Some of the brightest minds are combining climate science with machine learning to address the impacts of the climate crisis, from illegal logging to flooding. Priya Donti is leading the charge
What links artificial intelligence and the climate crisis? On the one hand, they’re both driven by fossil fuels. The link between dirty energy and an ever-heating climate is well-established. As for AI’s connection, energy hungry data centres and certain AI-powered tasks, such as generating images, require vast amounts of power.
On the other hand, AI has the potential to dramatically reduce our climate impact – not least by finding smart ways to cut energy consumption and generally shrink our carbon footprint.
From a climate perspective, AI is both a threat and a promise.
If we’re to harness that promise, we need to bring the brightest minds in AI and climate activism together. Those are two tribes that haven’t always been on the same page. But this is changing, and that’s thanks to people like Priya Donti, co-founder of global non-profit Climate Change AI (CCAI).
Her own back story encapsulates this shift. In high school, a biology teacher sparked in her a passion for climate and sustainability. Then at university, she fell in love with computer science – and was promptly struck by what seemed like a weird omission in the discipline. “Other technical subjects – engineering, biology, physics, chemistry – all seemed to have an application to climate. But I wasn’t finding that [with computer science]. I had these two things, climate change and computing, both of which I really like and care about. How do I bring them together?”
One answer came when she read a paper by researchers at the University of Southampton. It made the case for using AI to “put the smarts into smart grids”, by better managing electricity supply and demand, and so enabling faster take-up of renewable energy.
We organised a [climate-focused] workshop at one of the major machine learning conferences, and there were lines outside the door, people trying to get in
That was in 2012, in the early days of AI, and “I was really hooked,” says Donti, who soon started a PhD on the topic. Yet, she still felt a little lacking in community. At which point, while presenting her work at a conference, she bumped into David Rolnick. Formerly an intern at Google DeepMind, Rolnick is a fellow computer scientist with a passion for sustainability. “He pulled me into a lunch gathering, which was basically trying to rally people around the theme: how do we use AI for climate action?” And suddenly, there was Donti’s community. She quickly discovered how many others felt that way. “We organised a workshop at one of the major machine learning conferences, and there were lines outside the door, people trying to get in. There was a lot of excitement.”
And so, with the support of a range of funders and NGO partners, she and Rolnick set up Climate Change AI, a non-profit that “catalyses impactful work at the intersection of climate change and machine learning”.
With a small core team, it relies mainly on the enthusiastic engagement of volunteers from across academia, research and activism. (Donti herself combines her role at CCAI with that of assistant professor at MIT.) It stages workshops and online ‘happy hours’ (where AI and climate experts shoot the breeze), publishes papers, and runs seminars on everything from ‘AI-assisted discoveries in the soil carbon cycle’ to ‘multimodal AI approaches for urban microclimate prediction and building analysis’.
Donti combines her role at CCAI with that of assistant professor at MIT, pictured above
And if that sounds a touch geeky – well, that’s rather the point. CCAI is where geeks come to put their geekery to use. Because it’s a lot more than a talking shop: it’s about harnessing machine learning to develop specific solutions to thorny sustainability problems.
One of the most pressing of these, explains Donti, is data gaps: where we know there’s a problem, or even a potential solution, but we don’t have the granular data to act on it in a meaningful way. This was highlighted early on by CCAI’s third co-founder, Lynn Kaack, whose background is in climate policy but with an acquired AI expertise. “She was seeing [big] gaps in data: for example, in inventories of emissions from freight transport in various countries,” says Donti. “And she thought: ‘Well, we have satellite imagery: we can count trucks – can AI give us some answers?’”
It led to a strand of work, with initial funding from Google DeepMind, exploring how to fill gaps in data on everything from weather forecasts to species mapping. At a recent CCAI conference, DeepMind’s Anna Koivuniemi highlighted one such gap: knowing how much power to expect from a particular solar PV array at any one time.
Such knowledge is vital to optimising the way electricity grids work, and thus supporting the transition to renewables. In practice, though, predicting power output from arrays means knowing just how much cloud cover will be over them – a level of detail beyond the scope of standard weather forecasting, but one that could be captured via the vast data harvesting and interpreting capacity of machine learning.
By way of example, Donti cites the work of Open Climate Fix, a UK-based non-profit, which is doing just that. It combines data on actual solar power production, weather forecasts, video or other data on cloud cover and geographical information, with a view to providing the UK’s National Grid with a significantly more accurate – and localised – prediction of solar output, hour by hour.
Donti points to AI’s wider potential to match historical weather forecasts and climate predictions with actual recorded weather data: by ‘projecting [the forecasts] backwards’, it can work out how accurate they were, and what might be needed to improve their techniques for the future.
'We’re seeing more and more coming to the table, and they are some of the most intelligent and motivated people around,' says Donti
To outsiders, AI can sometimes seem lost in the IT clouds (computing and metaphorical), but the sort of work championed by CCAI epitomises its potential to get down and dirty with some real-world problems. With backing from funders such as Quadrature, Google DeepMind and Schmidt Futures, the non-profit has now awarded $3m (£2.26m) in grants to scientists exploring the application of AI to a whole array of real-world crises.
These include everything from tackling illegal logging, to boosting farmers’ resilience to floods in Texas and Fiji, through to helping Filipino shrimp farmers identify sites that combine high productivity with the potential to restore vital mangrove forests.
Plenty of detailed work on power systems and electricity grids is happening too, building on AI’s potential to save energy. These include a range of interventions to cut overall energy use across the economy and dramatic reductions in data centres’ energy consumption. For instance, when Google used the AI expertise of its newly acquired DeepMind to analyse the electricity used for cooling its data centres, it discovered it could cut it by a whopping 40% – and this was after it thought it had already made them as efficient as possible. At a stroke, that knocked 15% off the whole group’s data centre consumption.
Donti has been named one of Vox’s 2023 Future Perfect 50, and in MIT Technology Review’s 35 Under 35
It all adds up to a striking potential for AI to effect positive change with regards to the climate crisis, and could be on the cusp of what researchers are calling positive tipping points. Contrary to climate tipping points, which would trigger catastrophic, irreversible consequences, positive tipping points refer to fundamental shifts in the situation for the better.
In among all these techno-fixes, though, Donti and her colleagues haven’t lost sight of the ethical imperative at the heart of any humane AI strategy. CCAI’s mission statement cautions that machine learning ‘is not a silver bullet’ that can slay climate change singlehanded, and it doesn’t exist in a social vacuum. In something of a gesture of defiance in today’s political climate, it insists that ‘diversity, inclusion and equity are … fundamental to progress in addressing climate change.’
We’re seeing more and more coming to the table, and they are some of the most intelligent and motivated people around
Donti herself points out that, just as AI can be used to curb global heating, so it can and is being used to boost oil and gas exploration, or drive targeted advertising that is encouraging unsustainable consumption. “We shouldn’t push AI into all the places where it isn’t needed,” she adds.
It’s this combination of enthusiastic geek and standard bearer for social progress that surely inspired some of the accolades she’s received, including being named one of Vox’s 2023 Future Perfect 50 and MIT Technology Review’s 35 Under 35, among others. Those two facets of her character are summed up when I ask her what really excites her about her work with AI.
“What gets me most excited, more than the technology itself, is the people [involved]. We’re seeing more and more coming to the table, and they are some of the most intelligent and motivated people around. It makes me really optimistic to see all of the energy and motivation behind all this. It’s so exciting to see people throwing all their efforts at it. If this can continue, we’ll get it done.”
Photography: Cassandra Klos
