Golan Levin’s essay on computer vision really hammered home the fact that a camera isn’t an eye — it’s just a sensor feeding a math equation. Having done CS IBDP HL, I was already pretty familiar with how control systems work and how cameras process input, so honestly, none of the technical side was surprising to me. I’ve spent enough time looking at how sensors translate the physical world into data arrays to know that a computer doesn’t “understand” what it’s looking at; it’s just running basic tricks like frame differencing to see what moved or background subtraction to see what’s new in the shot. It doesn’t see a “person”; it just sees a bunch of pixels that changed from gray to brown. This makes the computer incredibly easy to trip up with something as simple as a flickering light or a shirt that’s the same color as the wall.
Because computers are so literal and limited, artists have to do a lot of the heavy lifting physically before the code even runs. Levin points out that things like controlled lighting or high-contrast backgrounds aren’t just aesthetic choices — they’re necessary “cheats” to help the computer distinguish a human from the floor. I think about projects like Myron Krueger’s backlit walls, where the environment is specifically engineered to give the computer a perfect silhouette. It makes me realize that successful interactive art isn’t just about writing clever software; it’s about designing a space that “explains” the world to the camera so the algorithm doesn’t get confused by visual noise.
There’s also a weird tension in using these tools because, at their core, they’re surveillance technologies. Levin mentions works like Sorting Daemon or the Suicide Box to show how artists can flip the script on military-grade tracking. It’s a bit uncomfortable to realize that the same tech making a digital shadow look cool in a gallery is what’s used to profile people in public spaces. It makes me wonder if we can ever fully enjoy “interactive” media without that nagging feeling of being watched and categorized. It’s a reminder that while the interaction feels like magic, the data being pulled is never really neutral — it’s always being filtered through whatever narrow definitions the programmer chose.