ImpactAlpha, Dec. 7 – Ask not what artificial intelligence can do for the climate. Ask what the climate challenge can do for AI.
By applying AI to meet the greatest existential threat facing humanity, AI can perhaps be absolved of itself being the greatest existential threat to humanity.
Slowing global warming is made to order as AI’s mission, the kind of huge, complex, multivariate, interdependent problem and solution set that can’t be fully grasped by a single human mind. Calling Sam Altman: Putting AI on the climate case in a major way could not only redeem the technology, but the humans. And the humans would be forever grateful.
The power of artificial intelligence (as well as its potentially massive energy consumption and carbon emissions) has been the talk of the COP28 global climate summit underway in Dubai. Corporate backers such as Google and Microsoft and global consultancies from BCG to PwC have been touting the efficiencies made possible in emissions monitoring and reduction, disaster forecasting and recovery and decarbonizing industrial processes.
The US is leading an effort “to put AI for climate literally into the text of the outcomes here at this COP,” US deputy climate envoy Rick Duke said in a panel discussion.
AI-driven insights, for example, have enabled American Airlines to cut its planet-heating contrails by more than half. Google claims its Maps users have avoided more than 2.6 million tons of greenhouse gasses since the introduction of AI-powered eco-friendly routing two years ago. And Al Gore’s Climate Trace monitoring tool is able to track the detailed emissions of more than 350 million sites thanks to AI and a network of satellites.
An even bigger prize may be in accelerating the painstaking research and development on emerging climate technologies and speeding them on the path of declining cost curves and commercial scale. AI models can scour scientific literature to home in on promising materials and processes in a fraction of the time it would take human researchers to do.
One example: nuclear fusion, long the holy grail of abundant, cheap, clean energy. Google’s DeepMind trained an AI learning model on the physics of plasma – the finicky superheated material that creates a fusion reaction. DeepMind’s learning model taught itself how to control plasma shapes inside a reactor, a complex key to commercializing fusion energy.
Climate social intelligence
And the even bigger prize would be forging a global consensus on the path forward and the most urgent action items. That consensus will be sorely tested at COP28 as 196 nations try to forge a final agreement that most likely will skirt the crucial question of phasing out fossil fuels because of objections from Saudi Arabia and other oil-producing countries.
To test that hypothesis, ImpactAlpha put some questions to ClimateGPT, a custom-built generative AI chatbot developed by Erasmus.ai, an Amsterdam-based technology company that presented the new tool at a session at COP28. COP President Sultan Al Jaber’s contention that there is no science or scenario that shows that the phaseout of fossil fuels will achieve the goal of limiting global warming to 1.5 degrees Celsius raised hackles at the conference. We asked, “Is this true?”
A: “No, the International Energy Agency has said that “phasing out fossil fuel use is the single most important step to limiting global warming to 1.5 °C”.
Even if the world does quickly phase out fossil fuels, ClimateGPT said we won’t be on track to stay within 1.5 degrees of warming. That will take “rapid and far-reaching” changes, according to the Intergovernmental Panel on Climate Change. The chatbot enumerated the well-known, if not yet well-implemented set of required actions. “The world is not on track to make the necessary changes to achieve this goal,” it said. “So should we just give up?” we asked.
A: “No, we should not give up…It is up to us to implement these changes.”
Q: “Does ‘us’ include the ClimateGPT chatbot?”
A: “No, the ClimateGPT chatbot is not a real person and does not have the ability to make changes to the environment. The changes needed to limit global warming to 1.5 degrees Celsius must be made by real people, such as governments, businesses, and individuals.”
The developer of the chatbot, Daniel Erasmus, says in an upcoming paper that filling the gap in large language models specifically tailored for climate contributes to “climate social intelligence.”
“That extends beyond the conventional understanding of artificial intelligence,” he says. “It seeks to build a community of learners and collaborators in the realm of climate information, rather than merely acting as a chatbot.”
No discussion of AI is complete, of course, without noting the risks. The BCG study, commissioned by Google, identified the huge amounts of water, as well as electricity, required to power AI. And the report warned, “AI can be applied to both climate-friendly and climate-unfriendly applications (and) can narrow or widen disparities between the Global North and the Global South.” For its part, Google says it won’t develop specialized AI systems for oil and gas extraction.
But at COP28, the risks of AI took a backseat to the risks of climate catastrophe.
“We’re following a pathway that takes us to 2.9 degrees Celsius. That is a path to disaster,” Johan Rockström of the Potsdam Institute for Climate Impact Research, said on a COP28 panel exploring AI’s role in accelerating climate action. “To be honest, we may not even need AI to assess the outcomes of that future.”
