AI Models' Training Process Can Emit Massive Amounts of Carbon Dioxide
Wed, April 21, 2021

AI Models' Training Process Can Emit Massive Amounts of Carbon Dioxide

Tech giants competing to build stronger AI models pose a greater threat to the environment / Credits: Panchenko Vladimir via Shutterstock

 

Artificial intelligence is an answered prayer for many industries. Research firm IDC reported that global AI spending on AI systems is expected to reach $97.9 billion in 2023, which is more than three times the global spending in 2019. The report also predicted 38% of growth by 2023 in AI used in services, such as IT, as companies seek outside expertise to design and implement AI projects. There will also be a 35% growth in software, while 27% of growth will be in hardware. 

"AI is one of those game-changers that's becoming ubiquitous across all industries. It's going to proliferate and become more embedded in all sorts of technology,” Marianne D'Aquila, an IDC research manager with expertise in cognitive and AI systems, said. 

However, the success of AI comes at the expense of the environment. A recent study conducted by researchers from the University of Massachusetts Amherst revealed that the training process for several common large AI models can emit over 626,000 pounds of carbon dioxide. This is the equivalent of about 300 round-trip flights between New York and San Francisco or almost five times the lifetime emissions of an average car. 

Roy Schwartz, a researcher at the Allen Institute for Artificial Intelligence, stated that tech companies competing to build stronger AI models pose a greater threat to the environment. "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. 

According to S&P Global, an American publicly traded corporation, the study discovered that testing AI models is energy-intensive for several reasons. For instance, machine learning is very "data-hungry," which means the more data it consumes, the more energy is required. Also, AI testing involves taking large matrices and multiplying those matrices to make them larger and more skilled.