Support independent, student-run journalism.  Your support helps give staff members from all backgrounds the opportunity to conduct meaningful reporting on important issues at Stanford. All contributions are tax-deductible.

Combining artificial and natural

Artificial Intelligence Critical in Fight against Climate Change

By

Earth’s most groundbreaking environmental photographers work out of sight, miles above our heads. Satellites continuously capture images of the entire world, creating an incredible but underused bank of Earth images.

Now, scientists are inventing new ways to analyze these photos using artificial intelligence and machine learning to answer the world’s most pressing environmental questions. By providing data that help scientists develop strategies to mitigate and adapt to climate change, artificial intelligence can provide a path forward. On Monday, Jan. 27, at Stanford’s most recent “AI for Good” seminar series event, an expert panel of industry leaders and computer scientists discussed this growing area of research.

Technology affords us many conveniences and benefits, but important questions about our surroundings remain unanswered. We can order food at the click of a button, but we don’t know how many trees or species there are, said Lucas Joppa, Microsoft’s chief environmental officer. Until we can establish good baseline information, finding a solution to problems can be challenging. Without knowing how many trees exist, for example, we cannot understand how much carbon our forests are able to store.

Artificial intelligence could hold the key to tapping into the satellite image repository. Advancements in the field have made computer vision more accurate, which allows for precise identification of even small objects in photos. It’s the same technology that recognizes people in the photos you take on your phone. 

Together, satellite images and artificial intelligence can identify objects on a larger scale. Satellite photos of Earth are taken at three-meter resolution, meaning that each pixel in the image represents a three-square-meter area on Earth. That’s like taking a photograph of your computer screen from 50 yards away and still being able to zoom in to see an individual pixel. Even small variations meters apart are visible, making it possible to fine-tune data.

Already, artificial intelligence algorithms predict how productive different agricultural fields will be — even before the harvest, said Stefano Ermon, an assistant computer science professor at Stanford. They do so by learning from satellite images and past data. Forecasting famine and anticipating which farmers might need extra resources could avert disaster before it strikes.

An algorithm can’t solve world hunger, but it can certainly help. Algorithms and models “will allow us to come up with more data-driven, more quantitative, more scalable solutions to these really big problems,” said Ermon.

But there is still progress to be made. The fields of environmental science and computer science often work without a clear understanding of one another’s skills and priorities, preventing collaborations that tap into the strengths of both fields. As the fight to tackle climate change grows more urgent, so does bridging the gap between environmental and computer science. If we are to do everything we can to fight climate change, said Joppa, we have to involve machine learning teams in that process.

Environmental start-ups are making strides in connecting technology and the environment. For example, iNaturalist allows users to identify plants and animals they encounter by taking a photo. The app’s artificial intelligence algorithm determines the species based on data it has studied and learned from. In turn, the application collects users’ photographs to continue expanding that dataset. Scientists use this data to better understand the natural world, whether by surveying biodiversity or identifying plant disease. And because that data is crowdsourced, scientists have more information than they ever could if they had to collect it alone.

Crowdsourced photos and satellite imagery are rich and unconventional data sources that lend themselves to analysis by artificial intelligence. Algorithms don’t have to be perfect to make a difference, said Ermon. Even with 95% accuracy, when scaled to billions of images, the results prove valuable to scientists. We’ve seen some of the benefits of this data already, from predicting famine to studying biodiversity. As scientists keep collecting data and working together across disciplines, they will continue to discover new ways to use data — some of which we can’t yet imagine.

And there’s no reason to worry about a technological takeover. Sentient robots remain unimaginable to computer scientists.

“The biggest threat is not what AI is going to do if it gets away from us,” said Joppa. “The biggest threat is that we’re not actually going to get around to deploying it in pursuit of the solutions to the problems that really matter.”

To have any chance of finding those solutions, we need to get started now.

Contact Devon R. Burger at drburger ‘at’ stanford.edu