Risk of Facial Recognition

Its important to remember that the latest emergence of AI methods are not perfect,  though much better than in the past, can produce erroneous results.   This can be for reasons of context in training, use or environment.    So there is risk involved.   So I always recommend at least a cursory look at risks analysis.  Much more if it shows higher indications of wrong predictions, sensitive domain or changing environment or regulation.

 Almost Everyone Involved in Facial Recognition Sees Problems   in Bloomberg    By Dina Bass

Facial recognition software is almost universally acknowledged by the scientific, technology, and legislative communities as flawed, with bias, mass surveillance, and other hazards making a strong case for regulation. In response, the Algorithmic Justice League and Georgetown University Law Center's Center of Privacy & Technology have introduced the Safe Face Pledge, urging companies not to provide facial artificial intelligence (AI) for autonomous weapons, and not to sell to facial recognition systems to law enforcement agencies unless explicit laws regulating their use are considered and approved. Facial recognition for surveillance, policing, and immigration is under scrutiny because scientists have demonstrated that the technology lacks sufficient accuracy for critical decisions, and performs worse on darker-skinned people. The Safe Face Pledge asks companies to "show value for human life, dignity, and rights, address harmful bias, facilitate transparency," by incorporating such commitments into business practices. The University of Washington's Ryan Calo said broad regulation and government oversight could complement pressure from workers and customers for companies to practice ethical AI deployment.  ... "