Demystifying Clearview AI: Summary
Demystifying Clearview AI Blog Series (Summary)
Facial recognition dates back to more than half a century and has seen recent breakthroughs with the advent of web platforms such as Google and Facebook. The deep learning models that underpin modern facial recognition technology continue to improve as their underlying dataset continues to grow, encouraging companies such as Clearview to amass continuously larger repositories of personal information from the public web.
Cloud platforms such as AWS make it incredibly easy to scale data collection from online sources. Open-source tools such as TensorFlow and PyTorch make it very simple to apply state-of-the-art deep learning object detection and recognition techniques.
Using off-the-shelf technologies such as ImageAI and publicly available CCTV feeds, we were able to design, implement, and deploy a deep learning application to effectively track vehicle movement across a stretch of a public road.
Despite recent breakthroughs, deep learning is flawed by design. Adversarial perturbations make it possible to evade face detection technologies and surreptitiously impersonate a target individual.
Methods such as infrared cloaking provide some level of practical protection against face recognition. Privacy laws like the CCPA and BIPA empower citizens to challenge companies such as Clearview, which are actively collecting personal information in ways that irreparably endanger privacy.
The use of facial recognition in law enforcement implicates important Constitutional values such as Due Process and Equal Protection under law. Advocacy by organizations such as the Electronic Frontier Foundation, the American Civil Liberties Union, and the Constitution Project has done a great deal to defend those values. While at the same time providing recommendations intended to guide policymakers on how to properly regulate facial recognition technology in a manner that both protects constitutional principles and aids in public safety.
