Sunsets Authored by Amazon
Why Amazon won’t extend its one-year ban on facial recognition tools.
** UPDATE: Since the writing of this post, Amazon has announced that it will extend its one-year moratorium on its facial recognition technology. In this article, however, you’ll learn about why we shouldn’t be fully satisfied with Amazon’s move.
First newsletter post! I had a blast making this, though I’m punting on the branding as I figure out what this can become. Thank you to everyone who gave encouragement along the way. If you enjoy reading, please share with friends :)
In response to nationwide protests and activism following George Floyd’s death last summer, Amazon announced a one-year moratorium on police use of its facial recognition software. While Microsoft walked away from the technology until regulations are in place, Amazon hoped that one year would give Congress enough time to pass appropriate rules to curb the harms of these tools.
For years, activists and researchers have raised the dangers of facial recognition tools in the hands of law enforcement. A landmark study in 2018 demonstrated that these tools make incorrect predictions at higher rates for darker-skinned people and females, with the highest error rates for darker-skinned females.1 Further research showed that these tools perform worse for Asian and Indigenous people as well. So, when these tools fail, it’s people of color whose rights are violated.
Within the last year, Robert Williams and Nijeer Parks have shared their stories of being wrongfully arrested due to false facial recognition matches. But because few places have regulations to inform victims when such tools are used, it’s unclear how many more people have been wrongfully condemned. Even if these tools were accurate for all people, facial recognition software would nonetheless give law enforcement authorities heightened surveillance capacity to invasively monitor peoples’ lives.
With Amazon’s one-year ban set to expire this coming June and no federal regulations in sight, the company must once again take a position on the use of its facial recognition technology. Given its decisions in the past few years, I think it’s unlikely that Amazon willingly extends the ban.
Amazon’s history with facial recognition technology
Days before announcing the ban last summer, Amazon posted on social media platforms that it stands “in solidarity with the Black community.” Resembling the dozens of other corporate commitments to racial justice, Amazon’s declaration of solidarity was antithetical to its past dealings in the law enforcement ecosystem.
Amazon had been selling its facial recognition software to police departments since 2018. During an interview in early 2020, incoming CEO Andy Jassy admitted that the company had no idea how many police departments were deploying Amazon’s facial recognition tools.
When asked about potential police abuse of the technology, Jassy threw his support behind facial recognition tools and gave the benefit of the doubt to cops instead of overpoliced Black and Brown communities. “Let’s see if somehow they abuse the technology,” said Jassy. “They haven’t done that. To assume that they’re going to do it, and therefore you shouldn’t allow them to have access to the most sophisticated technology out there doesn’t feel like the right balance to me.”
Performing solidarity
One might argue that the examples above took place before Amazon’s declaration of solidarity with Black lives—perhaps the company’s leadership has reckoned with the racial implications of its technology and will extend its ban. Unfortunately, Amazon’s actions since last summer still signal otherwise.
Although the company banned facial recognition, Amazon continues to sell Ring, which is its commercially available smart doorbell with security features. As of early 2021, Amazon has maintained partnerships with over 2000 law enforcement agencies (up from 1300 in 2020) to use footage from Ring to aid in criminal investigations. And because users can report “suspicious” activity on Ring’s neighborhood watch app, the product enables civilians to become the neighborhood cops, which has unsurprisingly led to racist comments and accusations.
Amazon’s continued involvement with Ring suggests that the company’s one-year ban on facial recognition technology served more to placate public demands. Its commitment to Black lives is palatable to the extent that it doesn’t cause more damage to the company’s bottom line than public backlash. One long-term threat to the bottom line is government bans of facial recognition software, which Amazon has worked to limit in the past.
In September 2020, Portland, Oregon passed one of the strongest facial recognition bans in the country. Quietly lobbying behind the scenes was none other than Amazon. The company spent at least $24,000 to soften how “facial recognition” was defined in the bill, presumably to open loopholes for continued use of its technology. To Amazon, selling contracts for its facial recognition technologies are more important than the harms posed to vulnerable communities. Although Amazon was unsuccessful in changing this law, it’s unclear how much money Amazon has spent to weaken other legislation on facial recognition.
When Amazon ends the ban
In the next few weeks, I expect Amazon to announce a raft of internal regulations on its facial recognition technologies—everything short of extending the current ban. It might call for stricter evaluations on police departments when deciding contracts, improvements to the technology’s accuracy on people of color, or independent audits on the risks of government use of facial recognition.
By taking these steps, Amazon would portray itself as a responsible vendor of facial recognition software when in actuality the responsible decision would be to extend the ban, at least until widespread guardrails are in place and the risks of this technology are clearly understood.
When Amazon ends the ban this summer, we must continue to demand change from Amazon and from our elected representatives2 as well. After years of work by activists and researchers, 2020 was a landmark year when several vendors like Amazon placed bans that finally recognized the harms of facial recognition software. If last summer taught us anything, it’s that in our current society companies like Amazon won’t change course unless outside forces will them to do so.
☀️ Summer tidbits
🛹 Big hills 1: me 0. Took a tumble while skateboarding this week. At least my penny board didn’t get banged up.
🏀 After years of disappointment, my favorite team the New York Knicks has made the playoffs for the first time since 2012-2013. Never give up, kids.
👀 Google’s proposal for AI models like GPT-3 to produce search engine results sounds radically awful. Have y’all seen the problematic stuff these models output as truth?
🤔 Princeton’s Engineering School announced its completed DEI plan (which took a year). It has both good plans and grave oversights. We need to talk about this more!
📚 Currently reading Socialism . . . Seriously: A Brief Guide to Human Liberation by Danny Katch. A lighthearted discussion on capitalism that still gets the mental gears turning.
🎧 Binging the podcast In Machines We Trust by the MIT Tech Review. Can’t believe it’s been around since last summer—I found out about it just last week. Incredible reporting, stories, and critiques on AI.
Cover Photo by Pawel Czerwinski on Unsplash.
The researchers investigated performance disparities over skin tone as a broad proxy for many races, due to skin tone variations among people of the same race and similarities between people of different races. This research by Joy Buolamwini, Timnit Gebru, and Deb Raji is fantastic. Learn more about the research here and here.