A squad of top AI researchers, including employees of Google, Microsoft and Facebook, have signed up to an open letter demanding that Amazon stop selling its facial recognition technology to law enforcement.
The letter, published on Medium, cited research by Inioluwa Deborah Raji and Joy Buolamwini it claims showed Amazon’s Rekognition tech – as of last August – showed higher error rate when classifying the gender of darker skinned women compared to lighter skinned men (31 per cent vs. 0 per cent).
It also rejects a series of blogs by Amazon’s GM of AI, Matt Wood, and Michael Punke, VP of global public policy, which sought to refute those findings. Wood in particular, said that Raji and Buolamwini’s research used an outdated version of the platform, focused on facial analysis not facial recognition, and did not use the 99 per cent confidence level which AWS recommends for use of Rekognition by law enforcement.
Wednesday’s open letter took issue with a number of Wood and Punke’s arguments. On the recognition versus analysis issue, it described Wood’s arguments as problematic, and said that Raig and Buolamwini “made no claim to gauging accuracy, but sought to highlight the need for rigorous intersectional analysis of commercial face recognition systems.”
It also said Raji and Buolamwini’s research “investigates Amazon’s products within the context and society they are used” and raised the spectre of operators outsourcing the final decision to the system. Amazon argues that all Rekognition should be just “one artifact of many in a human-driven decision”
Lastly, the letter says “companies such as IBM and Microsoft have reproduced comparable data and results…which reiterates that ethnicity cannot be used as a proxy for skin type while performing disaggregated testing.”
They conclude by calling on Amazon to “stop selling Rekognition to law enforcement as legislation and safeguards to prevent misuse are not in place”.
As of time of writing, the letter had 55 signatories representing a where’s where of the top academic institutions in AI.
Whilst cynics would say it’s easy for academics to decry the commercialisation of AI, it’s worth noting that at least 17 signatories appeared to be from commercial companies – along with one former AWS chief scientist – highlighting the increasing pushback by AI flavoured techies against the directions their employers would like to direct their intellects.
Google alone counted for eight. Google staffers have kicked up a stink about the makeup of its recently appointed ethics board, and last year forced the company to backtrack on a contract providing AI to the US Dept of Defense.