As image recognition is deployed throughout every aspect of our lives, the rise of adversarial input techniques has given rise to art and fashion meant to confound surveillance systems. This talk will begin by providing an overview of how optical illusions of various levels of technical complexity are leveraged by different artists as "black box" attack methods on various AI-driven surveillance systems. Then we'll explore how these techniques work on both computer vision and human vision in surprising ways, and give the audience opportunities to try and fool their own perception. Together we'll explore the promise that art offers in helping us better understand computer vision, and our own, so we can make informed choices about policies we need to define these technologies’ place in our world.
I'm part of an art vs surveillance working group, and would love to invite interested participants to join and continue to learn together, share projects, best practices, new literature, and more.
How will you deal with varying numbers of participants in your session?:If 30 participants attend, we'll constrain questions to the end, but most of the optical and audio illusions are DIY and report in chat, so should work fine with more folks.
If there are 3, I will really enjoy having some smaller group discussions, where folks are welcome to unmute and ask questions or talk into different topics as we go.
I work at the intersection of tech, art, and organizing. I currently lead a digital security lab for abortion access, and in my free time, conduct independent computer vision research.