AI is increasingly being used to tackle climate change, but it has its own emissions problem
On a farm in St. Peters Bay, PEI, a black four-wheeled rover with two outstretched arms drives through a row of thigh-high green leaves, its giant wheels kicking up the red dirt of a potato field. It looks more at home in the dusty, red Martian landscape than on a farm.
“In fact, there are quite a few people who stopped on the street to see what was going on,” said Aitazaz Farooque, the interim associate dean of the school of climate change and adaptation at the University of Prince Edward Island (UPEI ).
Meet AgriRobot, a robot trained using artificial intelligence to identify disease in potato plants.
Farooque leads a team of researchers at UPEI (in collaboration with the governments of PEI and New Brunswick) who are using AI in new and innovative ways. AgriRobot is the brainchild of Charan Preet Singh, who is a master’s student in the university’s sustainable design engineering department.
“It will create a map with location information so even if someone comes in, they don’t need to be trained … they can load the map on their cellphone,” Farooque said. “It will direct you to where the infected plants are and get them out.”
As the climate changes, farmers face more challenges than ever before. From floods, droughts and disease to warmer temperatures and shifts in growing and harvesting seasons, the business of agriculture is changing rapidly, which means farmers – and technology – must always continue.
But there is an irony: While AI is helping with climate adaptation and mitigation, it has its own emissions problem. And it’s one that will only grow as AI is used for more and more applications.
AI requires a lot of computers – and energy
“AI is being used in all kinds of ways to address climate action,” said Priya Donti, co-founder and chair of Climate Change AI, a global non-profit organization that explores the use of AI in climate change. climate.
“From helping us better predict solar and wind on the power grid to helping us better integrate those into power grids … in real time.”
AI runs on computers – most of them – hosted in data centers around the world. As AI models run, they need electricity. If that electricity comes from a grid that uses fossil fuels, it contributes to emissions.
At the same time, the computers in those data centers generate a lot of heat and need to be cooled — often requiring more electricity.
“Running AI is running any other computer program. You have an input, you want an output,” said Yacine Jernite, a New York researcher who works for Hugging Face, a company that hosts open-source platforms where AI models are shared.
“It’s going to do lots and lots of operations. And doing lots of operations for one answer means there’s a lot of energy and electricity being used by the computer running those operations.”
The problem is, no one knows how much AI accounts for emissions in data centers.
“We really need to watch out for the growth of the AI emissions footprint,” said Donti, who is based in Cambridge, Mass.
“And fundamentally, one thing that’s challenging and getting is that there’s not enough transparency among the data center providers, among the machine learning entities that are actually creating the machine learning algorithms in terms of actual monitoring and measurement of greenhouse gas emissions.”
Predict wildfires before they start
As we face an ongoing climate crisis, scientists are trying to develop ways to help us deal with the consequences.
In the Global South, locust outbreaks are increasing, threatening food security. A new tool called Kuzi is helping farmers by providing real-time data using satellites, soil moisture, surface temperature, humidity and more to predict potential outbreaks. It can then send a notification to the farmers on their cellphones.
And as the danger of fires increases, engineers and scientists are developing new tools to understand and predict when they will start.
Dryad Networks, a company based in Germany, has developed solar-powered sensors that can smell fire even before the flames erupt.
“In the back [the] membrane … is a gas sensor that is sensitive to hydrogen, carbon monoxide and volatile organic compounds, “said the company CEO Carsten Brinkschulte. “So it is actually like an electronic nose that can actually smell the fire . And this is where AI comes into play: We run AI on the sensor, so it can recognize pre-trained machine learning models trained for the smell of fire.
The company has already deployed 20,000 around the world, with a pilot project in part of the California forest. Brinkschulte said Dryad has also started a pilot project with an unnamed organization.
AI comes at an environmental cost
AI has great potential, experts say, but first there needs to be a better idea of how much it contributes to emissions – and the shift to renewables.
“We need to green the grid, and we need to make serious choices about how we make AI models more efficient for the places where we use them,” said Donti, of Climate Change AI.
“But also, as we must do for each sector, reassess which uses are suitable for the electricity that comes.”
And this extends to personal use of AI, because not all uses of AI are equal. Showing it two pictures, one of a dog and one of a cat, and asking it to choose the cat uses less energy than asking it to create or calculate something.
While we may enjoy creating filters for ourselves or asking generative AI like ChatGPT, it comes with a cost in terms of emissions. In fact, one study suggests that every time AI creates an image, it uses enough energy to charge a cellphone.
“We certainly shouldn’t think of AI as something that doesn’t cost anything,” Donti said. “I think it’s very easy to see with this abstract thing on your computer that it doesn’t have any effect, but it does.”