AI researchers ask Amazon to stop selling face recognition to law enforcement

AI researchers from industry and academia have signed an open letter calling on Amazon to stop selling its face recognition technology to law enforcement after apparent biases were discovered.

Face palm: This January, MIT’s Deborah Raji and Joy Buolamwini revealed research suggesting that Amazon’s Rekognition product misidentifies women and people with darker skin more often than other subjects. Buolamwini has previously highlighted the racial bias in other face recognition systems (Microsoft addressed the problems with its technology identified in that work).

Change afoot: Face recognition has become a banner issue for those concerned about irresponsible uses of AI, and it seems increasingly likely that some form of regulation will arrive. But the technology is spreading rapidly, and companies are struggling to adjust their positions. Microsoft has said it will continue to work with law enforcement but has also backed legislation that would require signs showing where face recognition is being used. Google has said it won’t supply face recognition until it can come up with an appropriate policy.

War of words: The letter also counters criticism of a study posted in January by two Amazon executives, Matthew Wood and Michael Punke. Amazon’s rebuttals claimed that the original research misrepresented the capabilities (and limitations) of Rekognition. They also noted that Amazon requires its technology to be used in accordance with the law and said that Amazon would endorse greater transparency from law enforcement about use of the technology.

Pioneering vision: Those who signed the letter include prominent voices in AI and ethics as well as Yoshua Bengio, a computer scientist who recently received the $1 million Turing Award with two colleagues for his role in developing the deep-learning technology that underpins modern AI—which is crucial to face recognition. Bengio has recently emerged as a key voice on the risks of AI.