Levin talks less about teaching computers to see our world and more about teaching humans to make the world easy for computers to read. I believe he adopted this approach because he wanted to demystify computer vision for novice programmers. And, to be honest, it worked! When I first thought of computer vision, it intimidated me. By acknowledging that “the discipline continues to confront numerous unsolved (and perhaps unsolvable) challenges” (1), while still breaking down the algorithms, he made an intimidating field feel slightly more approachable. Levin made me understand that a computer cannot tell whether a scene contains a person or whether it is day or night. Frame differencing, background subtraction, and brightness thresholding (the methods he describes) do not understand, per se. Instead, they compare pixels against a previous frame, a stored background, or a threshold. By being honest about how little these algorithms actually ‘know’, he makes them feel within reach, at least for me. For instance, in his Limbotime story, the students’ breakthrough came from finding a sheet of white foam core in a closet, rather than from them being programming-savvy. That detail stuck with me because it suggests you do not need to be an expert to build something that works in computer vision.
However, this also complicates this technology’s ethical use. Levin showed us how much the same simple tools can mean depending on who uses them. While he admits that computer vision was long limited to military and law-enforcement purposes, he emphasizes artworks that use that same technology in completely different ways. Rokeby’s ‘Sorting Daemon’, for example, turns automated profiling into a critique of profiling. Jeremijenko’s ‘Suicide Box’, on the other hand, uses motion detection to count something that, in her words, “was not being counted – that is, doesn’t count” (6). Levin demonstrates that these works range from the highly formal and abstract to the humorous and sociopolitical. I believe that this range exists because the techniques are somewhat ‘simple’, allowing almost anyone to use them. This is where my feeling of it being approachable got a little complicated. If tracking a person is mostly counting pixels, then the difference between an art installation and a surveillance system is largely the intention behind it. So the same accessibility that makes computer vision welcoming to beginners is also what makes it easy to misuse.