How Marketers Are Starting to Address Gen AI’s Green Problems

Despite predictions that generative artificial intelligence will drive a massive increase in energy demand, AI companies are only beginning to consider the climate impact of their products.

While some applications of the tech could support agencies’ sustainability goals, developing and using generative AI tools requires massive amounts of energy which, in turn, generates climate-warming greenhouse gas emissions.

“On the bright side, you’re going to have less travel, you’re going to have less human movement, because more of these mundane tasks are going to be done by AI,” said Larry Adams, CEO and founder of inclusion-focused AI platform X_Stereotype. “The flip side of that is the enormous computing power that is required to make these AI systems work.”

How Marketers Are Navigating AI’s Language Gap

Creating GPT-3—of which ChatGPT is a specialized variant—generated the same amount of emissions as driving 123 cars for a year, researchers found. Asking a question to an AI bot could require roughly 10 times the power needed for a standard search engine query.

As AI tools weave their way into the daily habits of workers and web users across the globe, the carbon footprint of the internet is poised to balloon well beyond its current estimated 4% of global carbon emissions. By one estimate, data centers and communication technology are predicted to reach 14% of global emissions by 2040.

Embed sustainability into the process

Because generative AI is relatively new, there’s little in the way of best practices when it comes to using the tech sustainably. Still, it’s important that marketing practitioners incorporating these tools consider the climate implications of all their work—including generative AI.

“[We need to be] making sure to work with partners that are taking [the climate impact of AI] into consideration,” Elav Horwitz, executive vice president and global director of applied innovation at McCann Worldgroup, said at Adweek’s NexTech.

Choosing partners based on their climate commitments is one way to ensure that sustainability will be part of the ongoing conversation. But there are practical ways to avoid excess emissions, too.

“Every prompt we put into ChatGPT creates carbon emissions,” Horwitz said. “Even training people on how to write the prompt … to think for a second about what you want to get out of the gen AI, and then go to it and get the right output.”

AI’s equity implications

While X_Stereotype doesn’t currently track the carbon emissions of its operations, it’s in the works. Because climate change is something that disproportionately impacts communities of color, measuring and reducing impact is a “priority problem” for the platform, Adams explained.

“As fast as AI moves, this next wave of conversation is going to come up,” Adams said. “How is this being powered? How is this being distributed? Is it ethically distributed AI?”

Avoiding the Jevons paradox

When Karan Walia first began developing the AI technology that now powers his company Cluep, he was training models that are far less efficient than today’s.

“It took us two years just to train our initial AI model to recognize human feelings within your textual conversations on social media,” Walia said.
“[Now,] you can get up and running with a model in two days.”

Those hardware efficiencies have major implications for the energy demands of AI tools. But as often happens, efficiency gains can be outpaced by wider adoption, resulting in an overall increase in demand on resources. This phenomenon, referred to as the Jevons paradox in economics, often stands between potential climate solutions and real emission reductions.

“Ultimately, the energy itself needs to be produced in sustainable ways that are less harmful,” Walia said.

But creating guardrails to prevent inefficient and harmful uses of AI can also promote better practices across the industry. “There can and there should be regulation of how much compute your AI models are utilizing in production,” he added.