美国国家公共电台 NPR--How AI is revolutionizing how governments conduct surveillance（在线收听）

How AI is revolutionizing how governments conduct surveillance

Transcript

Artificial intelligence is getting attention for its potential to bring huge changes to many different fields in the future, but experts say the AI revolution in surveillance is already here.

LEILA FADEL, HOST:

The world is full of cameras - on our phones, the corners of buildings, even satellites whizzing overhead. They're taking way more pictures than humans can look at. But now there's help from artificial intelligence, and intelligence agencies all over the world are getting in on the action. NPR's Geoff Brumfiel has more on how AI is revolutionizing surveillance.

GEOFF BRUMFIEL, BYLINE: Corey Jaskolski is the CEO of an AI vision company called Synthetaic. He's basically teaching machines how to look at images.

COREY JASKOLSKI: The way I think about it is, effectively, what we've come up with is a way to search massive amounts of data, almost like a search engine but for imagery, for visuals.

BRUMFIEL: It can work for everything from cellphone pictures to security camera footage. To show me, Jaskolski pulls open his laptop and brings up a top-down image of a city.

JASKOLSKI: What you're looking at here is high-resolution, aerial imagery of Milwaukee.

BRUMFIEL: Cities like Milwaukee regularly take this kind of imagery to check on zoning roads and other things. Normally, humans have to spend hours scrolling around. But Synthetaic's latest AI can find whatever we're looking for. We zoom in on a road.

What about - like, that looks like a delivery van or something.

JASKOLSKI: Yellow delivery van?

BRUMFIEL: Yellow delivery van.

JASKOLSKI: Sure.

BRUMFIEL: Jaskolski draws a little box around the yellow van, and suddenly the program starts finding more.

Oh, my gosh. They look like - I mean, that looks like exactly the same van all over town, right?

We scroll over to a busy intersection with another one of the vans moving between a lot of other vehicles. There's a white van, a yellow garbage truck. But the program only marks the yellow van.

JASKOLSKI: We've taught it to be a yellow van detector.

BRUMFIEL: I mean, I do have to say there is something creepy and Big Brother-ish about this. In 30 seconds, you've literally found a good dozen yellow vans in a city. And you can imagine how long it would take a human squinting at all those images like...

JASKOLSKI: No, that's right. That's right. I mean, it is a state change in the ability to find things quickly in large data sets.

BRUMFIEL: AI has been getting attention for its potential to bring huge changes to lots of different fields in the near future, but the AI revolution in surveillance is happening now. For decades, cameras have been watching over cities, businesses and even homes. But that footage has mainly been stored locally, and reviewing it took a pair of human eyes. Not anymore. AI systems can now hunt for a van in a city, scan license plates and even faces in real time. The system being developed by Synthetaic has many possible uses. An environmental group, for example, is trying to use it to track large livestock operations globally to monitor greenhouse gas emissions. Synthetaic's system really can find anything you want anywhere in the world.

JASKOLSKI: We've run searches, as an example, across the entire eastern seaboard of Russia for ships, and we can find every ship in a few minutes. It's pretty remarkable.

BRUMFIEL: Being able to scan the vast coastline of a nation like Russia is why this kind of technology has caught the eye of big government intelligence agencies. Watching everything that needs to be watched has always been a labor-intensive business. Even in George Orwell's famous novel "1984," the all-seeing thought police struggled to keep up.

GREGORY ALLEN: In "1984," there weren't enough people to watch all the surveillance cameras all of the time. But artificial intelligence can watch all of the surveillance cameras all of the time.

BRUMFIEL: Gregory Allen is director of AI and Advanced Technology at the Center for Strategic and International Studies. In the digital age, intelligence agencies are drowning in photos and videos.

ALLEN: There's just so much more data being collected than can realistically be analyzed by a human analysis workforce.

BRUMFIEL: Nations like China are investing heavily in AI for both domestic and international surveillance. In the U.S., intelligence agencies are already using it, though only to look overseas.

MARK MUNSELL: We do not spy on Americans.

BRUMFIEL: Mark Munsell heads the Data Digital Innovation Directorate of the National Geospatial-Intelligence Agency. It collects and analyzes huge quantities of satellite and drone imagery for the Pentagon. He wants to make very clear that this is not an Orwell novel. At the same time, he says, with a war in Ukraine, tensions rising with China and challenges all over the planet, agencies like his need AI.

MUNSELL: We have about the same amount of humans. And so there's only really one way to keep up with more and more sensors, and that's to apply as much automation as we can.

BRUMFIEL: Munsell's agency is currently using a set of AI tools called Maven to analyze several different kinds of imagery. It could let human analysts quickly spot potential targets, like tanks in a field or planes at an airbase. The exact details of how it works and what they're looking at remains classified.

MUNSELL: It has several lines of effort, we'll call it. I think that's the term that we use.

BRUMFIEL: But Maven has also stirred controversy. Google was involved with the project until its workers launched a protest over growing fears of weaponized AI. In a letter, they wrote, quote, "building this technology to assist the U.S. government and military surveillance and potentially lethal outcomes is not acceptable." It got thousands of signatures, and the tech giant eventually pulled out of Maven. Gregory Allen, who's been watching AI change the face of surveillance, says it's unrealistic to think the technology will go away.

ALLEN: Yeah. I mean, China is not going to give this up under any circumstances.

BRUMFIEL: At the same time, he says he thinks AI should be regulated, especially domestically.

ALLEN: I think it's entirely appropriate that there be safeguards and restrictions.

BRUMFIEL: As these tools grow ever more powerful and more available, many experts believe it's time to strengthen laws around using AI in surveillance.

Geoff Brumfiel, NPR News.

(SOUNDBITE OF EVOCATIV'S "TURMOIL")