Climate Change: How Can AI Help?
Alexandre Lacoste Alexandre Lacoste
September 26 7 min

Climate Change: How Can AI Help?

The summer of 2019 gave us some of the clearest examples yet of how climate change is transforming our world. The hottest June ever was followed up with the hottest July ever — which also turned out to be the hottest month in recorded history. Scientists memorialized the first Icelandic glacier to lose glacier status and predicted the country would be glacier-free in 200 years. And unprecedented wildfires raged in the normally frozen Arctic, throwing up a smoke cloud nearly the size of Europe.

The 2018 report from the International Panel on Climate Change gives us a stark time horizon: 20 years. Within two decades, we need to hit net zero greenhouse gas emissions if we are to avoid the potentially catastrophic consequences of significant global temperature shifts.

That 20-year time horizon is useful when looking at solutions, too. Some are long-term, over that 20-year horizon, whereas others are achievable within the next two decades. And to make an impact on climate change, we need to be making radical changes within the next 20 years.

For many people, reading these headlines leads to a simple question: what can we do? Eating less meat and biking to work doesn’t feel like quite enough, and the magnitude of the challenge is immense. As an AI scientist, I know that machine learning can give us new insights into challenging problems. I wanted to find out how AI can help in the efforts to mitigate the impact of climate change, so I reached out to a few close collaborators — and it turned into a much bigger project than we anticipated.

Bringing scientists together

NeurIPS 2018 was a chance to welcome the top AI scientists in the world to Montreal for one of the field’s premier conferences. It can seem like more happens on the sidelines, however, than in the conference hall itself.

At an informal meetup over lunch focused on climate change and disaster response, I met with American researcher David Rolnick, now a postdoctoral fellow at the University of Pennsylvania. We agreed that AI could have a significant impact on climate change if we brought the community together to develop solutions, and there was more work to do in answering the key question of what we can do.

We soon gathered a group of like-minded people and decided to organize a workshop at ICML in June 2019 to hear from scientists around the world what AI can do to help solve climate change. In order to guide the discussion, Rolnick proposed that we put together a whitepaper compiling input from some leading thinkers in the field. We wanted to know: what are the most important levers that involve AI we can use to have an impact on climate change?

What started as a lunch conversation ended up as a 55-page paper (with 42 additional pages of references), “Tackling Climate Change with Machine Learning,” featuring 22 authors including myself and Element AI co-founder and Turing Award winner Yoshua Bengio as well as Google Brain co-founder Andrew Ng, DeepMind founder Demis Hassabis, and Microsoft Research co-founder Jennifer Chayes. We released it on the first day of the ICML workshop and it received coverage in major publications such as National Geographic, MIT Technology Review, The Verge, and India Times.

Our workshop was a success, featuring Bengio, Ng, and others, including University of Colorado computer science professor Claire Monteleoni, whose pioneering work in data science and climate informatics helped lay some of the groundwork for our thinking. It was the first climate change workshop hosted at one of the main machine learning conferences, and you can still view a recording of the entire event.

The findings

There are many ways that AI can have an impact on tackling climate change. That said, it is not a silver bullet. There is no single way to solve climate change all at once, and AI is not a panacea for the planet. Many kinds of solutions and actions will be needed to reduce the extent and adapt to the impacts of climate change, and we need to work on many different approaches in parallel.

In our paper, “Tackling Climate Change with Machine Learning,” we covered a broad spectrum of areas where AI can help to reduce greenhouse gas emissions, such as energy, transportation, and agriculture, and where it can help in adapting to the consequences of a changing climate. Many climate-change solutions would not benefit from AI; we focused on those that could. The authors for each individual section gathered input from experts in each of these fields. We included discussion on how AI can help inform individuals and policymakers, because many of these solutions require action from governments, NGOs and civil society.

We included many specific examples in our paper where AI can help drive solutions for climate change. AI-driven research can lead to accelerated experimentation, allowing scientists to discover new approaches and test new ideas through software. This accelerated research can help in many areas including materials science, where scientists are working to discover new materials such as next-generation batteries and solar fuels that can better store or harness energy from variable natural resources.

Remote sensing is another area where AI could have an impact. Satellites now blanket the Earth, yet there is a lot of work to be done to use these images for tracking important indicators such as forest density, ecosystem biodiversity or agricultural yield. AI systems could aggregate all that information for constant monitoring of environmental indicators around the world, so we can better adapt to a changing climate. Also, since greenhouse gases by definition interact with light, satellites equipped with cameras that see beyond the range of human vision could give us more timely and localized information on emissions in a matter of hours.

The other potential applications are numerous: precision agriculture, decreased emissions from cement production, even nuclear fusion — where AI can help accelerate experimentation around controlling fusion reactions. Already, AI is being used to detect methane leaks in industrial processes, optimize freight routing, reduce building power consumption, and forecasting of renewable electricity production.

We hope our paper provides a starting point for those looking to apply AI to climate change mitigation and adaptation and provides value for our audience, including researchers, engineers, entrepreneurs, established businesses, and policymakers.

For those who want to apply AI to climate change, who want to do something, we provide a roadmap: learn, collaborate, listen and deploy. Identify and learn how your skills may be useful, hopefully with the paper as a starting point. Find collaborators and other experts to work with. Listen to what your collaborators say is needed, and gather enough input to ensure your work will have the desired impact. And make sure to deploy your work where its impact can be realized and maximized.

The future of AI and climate change

I believe that AI and climate change represents a vital but too often overlooked area of research with enormous potential. Our workshop and paper will, I hope, help draw attention to the important work already in this space and encourage new approaches. We need to find high-impact solutions and make sure that many of them can be implemented within 20 years.

For me, the biggest metric for the field isn’t the number of citations, or how big a person’s network gets, or how many co-authors you can find. It’s about impact: how many tonnes of greenhouse gas emissions are we preventing? How are we building tools for adapting to climate change? How are we making a difference, and how can we scale it up?

Collaboration is key. Effective solutions require cooperation between AI experts and other stakeholders working in areas relevant to climate change.

To learn more, please visit our website, which includes the submissions and videos from the ICML workshop, a link to the discussion paper, and our other initiatives. Engage with us and let us know your ideas about AI and climate change by getting in touch and subscribing to the newsletter.

We will be running a workshop at NeurIPS 2019 and are always looking for new people and ideas in our ever-growing community.