The machine simply runs, and the public record accumulates.
In the summer of 2021, a Belgian artist turned the tools of surveillance back on those who govern, building a system that watches Flemish parliamentarians for signs of distraction and broadcasts what it finds to the world. Dries Depooter's 'The Flemish Scrollers' does not judge intent — it only records posture, the downward glance, the divided attention — and in doing so asks an old question in a new form: who watches the watchers, and what does watching actually tell us? The project sits at the uneasy boundary between accountability and automation, where transparency becomes continuous and context disappears.
- An AI system built by a single artist is now doing what entire press corps once struggled to sustain — monitoring elected officials in real time, without rest, without editorial discretion.
- Every time a Flemish lawmaker's eyes drop toward a phone during a live session, a video clip is extracted, named, and published to social media automatically — no human hand required.
- The system cannot distinguish between a legislator reading urgent policy documents and one scrolling idly through feeds, flattening all distraction into the same public accusation.
- No government response has come, no scandal has erupted, yet the precedent quietly hardens: if one artist can build this for one parliament, the architecture of automated institutional surveillance is now open-source.
- The deeper disruption is not the tool itself but what it displaces — the journalist's judgment, the editor's discretion, the human decision about what is worth making public and why.
Belgian artist Dries Depooter has built a system that watches politicians watch their phones. Called 'The Flemish Scrollers,' it monitors live-streamed sessions of the Flemish government on YouTube, using facial recognition and machine learning to detect the moment a lawmaker's gaze drops toward a device. When it does, the software automatically extracts a video clip, attaches the politician's name, and posts it to Twitter and Instagram. The accounts went live on July 5, 2021, and have been running without interruption since.
What the system can do is precise — it measures the posture of distraction and quantifies how long it lasts. What it cannot do is equally significant. The AI has no way of knowing whether a legislator is scrolling social media, responding to an urgent message, or reviewing legislative documents. It sees only the downward glance, not what fills it.
The project occupies an uncomfortable position. Elected officials conducting public business in already-public sessions are, in one sense, fair subjects for this kind of scrutiny. But the system offers no editorial judgment, no context, and no opportunity for explanation. A lawmaker checking a family emergency receives the same treatment as one mindlessly scrolling. Automation removes nuance and replaces it with scale.
There is also something revealing in the architecture of the project itself. Rather than relying on journalists to observe, synthesize, and report, Depooter has outsourced the watching to an algorithm and the publishing to social media. Scrutiny becomes frictionless and continuous — no editor decides what matters, no reporter chooses what to investigate.
The Flemish government has not publicly responded. No named lawmaker has triggered a broader scandal. But the precedent is established. The question is no longer whether the technology works — it clearly does — but whether institutions will tolerate being watched this way, and whether the public will find the watching meaningful or merely a performance of accountability.
A Belgian artist has built a machine that watches politicians watch their phones. Dries Depooter, who works at the intersection of artificial intelligence, surveillance, and social media, created a system called "The Flemish Scrollers" that monitors live-streamed sessions of the Flemish government in Belgium, automatically detecting lawmakers who look down at their devices and posting video evidence to Twitter and Instagram with their names attached.
The mechanics are straightforward. When the government broadcasts a session live on YouTube, Depooter's software—written in Python and powered by machine learning and facial recognition—scans the feed continuously. The moment a politician's gaze drops toward a phone in their lap, the system registers it. A video clip gets extracted, timestamped, and published to the project's social media accounts, complete with the lawmaker's name tagged in the post. The accounts went live on July 5, 2021, and have been running automatically ever since.
What the system can do is precise: it identifies the moment someone looks down at a screen and measures how long they stay focused on it. What it cannot do is equally important. The AI has no way of knowing whether a legislator is scrolling through social media, checking urgent messages, reviewing legislative documents, or reading news. It sees only the posture of distraction, not its cause or content. The tool quantifies screen time but remains agnostic about what fills that time.
Depooter's project sits at an uncomfortable intersection. On one hand, it offers a kind of radical transparency—elected officials conduct public business in public sessions that are already broadcast publicly, and this simply makes one dimension of their attention (or inattention) visible in a new way. The data is real. The behavior is observable. The public has a reasonable interest in whether their representatives are engaged during legislative work.
On the other hand, the project raises immediate questions about surveillance, consent, and what it means to automate public judgment. The system doesn't flag a lawmaker once and let the matter rest. It continuously monitors, continuously records, continuously publishes. There is no human editorial judgment, no context, no opportunity for explanation. A politician checking a message about a family emergency gets the same treatment as one mindlessly scrolling. The automation removes nuance and replaces it with scale.
The project also reveals something about how we now think about accountability. Rather than relying on journalists to observe sessions, synthesize patterns, and report findings, Depooter has outsourced the watching to an algorithm and the publishing to social media. The work of scrutiny becomes frictionless and continuous. There is no editor deciding what matters, no reporter deciding what to investigate. The machine simply runs, and the public record accumulates.
What happens next is unclear. The Flemish government has not publicly responded to the project. No lawmaker has been named in a way that has triggered a broader scandal. The accounts post regularly into what may be a relatively small audience. But the precedent is set. If one artist can build this for one parliament, others can build it elsewhere. The question is not whether the technology works—it clearly does—but whether institutions will tolerate being watched this way, and whether the public will find the watching meaningful or merely performative.
Notable Quotes
When a live broadcast starts, the software searches for phones and tries to identify a distracted politician using AI and facial recognition, then publishes the video to Twitter and Instagram with the politician tagged.— Dries Depooter, via his blog
The Hearth Conversation Another angle on the story
Why did Depooter build this? What was he trying to prove?
He's an artist who works with surveillance and AI as materials. This isn't a crusade against distracted politicians—it's an exploration of what happens when you automate accountability and let algorithms decide what the public should see.
But doesn't it feel like a violation? These sessions are already public, but there's a difference between being broadcast and being algorithmically hunted.
Exactly. The broadcast is passive. You can watch if you want. This is active surveillance—the machine is always looking, always recording, always publishing. It removes the human judgment that normally filters what matters.
Can the system tell the difference between someone checking their email about a sick child and someone playing Candy Crush?
No. That's the whole problem. It sees the posture of distraction but not the reason for it. Everything looks the same to the algorithm.
So is this accountability or just performance?
Maybe both. It's real data about real behavior. But without context, it's just a stream of clips. True accountability usually requires someone to synthesize the information and make a judgment about what it means.
What happens if other parliaments adopt this?
Then you've created a system where elected officials are under continuous algorithmic surveillance, and their attention is quantified and published automatically. That changes the nature of what it means to be watched in a democracy.