Nachhaltigkeit ist heute mehr denn je ein zentrales Thema, das nicht nur die Umwelt, sondern auch die soziale und wirtschaftliche Entwicklung unserer Welt betrifft. In diesem Interview gab uns Dr. John Platt Einblicke wie Künstliche Intelligenz helfen kann, die Welt nachhaltiger zu gestalten, oder auch nicht.

How can AI fight climate change?  

AI will be a crucial tool in reducing greenhouse gas emissions and achieving net-zero emissions by 2050. AI can be an enabling technology for both mitigating climate change and helping reduce the harm caused by climate change. 

One example of mitigation is our contrail avoidance project, a project that I participated in for the past 3 years. The 2022 IPCC report noted that cirrus clouds created by jet engine exhaust account for roughly 35% of aviation’s global warming impact, which is almost as much impact as produced by jet fuel. We have used AI to analyze weather and satellite data so that we can detect and predict contrails, and match them to flights. The Applied Science team in Google Research teamed up with American Airlines and Breakthrough Energy showed that we can use the AI to reduce contrails on test flights by 54%.  

We are partnering with EUROCONTROL, the European Organization for the Safety of Air Navigation, to help bring contrail avoidance to Europe. Locally over Europe, contrails may cause as much as one-third the warming (measured in watts/square meter) as the rest of anthropogenic causes. 

We have also applied AI to helping people and businesses decide whether to install solar power on their own rooftops. This capability will make it easier for planners to determine the best building designs and solar options for urban areas. As we continue working toward a carbon-free world, making solar technology more accessible for cities and organizations will be important for reducing global emissions. 

Another example of AI fighting climate change is in transportation optimization. Google Maps offers routes to drivers in order to reduce their emissions, by avoiding hills and traffic. This has already saved 2.4 million metric tons of carbon dioxide. Also, we have just announced Green Light, a system that makes recommendations to cities for tuning their traffic lights. This traffic light tuning smooths out the flow of traffic and reduces both carbon emissions and air pollution. 

Will using AI make the problem worse?  

AI and ML can have a measurable impact on the climate, because of possible emissions due to the electricity needed to run them. One solution is to ensure that  AI runs on a cloud service that is carbon-neutral (or zero carbon), to limit the impact on the environment. 

That is why Google has matched our energy supply with renewable energy since 2017: we purchase as much renewable energy over a year as we consume in our operations. In Germany for example, last year our carbon-free energy score was at 96% while the German grid was at 56%. We are also working on being zero-carbon 24/7 by 2030: making sure that our purchases of zero-carbon energy matches our consumption, by region, for every hour of every day.  

What should we do first to make AI more sustainable?  

All energy-heavy computation should ideally take place on cloud services that are carbon-neutral today and zero-carbon in a few years. The advantage of running AI on carbon-neutral and zero-carbon cloud services is that it will encourage the development and construction of renewable energy sources around the world and Google is proud to lead the way here. Economic models have shown that large purchases of zero-carbon electricity has positive network effects and reduces emissions for everyone. 

With AI at an inflection point, predicting the future growth of energy use and emissions from AI compute in our data centers is challenging. Historically, research has shown that as AI/ML compute demand has gone up, the energy needed to power this technology has increased at a much slower rate than many forecasts predicted. We have used tested practices to reduce the carbon footprint of workloads by large margins; together these principles can reduce the energy of training a model by up to 100x and emissions by up to 1,000x. We plan to continue applying these tested practices and to keep developing new ways to make AI computing more efficient. Google data centers are designed, built, and operated to maximize efficiency—even as computing demand grows. On average, a Google-owned and -operated data center is more than 1.5 times as energy efficient as a typical enterprise data center and, compared with five years ago, we now deliver approximately three times as much computing power with the same amount of electrical power.  To support the next generation of fundamental advances in AI, our latest TPU v5 is proven to be one of the fastest, most efficient, and most sustainable ML infrastructure hubs in the world. 

We’re continuously investing in new ways to make AI itself more efficient and sustainable and to apply AI to projects that protect the environment and the climate.  

Thank you for the interview!