Several public groups have written an open letter to President Biden and his administration, highlighting their concerns about the lack of action regarding the large amounts of energy used by generative artificial intelligence tools and their ability to create environmentally relevant misinformation. As early as October this year, the Biden administration attempted to resolve regulatory challenges in the development of artificial intelligence through an executive order signed by the president, which promised to manage the risks of the technology.
The order covers multiple areas, including the impact of artificial intelligence on employment, privacy, security and other aspects. It also mentioned using artificial intelligence to improve the national grid and review environmental research to help combat climate change.
Seventeen organizations, including Greenpeace USA, Tech Oversight Project, Friends of the Earth, Amazon Employees for Climate Justice and the Electronic Privacy Information Center (EPIC), wrote in an open letter to Biden that the executive order does not do enough to address the impact of technology on the environment, whether literally or through misinformation.
"…widespread use of LLM will increase carbon emissions due to its enormous energy needs and the carbon footprint of its growth and proliferation," the letter reads.
A recently published study by AI startup HuggingFace and Carnegie Mellon University highlights the significant amount of energy consumed by generative AI models during the training phase and frequent daily applications. The carbon footprint of generating 1,000 images using a model like StableDiffusionXL is equivalent to driving a regular gasoline car 4.1 miles.
"Silicon Valley's baseless hype claims that artificial intelligence will save the planet at some point in the future, but research shows that the opposite is actually happening now," the groups wrote.
In addition to concerns about the impact of AI on the planet, the organizations expressed concerns that AI could be used to spread climate disinformation: "Generative AI has the potential to amplify the climate disinformation that plagues the social media age and slow down efforts to combat climate change. Researchers have been able to easily bypass ChatGPT's safeguards to produce an article from the perspective of a climate change denier, arguing that global temperatures are actually falling."
We’ve already seen generative AI used to create disinformation about the upcoming US election and a number of other areas, where the main challenge is often convincing people that the generated content isn’t true.
The groups want the government to require companies to publicly report the energy use and emissions generated by AI models throughout their lifecycle, including training, updates and search queries. They also called for a study of the impact of AI systems on climate disinformation and the implementation of safeguards to prevent its creation. Finally, companies are also required to explain to the public how AI-generated models generate information, how accurately they measure climate change, and the sources of evidence for the claims they make.