Artificial intelligence can find your location in photos, worrying privacy experts
AILSA CHANG, HOST:
Artificial intelligence can recognize your face, your voice, even your writing style. Now it soon may be able to figure out where you are standing. NPR's Geoff Brumfiel has the latest on how AI can identify a location based on a photo and what that could mean for privacy.
GEOFF BRUMFIEL, BYLINE: A year ago, a group of graduate students at Stanford were taking a course on AI.
MICHAL SKRETA: CS 330, Deep Multitask and Meta Learning.
BRUMFIEL: Michal Skreta was one of the students. He says he and his partners needed a project, and they shared a common hobby.
SKRETA: During that time, we were actually big players of a Swedish game called GeoGuessr.
BRUMFIEL: GeoGuessr is an online game that challenges players to geolocate photos. It has a pretty straightforward setup.
SKRETA: You enter the game. You're placed somewhere in the world on Google Street View, and you're supposed to place a pin on the map that is your best guess of the location.
SILAS ALBERTI: The game is pretty popular, I'll add. It has 50 million players. It has world championships. It has YouTubers, Twitch streamers, like, pro players.
SKRETA: Silas Alberti is another member of the student AI project. Players race against the clock and each other. They try to pick a location based on clues, like street signs, how people are dressed, what kind of landscape is visible. The students wanted to see if they could build an AI player that could do better than humans. They started with an existing system for analyzing images that was built by OpenAI, the same company that makes ChatGPT. Then they trained that system with images from Google Street View.
ALBERTI: We created our own dataset of about 500,000 Street View images. That is actually surprising. It's actually not that much data. We were able to get quite spectacular performance.
BRUMFIEL: With a few additional modifications, the team's AI could figure out the location of any Street View photo anywhere on earth. It guessed the correct country around 95% of the time and usually guessed the location to within about 25 miles. Next, they pitted their algorithm against a human - specifically, a really good human named Trevor Rainbolt.
TREVOR RAINBOLT: Today, I got reached out to by these students from Stanford University that said they built a GeoGuessr AI for their class that I will not be able to beat.
BRUMFIEL: Rainbolt is a legend in GeoGuessing (ph) circles, but he'd met his match.
RAINBOLT: Pretty big city. I'm going to go over here. Wait, what? I thought there was a lake to the south, and it just 5-K'ed (ph), bro. I don't even want to win anymore. Well, I can't win, but...
BRUMFIEL: He lost multiple rounds.
SKRETA: We weren't the first AI that played against Rainbolt. We were just the first AI that won against Rainbolt.
BRUMFIEL: AI excels because it can pick up on all the little clues humans can and many more subtle ones, like foliage, soil, weather. The group says the technology has all kinds of potential applications. It could identify roads or power lines that need fixing, spot invasive vegetation. Skreta says it could be a great teaching tool.
SKRETA: But it's also about things like - you know, you have a photo. You like this destination in Italy. Where in the world could you go, you know, if you wanted to see something similar?
BRUMFIEL: This program can already geolocate photos not on Street View. I gave it a few from a road trip I took more than a decade ago, and it found most of them. It guessed a campsite in Yellowstone to within about 35 miles. Another photo taken on a street in San Francisco it placed to within a few city blocks. That's got some experts worried.
JAY STANLEY: From a privacy point of view, your location can be a very sensitive set of information.
BRUMFIEL: Jay Stanley is a senior policy analyst at the American Civil Liberties Union who studies technology. He worries companies might soon use AI to track where you've traveled. Governments might check your photos to see if you've visited a country on a watchlist.
STANLEY: Stalking and abuse is an obvious threat.
BRUMFIEL: In the past, Stanley says, people have been able to remove GPS location tagging from photos they post online. That might not work anymore. The Stanford graduate students are well aware of the risks. They've held back from making their full model publicly available precisely because of these concerns, but Stanley suspects the cat's out of the bag.
STANLEY: The fact that something of this degree of power was created by a student project makes you wonder what could be done by, for example, Google.
BRUMFIEL: In fact, Google already has a feature known as location estimation. Right now, it only uses the catalog of roughly a million landmarks - not the hundreds of billions of Street View images in Google's catalog. The company told NPR that users can disable the feature if they want. Stanley thinks the use of AI for geolocation will become even more powerful. He doubts there's much to be done, except to be aware of what's in the background of the photos you post online.
Geoff Brumfiel, NPR News.
(SOUNDBITE OF MOONSTARR'S "DETRIOT")
Copyright © 2023 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.