The Art of Surveillance: Exposed by Dries Depoorter

Art & Sculpture Artificial Intelligence
A person in the foreground is viewing a wall of 10 screens with exposed wires, each displaying security camera footage.
This article appeared in Make: Vol 87. Subscribe for more great projects.

You’re out sight-seeing and you post a selfie to Instagram. Before long, an unknown computer posts a video of you — taken before, during, and after your selfie! — from a nearby security camera. What the? Is Interpol after you? The NSA?

Actually it’s just one guy. The Follower (2022) was the provocative work of Belgian technology artist Dries Depoorter, who exploits unsecured cameras and APIs to expose the dark sides of cybersecurity, privacy, social media, and artificial intelligence.

Featured image: Seattle Crime Cams (2017) — City police share data on where they’re going in real time. This project finds live cameras nearby, with live police radio.

Dries Depoorter presenting in Copenhagen, 2018. Photo by IO

With a background in electronics, Depoorter also studied media arts and worked in advertising. Now he makes art for a living, and one of his main mediums is AI, applied to ubiquitous internet-connected surveillance cameras: security cams, traffic cams, doorbell cams, cameras in somebody’s living room, lobby, or legislature.

I chatted with Depoorter to learn more about how he deploys AI to spy, and why.

How did you become focused on surveillance? Was there an incident that turned you in that direction?

Surveillance, privacy, AI, and the role of technology in society have always intrigued me. While there’s no singular event in my life that steered me toward this topic, I’ve always been fascinated by the tension between the benefits of technology and the potential threats to our privacy.

Border Birds (2022–2023) — “I made this project with my sister Bieke Depoorter of Magnum Photos. I found a lot of open cameras on borders of two countries. I use machine learning to capture images of birds that are crossing those borders, then we print them out and sell them.”

How did you find open cameras for your work? Lists of public cams? Or did you hunt them down?

I mainly use Shodan, a search engine for internet connected devices. Shodan provides an extensive database of cameras and other devices that are openly accessible on the internet. I haven’t had the need to resort to wardriving.

What’s your go-to technology for image recognition?

In most of my installations I use Raspberry Pi. If I need something more powerful I use Nvidia Jetson Nano; if I still need more power I will use AWS. In most of my work I use Keras for face recognition, and for object detection I use Yolo.

The Follower

The Follower (2022) — “Uses open cameras and AI to find out how an Instagram photo was taken. I recorded a selection of open cameras for weeks, then scraped all Instagram photos tagged with the same locations. Image recognition software compares the two.”

So what was the reaction to The Follower?

It was intense. The reaction was, “If only one guy achieved that with simple open source software, imagine what a government can do with access to way more cameras, way more computer power, and way more resources.”

I had to take down the video because of copyright issues with public cameras, some companies claimed they had rights to the images.

In your view what are the most highly surveilled cities? I’ve heard London? Does the USA compare?

Yes, I think it’s London. But in some countries it’s easier to find open cameras.

Jaywalking

Jaywalking (2016) uses live cameras at street intersections, some public, some just had the default password!”

Jaywalking shows how pervasive cameras are, but also how easily we turn into informants in a surveillance society. It reminded me of that psychology experiment where obedient subjects were willing to press a button to torture someone!

Absolutely. Jaywalking highlights not just the reach of surveillance but also how it nudges us to police each other. It’s reminiscent of the Milgram experiment, where authority influenced people to act against their morals. In both we see the power of external forces shaping individual behavior.

“I wrote image recognition software to detect when people are jaywalking. Then the screen asks viewers whether they want to report the jaywalker; if they press the big button, it emails the local police station with a photo of the jaywalker.”

Quick Fix

Quick Fix points out problems of trust and fraud in social media: bots, fake accounts, inflated user statistics. But it also gets at the influencer economy specifically — if attention is money, then why not cheat to keep those numbers up?

You’ve hit the nail on the head! Quick Fix shines a light on the hidden corners of social media. If I can manipulate numbers and attention, imagine what bigger entities with more resources can do.

A vending machine with a metal keyboard, coin slot, and screen showing prices for Instagram likes and followers.
Quick Fix (2019–2023) — “Installation allowing viewers to buy likes and followers. I currently have 50,000 fake accounts on Instagram and I can make sure they like or follow you. If I get asked by exhibitions I can produce more of these — I’m a huge fan of Raspberry Pi. No, the software is not open source!”

It makes us question what’s real and what’s not, especially when popularity and money are so closely linked. It’s a peek behind the curtain of the social media game, where sometimes, things aren’t as they seem.

Surveillance

A child in a gallery is looking up at a security camera and loudspeaker.
Surveillance Speaker (2018) — “Tries to describe what the camera is seeing, with a really natural voice. One of the most beautiful moments I had with this piece was, there was a [set of] twins walking around the camera, and the piece was saying, ‘I see a girl taking a selfie in a mirror.’ So sometimes it’s really poetical, and sometimes it’s creepy accurate.”

Your Surveillance pieces give us the uneasy feeling that any camera might be connected to AI that is classifying and reporting what is sees — even hunting for someone specific. It’s like the fantastical all-seeing NSA in the spy movies, but you’re showing that potentially anyone can do it.

Surveillance Paparazzi (2020) — “Microsoft has a paid API where you send in an image and it will recognize 200,000 celebrities around the world. So I made a bunch of computers that are constantly checking open cameras around Hollywood trying to find a celebrity. One screen shows the process, the other shows the results. They’re all false positives. Includes doorbell cameras!”

Sure thing! These projects show us just how much surveillance in combination with AI is a part of our everyday life. It’s clear that cameras and technology are watching in many places we go. It’s a reminder for us to be aware and think about how much we’re being seen.

The Flemish Scrollers

Screenshot of politicians during a meeting. Some have boxes around their hands and faces, indicating AI detected they're using their phone.
The Flemish Scrollers (2022) — “Tagging politicians with AI and face recognition when they use their phones during government meetings livestreamed on YouTube. A Python script uses Keras software to search for phones and identify a distracted politician, then a video is then posted to Twitter and Instagram with the politician tagged, and the message: ‘Dear distracted, pls stay focused!'”

With The Flemish Scrollers are you intentionally turning the tables? Governments can use AI to spy on us to make sure we’re not breaking the laws, but here we can use AI to keep an eye on the government while they’re making the laws!

A server rack with multiple screens. Two large ones show video of politicians, smaller ones below show the process behind the AI script.
The Flemish Scrollers (2022) — “In galleries it’s shown in a server rack that makes a lot of noise; I really like to show the hardware behind this kind of calculations. After I launched it, they changed the way they livestream — normally you had an overview of the whole room, but now they only film the speaker!”

It’s interesting, right? The Flemish Scrollers plays with that idea, flipping the script on surveillance. It’s about empowering the public and holding authorities accountable, reminding everyone that transparency should be a two-way street.

Short Life

A device that looks like a clock radio with exposed components. The display shows 40.30%
Short Life (2020–2023) — “A clock you can buy that shows how much percentage of your life is already done. I prototyped it with Arduino and then made my own PCBs. It uses [life expectancy] data from WHO for how old people get in every country. When you order it, you fill in your birthday and where you’re from, and it’s personalized; on the back I write your name. It’s an ‘edition of 1 million’ — we will only sell 1 million of it!”

This article appeared in Make: Vol 87. Subscribe for more great projects.

What will the next generation of Make: look like? We’re inviting you to shape the future by investing in Make:. By becoming an investor, you help decide what’s next. The future of Make: is in your hands. Learn More.

Tagged
Discuss this article with the rest of the community on our Discord server!

ADVERTISEMENT

Escape to an island of imagination + innovation as Maker Faire Bay Area returns for its 16th iteration!

Prices Increase in....

Days
Hours
Minutes
Seconds
FEEDBACK