As climate scientists and advocates scramble to find ways to turn the environmental tide for the better, Google DeepMind Climate Action Lead Sims Witherspoon believes the answer could be found in artificial intelligence—by starting with the question.
“I approach climate change as a scientific challenge,” Witherspoon said from the stage at the annual Wired Impact Conference in London, focused on sustainability and environmental, social, and corporate governance (ESG). “I’m also a techno-optimist and an artificial intelligence product manager, so I can also approach it as a technological one.
“The first thing we do as scientists is we try to understand the problems we face,” she proposed. “Problem definition is where we start when we begin to come up with a solution.”
Earlier this year, Google merged its Brain and Deepmind AI teams under a single banner called Google DeepMind.
Witherspoon proposed a three-step plan for addressing climate change using AI called the “Understand, Optimize, Accelerate” framework. First, speak with the people experiencing the problem—in this case, climate change. Next, determine if AI applies to the problem and, if so, find an AI solution. Finally, focus on a path to deployment and impact.
“AI can be helpful in helping us understand climate change and its effects on Earth’s ecosystems,” Witherspoon said. “It can also help us optimize current systems and infrastructure because we can’t just start over from scratch today.”
What’s the Environmental Impact of Generative AI Tools?
Looking at the deployment pathway, Witherspoon noted that options begin to fall off the table due to the current regulatory environment, infrastructure limits, or other constraints and dependencies like limited data or viable partners. Even so, “it can help us with the breakthrough science and technology we need for a more sustainable tomorrow,” she said.
Witherspoon emphasized the need for a collective effort, adding that despite the strength of individual expertise, collaboration is essential and would require the combined efforts of academics, regulatory groups, corporations, non-governmental organizations (NGOs), and affected communities.
“Without talking with these folks—the ones who are experiencing or working on the challenges, every organization, whether it’s corporate, academic, NGO—they risk wasting time and resources,” Witherspoon said. “Which in the climate crisis are both incredibly valuable in this urgent task to solve the problems at hand, and we just can’t do that.”
Working with the United Kingdom’s National Weather Service Meteorological Office in 2021, Witherspoon said Google Deepmind utilized their extensive radar data to analyze UK rainfall with AI. Google fed the data into its Deep Generative Model of Rain (DGMR) generative AI model.
“We did a qualitative assessment with 50 meteorological experts at the UK Met Office, and over 90% of them preferred our methods—ranked them as their first choice—to the traditional methods they used before,” Witherspoon said. She added that the source code data and verification methods are freely available.
But while Witherspoon said AI can play a role in solving climate change, she also cautioned that the emerging technology is no panacea.
“AI is not a silver bullet,” Witherspoon said. “It’s important to say that AI will not solve all challenges driving the climate crisis; it isn’t even the right tool for many of the challenges that we face.
“AI also needs to be deployed safely and responsibly,” she continued. “Not to mention, until our grid itself is run on carbon-free energy, every energy-intensive technology will carry a carbon footprint, and that includes artificial intelligence.”
“Sometimes a simpler solution is better than a high-tech one,” Witherspoon added.
In May, Boston University professor Kate Saenko sounded the alarm on the carbon footprint of AI models and the impact on the climate of mass adoption of AI chatbots like ChatGPT.
In the report, Saenko noted that OpenAI’s GPT-3 model, with 175 billion parameters, consumed an equivalent amount of energy as 123 gasoline-powered passenger vehicles driven for one year, or around 1,287 megawatt hours of electricity. It also generated 552 tons of carbon dioxide, which Saenoko said came before the AI model was released to the public.
“If chatbots become as popular as search engines, the energy costs of deploying the AIs could really add up,” Saenko said.
Edited by Ryan Ozawa.
