new technology and its benefits
Maybe to solve the problems it causes, maybe not.
Paul Solman joins us from Silicon Valley on the PBS “News Hour.”
