The best post-apocalypse movie of 2021 reveals one vital reason we need A.I.
In Finch, Tom Hanks relies on a robot that predicts superstorms. IRL scientists are working on similar technology.
It’s been ten years since a fateful event devastated Earth, but Finch isn’t miserable.
He’s actually holding pretty well with his dog, Goodyear, in an underground compound powered by wind farms that shield his home from the intensely hot and dusty atmosphere above.
In the new sci-fi Apple TV+ movie, Finch, the titular character (played by Tom Hanks) isn’t overly preoccupied with the past so much as he is concerned about the future. But it’s not all post-apocalyptic roses: When Finch builds a C3PO-like robot with artificial intelligence and connects him to a local weather station, the A.I. informs him that a dangerous, end-of-the-world storm is coming very soon.
“A.I. forecasting is not perfect, but it is the most accurate method available today.”
As it turns out, you might not need to live in a turbulent post-apocalyptic world to experience the benefits of A.I. weather forecasting — especially as the climate crisis makes extreme weather events more common.
Inverse spoke with weather forecasting experts to fact-check the science behind Finch and discuss the ways we can use A.I. to predict the weather in real life.
“A.I. forecasting is not perfect, but it is the most accurate method available today. And it's fast,” John K. Williams, Head of Weather A.I. Sciences at IBM’s The Weather Company, tells Inverse.
Can A.I. predict extreme weather?
In the movie, Finch connects his robot to a local weather station. “I need you to use this data to make a forecast,” he commands.
It’s a plot device, but one rooted in real-life science. Right now, atmospheric science researchers are using A.I. to improve weather forecasting. That includes Sue Ellen Haupt, senior scientist at the National Center for Atmospheric Research (NCAR), tells Inverse.
“It is already state-of-the-art science to use A.I./machine learning in weather forecasting,” Haupt tells Inverse.
A popular website you’ve probably heard of — weather.com — has been using A.I. machine learning for decades to forecast weather. The Weather Company runs the site.
“Perhaps the most important way we use A.I. in the forecasting process is to intelligently combine information from multiple sources,” Williams says. He claims that “in general, our approach of using A.I. coupled with human forecasters… produces the most accurate forecasts in the world.”
Commenting on this system, Haupt says it takes “model simulations and past observations and trains an A.I. system to improve on the NOAA weather model simulations.”
In other words: A.I. is already using machine learning and pattern recognition to better predict the weather.
“This means that the [A.I.] algorithms will learn from past behaviors of the atmosphere — and how traditional numerical models performed to improve forecasts,” Jonathan Weyn, who studied A.I. weather forecasting during his Ph.D. research at the University of Washington, tells Inverse.
Google is also in the weather forecasting game.
In 2020, Google’s research team began collaborating with National Oceanic Atmospheric Administration to “explore the benefits of artificial intelligence (A.I.) and machine learning (ML) for enhancing NOAA’s use of satellite and environmental data.”
“The beauty of these methods is that they can process enormous amounts of data quickly to get near-term forecasts,” Jason Hickey, senior software engineer at Google Research, tells Inverse.
But A.I. weather forecasting isn’t perfect. Scientists lack sufficient on-ground observational data to train A.I. algorithms, requiring scientists to use satellite data to train A.I. This is limited by the scope of the satellites.
“In my mind, the biggest limitations are sufficiently dense and frequent observations to train the A.I. models,” Haupt says.
Is the robot in Finch realistic?
Finch’s robot scans data from a local weather station to make a lightning-fast prediction: several intense weather events — 155-mile-per-hour storms — will collide with each other the next 24 hours, creating a not insignificant probability of a superstorm.
The storm will last “40 days, approximately,” the robot drones, sending Finch into a tailspin of dread about how he and his dog will survive.
The short scene catalyzes the movie’s action-packed second act, but it also raises several questions, including: Can A.I. really make all the predictions we see in Finch?
Not with our current technology, which has taken decades to get “high-quality atmospheric measurements and data processing,” Weyn says. “The notion that a robot could do everything from data collection to forecasting remains quite far-fetched.”
But experts say Finch has two things going for it in terms of accuracy: the robot makes a short-term prediction and focuses on a large-scale weather event.
“The accuracy of weather forecasts depends on a lot of things, like the weather variables you want to predict, how far in advance you're forecasting [the forecast horizon], and even the particular environmental setup,” Williams says.
“The notion that a robot could do everything from data collection to forecasting remains quite far-fetched.”
“On average, shorter-term forecasts are more accurate than longer-term ones, and predictions covering larger-scale events are more accurate than small scale ones.”
Google’s research team recently wrote about using A.I. to predict rainfall within a 12-hour period.
“We can use A.I. to make high-quality weather predictions in the near term,” Hickey says. “For medium-range predictions of one to 10 days, traditional physics-based methods currently tend to do better than A.I.”
Williams thinks Finch’s biggest limitation is the set-up: It would be extremely difficult to forecast weather in a post-apocalyptic environment, where dust clouds make it difficult to see, monitor, and gather data from satellites.
“In a post-apocalyptic world, I suspect the biggest obstacle to accurate forecasting might be obtaining good data on the initial state of the atmosphere,” Williams says.
How can A.I. help us deal with climate change?
A.I. isn’t just helpful for forecasting fictional apocalyptic events, it can help us get ahead of extreme weather events in real life, too. According to a 2021 UN report, extreme weather events — especially excessive flooding — are becoming more frequent as a result of climate change.
“Researchers have shown lots of promising results on how A.I. can help forecast extreme weather events, including those that are becoming more frequent as our climate changes,” Williams says.
For example, Google’s Flood Forecasting Initiative uses A.I. to predict flooding and send alerts to people in flood-prone countries like India and Bangladesh, sending out 115 million warnings in 2021.
Hickey believes A.I. can be helpful in both the short term for crisis responders and for determining the probability of extreme weather events several weeks in advance.
“While A.I. methods are still in their infancy, they are proving to be very helpful for predicting such extreme events,” Hickey says.
The National Science Foundation also established the A.I. Institute for Research on Trustworthy A.I. in Weather, Climate, and Coastal Oceanography (AI2ES) with the aim of developing trustworthy A.I. weather predictions to help protect people and the environment.
“While it is good to recognize this potential, it is also important to understand that A.I. and machine learning are relatively new methods to the field,” Hickey says.
Finch is now streaming on Apple TV+
Reel Science is an Inverse series that reveals the real (and fake) science behind your favorite movies and TV.