Climate activist groups are pressuring the US government for legislation that forces tech companies to disclose the carbon emissions from their generative artificial intelligence systems.
In a signed letter to Senate majority leader Chuck Schumer, 22 groups including Amazon Employees for Climate Justice, Fight for the Future, Green America, the Union of Concerned Scientists, Friends of the Earth, and Accountable Tech ask the Senate to find ways to ensure that AI’s carbon impact doesn’t undermine the fight against climate change.
Schumer has been pushing to demystify the workings of AI amid the debate over how best to regulate the technology. With the Senate planning legislation, Schumer wants the US government to “treat AI with the same level of seriousness as national security.”
While acknowledging that AI could benefit American society, the activists warn that the widespread use of large language models (LLMs) will increase carbon emissions. Among their policy demands: Companies must publicly report energy use and emissions from the full life cycle of AI models. That includes training, updating, and search queries, along with all the data required.
The groups also want companies and their executives to be held liable for any environmental or other harm that results from the use of generative AI.
Technology giants such as Google, IBM, Microsoft, and SAP were underreporting their carbon emissions long before the current AI boom. Google says it’s been working to address the emissions from developing its AI models, even claiming to have have hit zero emissions in 2020, but there have been inconsistencies in how it declares its carbon footprint. Some companies only manage to disclose half of their emissions. In July, IBM launched a cloud carbon calculator tool to “spot patterns, anomalies and outliers in data that are potentially associated with higher greenhouse gas emissions.”
Tech companies burn an enormous amount of energy. That’s largely thanks to data centers, which need water to keep them cool. Google’s data centers alone used twice as much energy as San Francisco in 2020, and the company’s water consumption rose 20% last year.
Training a single AI model or chatbot can use more electricity than 100 US homes do in a year. ChatGPT used 1.287 gigawatt-hours of energy during training, about the same as the annual consumption of 120 American homes. A 2019 study found that creating GPT-3, which has 175 billion parameters, was the energy equivalent of driving 123 gasoline-powered vehicles for an entire year.
Meanwhile, the world’s data centers emit more carbon than the commercial airline industry. In the US, data centers use about 2% of all electricity, according to the Department of Energy. One forecast sees computing technology consuming 8% to 21% of global energy by 2030, with data centers accounting for a third of that usage.
The climate activist groups behind the letter to Schumer aren’t just raising concerns over massive energy use by LLMs like ChatGPT, which they want “monitored and disclosed transparently, to allow both consumers and policymakers to understand the trade-off of such technology.” They also highlight AI models’ ability to amplify disinformation about climate change. “The ease and speed with which people and organizations can use LLMs to produce and distribute climate disinformation threatens to perpetuate climate denialism and slow efforts to fight climate change.”