Technological solutions to climate change can be put into two categories. Vertical solutions that tackle pollution in one sector, say low-carbon fertilizers that help reduce emissions in agriculture. Or horizontal solutions that address issues across many different industries, say lithium-ion batteries that electrify cars but also better integrate renewables in the electricity mix.

Artificial intelligence is one such horizontal technology that has big potential to help cut plenty of planet-warming emissions. The list of problems AI can solve is long, and the number of startups using AI to solve some of those problems is longer still.

As with any new technology, there’s also a lot of hype. To cut through the noise, I spoke with Priya Donti of Carnegie Mellon University and David Rolnick of McGill University, two of the three co-chairs of the group Climate Change AI, which brings together academic and industry experts.

What exactly is AI?

“It’s a very broad term that basically covers any computational algorithm that can perform some kind of complex task,” Donti says. “Typically, tasks that humans can do like vision, speech, reasoning.”

There’s still a philosophical debate among AI researchers whether the goal of AI is to do things as well as a human — or achieve superhuman performance.

Machine learning is a type of AI application that is narrowly focused on drawing insights from large datasets. It’s probably what a human could have done, Donti says, but machine learning helps speed up the process.

How can AI help reduce emissions?

There are five broad ways to think about AI’s climate applications:

1. Distilling data into actionable insights. For example, data analytics company Kayrros uses satellite images and machine learning to help spot methane leaks.
2. Optimizing complicated systems. Fero Labs, based in the U.S. and Germany, uses machine learning to improve energy efficiency at cement, steel and chemical companies. WeaveGrid is helping electric grid operators better integrate electric-vehicle charging.
3. Accelerating scientific discovery. The startup Aionics helps speed up the experiments needed to find a new battery material.
4. Making climate simulations quicker. Researchers are using AI to cut the time needed to run big, complex models.
5. Improving predictions. The startup Kettle uses neural networks to improve forecasts of wildfire risk.

Could AI help increase emissions?

Yes. “AI accelerates whatever you ask it to accelerate,” Rolnick says. That means there are applications of AI that can increase emissions, too.

The biggest tech companies provide AI solutions to oil and gas firms to optimize the exploration and production of fossil fuels. Algorithmic recommendation tools on retail platforms likely help increase consumption. And self-driving cars may end up substantially increasing the number of miles driven.

So it’s important that those deploying AI solutions find ways of ensuring that, wherever possible, there’s alignment with global climate goals.

What are other potential pitfalls?

The sheer interest in deploying AI means that its applications are set to explode, says Donti. A recent virtual Climate Change AI workshop drew 2,000 participants. Researchers submitted 300 proposals to put AI to use tackling global warming.

“The big mandate for the community is that, even though AI is a flashy technology, there are many situations when a less flashy solution might be the correct one,” she said. “It’s important to make sure that AI doesn’t end up serving as a diversion.”

The goal would be to make the use of AI as widespread as spreadsheets are today. But to really help fight climate change, users need to be skilled and aware of the technology’s potential and pitfalls, Rolnick says.