The artificial intelligence industry has skyrocketed in recent years, powering technologies behind smart speakers and self-driving cars, but that growth is coming at a cost.
Researchers at the University of Massachusetts Amherst recently conducted a study assessing the energy consumption required to train several common large AI models. The study revealed that the training process can emit over 626,000 pounds of carbon dioxide, nearly 5x the lifetime emissions of an average car, or the equivalent of about 300 round-trip flights between New York and San Francisco.
The benefits from the advancements in AI and other emerging technologies at the expense of the environment are simply not worth it, say many industry experts who are urging big tech companies to ramp up their sustainability efforts. Failing to do so could leave the companies' reputations at risk, they said.
Global spending on AI systems is expected to reach $97.9 billion in 2023, over 2.5x the $37.5 billion that will be spent in 2019, according to research firm IDC.
IDC predicts that 27% of the growth in AI by 2023 will be in hardware, which includes devices such as Amazon.com Inc.'s Alexa-powered smart speakers; 38% will be in services, such as IT, as companies seek outside expertise to design and implement AI projects; and 35% will be in software.
"AI is one of these game-changers that's becoming ubiquitous across all industries," Marianne D'Aquila, an IDC research manager with expertise in cognitive and AI systems, said in an interview. "It's going to proliferate and become more embedded in all sorts of technology."
Smarter, not greener
The push to advance AI is turning into an arms race of sorts among big tech companies, which are competing to build stronger models that deliver better performance. But they also pose a greater threat to the environment, said Roy Schwartz, a researcher at the Allen Institute for Artificial Intelligence, an organization that seeks breakthroughs in AI through research.
"The larger you make these models, the more energy they consume. If we continue this growth, we will see a much more significant negative impact on the environment," Schwartz said in an interview.
In July, Schwartz published a study indicating that the computations required for deep-learning research have increased 300,000x from 2012 to 2018. Deep learning is a subset of machine learning, a branch of artificial intelligence that configures computers to perform tasks through experience.
Testing AI models is energy-intensive for several reasons, Schwartz said. For one, machine learning is very "data hungry," meaning the more data it consumes, the more energy is required.
The fundamentals of AI testing also involve taking large matrices and multiplying those matrices to make them larger and more skilled. As these matrices grow, the model becomes more accurate. However, the larger the model, the more energy is wasted since more computations are needed to execute the training, Schwartz said.
An AI model's massive energy consumption also does not stop after training, Schwartz noted. This makes it difficult to determine whether its potential benefit to an organization will ever compensate for its never-ending consumption. As the research is still in its early days, there is no way to quantify which countries are consuming the most energy by running AI models, Schwartz said.
Referring to the University of Massachusetts Amherst study as "shocking," Schwartz said his main goal is to simplify the entire AI testing process, by helping organizations to make better use of their data and reduce the amount of energy needed.
Push for change
The United Nations Intergovernmental Panel on Climate Change issued a report in 2018 underscoring the need for a 45% reduction in greenhouse gas emissions by 2030 in order to combat the impacts of climate change.
The U.S. soon after released its National Climate Assessment as mandated by law under the Global Change Research Act of 1990. The assessment warned that failure to significantly reduce greenhouse gas emissions in the U.S. could have detrimental impacts on the economy, agriculture and public health.
Emily Kreps, global director of investor initiatives at CDP, a nonprofit global environmental disclosure platform, said asset owners are urging investors and asset managers to look more closely at how their own portfolios could be at risk.
"They want to ensure that 10, 20, 30 years down the line, their portfolios are resilient against a lot of the environmental issues that we're faced with now," Kreps said in an interview.
CDP in June joined forces with a group of 88 investors that have a combined $10 trillion in assets to target companies that do not disclose their sustainability efforts. Amazon was included on the list, along with 707 other companies, for its "high environmental impact and low transparency."
Amazon shareholders in May urged the company to publicly report how it is preparing for climate disruptions, but the proposal was defeated. Amazon's board argued in a proxy statement that doing so is unnecessary, as it already reports adequate information in this area.
Big tech takes action
In September, Amazon pledged to reach 80% renewable energy by 2024 and 100% renewable energy by 2030, with the ultimate goal of net zero carbon by 2040. Net zero carbon is when the amount of carbon dioxide emissions released annually is zero or negative.
“It’s going to be challenging, but we know we can do it, and we know we have to do it,” Amazon CEO Jeff Bezos said at an event announcing the climate pledge. CDP's Kreps applauded Amazon's latest efforts but urged the company to disclose its efforts through CDP, saying this action would promote further transparency and accountability.
Amazon competitor Microsoft Corp. powers 100% of its global operations by renewable electricity already, and fellow tech giant Apple Inc. reached its goal of powering its own facilities with 100% renewable energy in 2018. Similarly, Alphabet Inc.'s Google LLC in 2018 achieved 12 consecutive years of carbon neutrality and matched 100% of the electricity consumption of its global operations with renewable energy.
Sue Reid, vice president of climate and energy at Ceres, a sustainability nonprofit organization, said these companies must continue to prioritize climate change or face significant financial and reputational consequences.
"If they're investing in the wrong things now that aren't consistent with decarbonization and failing to seize opportunities that they can benefit from in terms of clean energy and climate solutions, they're going to be in a bad place,” she said.