President Biden recently issued an Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. The order describes the purpose of artificial intelligence (AI) and details “guiding principles and priorities” for future AI policy. The 63-page document also orders nearly a hundred actions to be completed in the next 30 to 540 days. While focused primarily on security– individual, governmental, and national– and expanding AI research, there are also more specific action items targeting key concerns, such as how to guarantee civil rights and promote responsible use of AI in healthcare.
The executive order mentions the environment only briefly. One action item emphasizes the importance of AI in “strengthening our Nation’s resilience against climate change impacts and building an equitable clean energy economy for the future”. The order also includes the soon-to-be-created Global AI Research Agenda, which will include, among many other goals, the goal of advancing sustainable development. No further detail is given.
The inattention to environmental issues should be cause for concern. There are currently no environmental regulations in place for AI. There needs to be. Yes, AI may contribute to climate change resilience, but that doesn’t mean potential environmental consequences of AI can be overlooked. With this executive order calling for considerable amounts of new research and AI funding, including the establishment of several new AI research centers, environmental impact must also be addressed.
Training AI models has an intense environmental footprint due to the large amount of data processing. Since 2012, AI models in training have doubled their use of computing power every 3.4 months. Currently, training just one AI model can produce upwards of 626,000 pounds of carbon dioxide. That’s about the amount the average person emits in 67 years. And there are now thousands of new AI tools entering the market every day. It adds up fast.
AI drives other environmental impacts too. Training advanced AI models consumes hundreds of thousands of liters of freshwater used to cool computing servers. This water typically evaporates, meaning it is not then reused. This is particularly bad considering most water used to cool servers started as potable water. Some other issues include electronic waste disposal and natural ecosystem impacts of specific AI applications such as AI in agriculture, which could increase fertilizer and pesticide use.
This isn’t an argument against AI. There are numerous ways that AI can have environmental benefits. But AI is a new and unique threat. More research on the environmental impacts of AI and how to mitigate them is needed. As the executive order already calls for considerable research on AI, environmental impact should have been included in that call.
This executive order could have shown US leadership in addressing AI’s potential environmental impacts. Experts at Brookings, an independent research institute, agree. Joseph Keller, a visiting fellow, notes that the EU and U.K. are now asking for more transparency from their tech companies regarding carbon emissions and water usage. In response to President Biden’s executive order, Keller urges for the U.S. to initiate this practice as well. He specifically hopes that U.S. tech companies will “pledge to improve their transparency around the environmental impact of AI” and that the U.S. can become a leader in upholding AI sustainability.
Right now, the US is playing catch up to the EU. The EU continues to work towards establishing their AI Act, which guarantees “robust protection” for the environment. It also states that the environment is a ‘high-risk area’ for AI impacts. For all applications of AI, the EU asserts that environmental requirements must be met, leaving no doubt that protecting the environment is a priority.
It could be argued that there are too many complexities related to AI to include in Biden’s recent executive order. Here’s what could have been included for a start: acknowledge that AI presents environmental concerns and call for more research. This would have helped open up the conversation for future AI environmental regulations, like those coming in the EU’s AI Act.
While President Biden’s recent Executive Order regarding AI is an important start toward using AI safely and securely, the omission of environmental concerns is troubling. As nations explore AI’s potential and risks, we must learn from each other’s successes and mistakes. Establishing the environment as a priority is key as the AI regulation conversation continues.