Amazon is marketing an AWS-based facial recognition service to law enforcement agencies that purports to be able to identify people in real time, according to documents obtained in a Freedom of Information Act request by the American Civil Liberties Union.
The service, called Amazon Rekognition, is claimed to be able to detect up to 100 faces per image, as well as compare images to databases of photos containing tens of millions of faces, according to the documents. Amazon representatives cite law enforcement purposes as a common use case in this promotional video.
Amazon is not being at all secretive about these practices. This press release from November 2017 noted that the Washington County, Oregon Sheriff's Office "has been using Amazon Rekognition over the past year to reduce the identification time of reported suspects from 2-3 days down to minutes and had apprehended their first suspect within a week" of adopting the technology. Amazon also lists the City of Orlando Police Department as a customer, which uses the technology for "real-time detection and notification of persons-of-interests."
SEE: Artificial intelligence: Trends, obstacles, and potential wins (Tech Pro Research)
This public disclosure is what prompted the ACLU to investigate Amazon's courting of law enforcement agencies. Documents uncovered by that organization indicate that Amazon is offering free consulting to law enforcement in Orlando for their deployment, and that the city of Orlando is using the technology to analyze footage for "people of interest" from "cameras all over the city."
Amazon had also touted the ability to use Rekognition with footage from police body camera systems, though the ACLU notes that mentions of this type of interaction were scrubbed from the AWS website "after the ACLU raised concerns in discussions with Amazon," adding that this capability is still permissible under Amazon's terms of service. This change "appears to be the extent of its response to our concerns," according to the ACLU.
Naturally, using cloud services to build a panopticon is likely to generate concern among residents of the localities that have deployed the technology. Under optimal circumstances, this would be implemented following a referendum or, at a minimum, a period of public comment about combining surveillance technology with mass facial recognition. The ACLU sought documents indicating that any such outreach was attempted, though no documents were discovered. It does, however, point out the existence of an internal email from a Washington County employee stating that the "ACLU might consider this the government getting in bed with big data."
A report in Motherboard notes that Amazon's facial recognition technology has noticeable gaps in reliability, citing training documentation showing the software comparing a picture of O.J. Simpson with a picture of a white man with a mustache and long hair as a 93.53% match. Practically speaking, the only visual similarity between the two pictured men is that they both have faces.
Amazon's courting of law enforcement appears at odds with their corporate values. The ACLU, alongside various other civil rights groups including the Electronic Frontier Foundation (EFF) and Human Rights Watch, underscores this in a letter to Jeff Bezos, noting that "Amazon has opposed secret government surveillance. And you have personally supported First Amendment freedoms and spoken out against the discriminatory Muslim Ban."
This is a growing problem across the technology industry. Google employees are calling on the company to stop their collaboration with the US military on a project to analyze video taken from drones, while Google itself has scrubbed "don't be evil" from the opening of their code of conduct.
The growing concern of the use of such technologies highlights the potential controversy that could be caused by deploying artificial intelligence (AI) in both the public and private sector. Organizations interested in the technology should take their deployment slowly, examining all possible ethical ramifications before moving forward.
The big takeaways for tech leaders:
- Amazon is courting controversy by working with law enforcement to implement their Rekognition identification service.
- In a training slide, Amazon Rekognition software was unable to correctly differentiate between O.J. Simpson and a mustachioed white man with long hair.
- Special report: Turning big data into business insights (free PDF) (TechRepublic)
- FBI inflated encrypted device figures, misleading public (ZDNet)
- Amazon Web Services: A cheat sheet (TechRepublic)
- ACLU tells AWS: Stop selling facial recognition tools to the government (ZDNet)
- Phone tracking service LocationSmart exposed API, allowing anyone to track you (TechRepublic)
James Sanders is a Tokyo-based programmer and technology journalist. Since 2013, he has been a regular contributor to TechRepublic and Tech Pro Research.