AI & climate: a first of its kind conversation on the hill

Last week, Ai2 partnered with the Bezos Earth Fund on a landmark event focused on AI & Climate with Senator Martin Heinrich (D-NM) as our honorary co-host. From the top of the Hart Senate Building with the beautiful Capitol dome as the backdrop, nearly 100 members of Congress and their staff, media, and think tanks gathered to learn from leading experts about the enormous potential of AI as well as its impact on our climate. AI and climate change are two of the most defining challenges of our time, and it's crucial we explore how they interact and influence each other.

Senator Ed Markey (D-MA), a champion for climate action, began with a stark reminder: "There is a Dickensian quality to AI. It is the best of technologies and the worst of technologies simultaneously. It's a technology that can enable and ennoble or degrade and debase. It has a dual personality…Artificial Intelligence has real costs."

Moderated by Politico Tech Correspondent Steven Overly, our panelists illuminated this concept with tangible examples from climate modeling and agriculture to power consumption and CO2 emissions. We heard from Andre Perkins, Senior Research Engineer on Ai2's Climate Modeling team; Jesse Dodge, Senior Research Scientist and Author of Green AI at Ai2; Bruce Erickson, Professor of Digital Agriculture at Purdue University; and Priya Donti, Assistant Professor at MIT and co-founder and Chair of Climate Change AI.

Transparency was a key theme of the discussion, particularly the lack of and need for transparency throughout the machine learning pipeline. As Perkins said, "We don't know most things about these models right now…what they are keying in on and how they are making their predictions." And while this level of explainability may not be achievable, reasonable disclosures such as how the model was trained and how much power it used are very useful - but there are no requirements or incentives for sharing this information. Dodge shared that "the CO2 emissions from training small AI systems consumed about as much electricity and released about as much CO2 emission as charging your phone. We don't expect that after somebody charges their phone they buy carbon offsets. But at some point, when you're training a massive model that might release as much emissions as burning a rail car full of coal, the responsibility lies with you to be transparent about what you're doing."

"[AI] is a technology that can enable and ennoble or degrade and debase. It has a dual personality…Artificial Intelligence has real costs. " - Senator Ed Markey (D-MA)

The discussion also surfaced examples of how AI is being used today. Donti highlighted the use of AI to accelerate the development of next generation batteries, such as those that power electric vehicles, and satellite imaging to support things like flood detection which can lead to better resource mobilization in affected areas. Erickson reminded us that "one of the key things of agriculture is that it controls about half of the environment in the US…Agriculture contributes 10% of greenhouse gasses, and we are one of the great possibilities for solutions." One such example is leveraging computer vision to target the use of herbicides and pesticides to specific areas instead of evenly treating entire fields.

Given the resource intensiveness of AI and the potential for electricity demand to double in the next two years, discussions of alternative energy emerged. Perkins commented, "Fusion is the moonshot candidate - if it does end up working it would be amazing and completely transform everything. But the likelihood of that coming to pass is extremely small, especially on the timescale that matters for the climate that is changing right now." Donti cautioned, "We are already in a situation where aggressive energy efficiency is necessary to decarbonize our power sector. Relying on moonshot supply side solutions as a sole solution to dealing with the growth of energy and AI use might be viewed as a diversion tactic to avoid thinking about the real problem."

The real problem and the significant opportunity must equally be considered as we move forward. Recent legislation, introduced by Senators Markey and Heinrich and Representatives Eshoo and Beyer is a big step in the right direction. The legislation calls for an interagency study to investigate and measure both the positive and negative environmental impacts of AI while also aligning on metrics and standards. The bill is widely endorsed by AI and climate experts, and as our colleague, Emma Strubell, said, "provides a roadmap for elucidating the complex relationship between AI and the environment, which will be critical to ensuring that AI technology is developed and deployed so as to have a net positive impact on our environment and society."

AI and climate change are intertwined, presenting both risks and opportunities. By embracing transparent and responsible development, harnessing AI's potential for good, and working collaboratively, we can turn this challenge into a catalyst for a more sustainable future.