The Safe Face Pledge
launched last month as a “pledge to mitigate abuse of facial analysis
technology,” with four themes: “Show Value for Human Life, Dignity, and
Rights;” “Address Harmful Bias”; “Facilitate Transparency”; and “Embed
Commitments into Business Practices” (SAFE).
The full pledge is inspirational and comprehensive, covering bias,
secret and discriminatory state surveillance, risking human life, law
enforcement abuse, auditing customer compliance, communicating the
systems’ workings, and making your legal documents (from vendor
contracts to terms of service) reflective of your values.
The pledge’s announcement describes how the UK’s notoriously inaccurate
police facial recognition systems are more likely to falsely accuse
black people of being a match for a criminal than people of different
ethnic or racial backgrounds.
That reminded me of something that EFF executive director Cindy Cohn described on a panel last month:
Cindy pointed out that there’s a danger in centering the critique of
facial recognition in racial bias, because this bias is the result of
the systems not being trained with enough images of racialized people.
When a Chinese state facial recognition system ran into this problem,
the Chinese government simply bought the driver’s license database from
an African client state and used it as training data, eliminating bias
in the algorithm’s false positive rate, by massively invading the
privacy of millions of African people, and now the system is even better
at tracking black people.
