In 2016, Amazon Web Services released Rekognition, a service that lets users analyze digital imagery in order to identify objects, including faces, based on a machine-learning algorithm.
The company has touted uses for this product that include identifying triggering content in images, determining emotions in an image, cashier-less grocery stores (which they have implemented in Amazon Go) — and, most worrisome, “public safety.”
The implications of this technology are eerily dystopian. The software is designed to identify up to 100 faces in an image or video, and can identify the emotions and actions of subjects. In the hands of the police, Immigration and Customs Enforcement (ICE) or the National Security Agency (NSA), this poses an existential risk to privacy, oppressed groups, left organizations and any other targeted person.
And the unfortunate truth is that it is already being licensed to police departments. As an ACLU report reveals, police departments in Orlando, Florida, and Washington County, Oregon, are already using it. If legal restrictions are overcome, this could easily be licensed to a company like Axon to be used in police body cameras in numerous cities.
Even scarier, consider if ICE were able to license this technology. It could use it to hunt down any undocumented immigrant it has a photo of. Even worse, it could cross-list any photo to which it has access against a database of identification images — and mark anyone who isn’t in its system for tracking and detention.
The ACLU only listed agencies to which it submitted public records requests. Amazon Web Services (AWS) has specially tailored contracts for federal agencies, including the Department of Homeland Security, the IRS, the US Air Force, the US Navy, and many others.
The truth is, we don’t know which of these groups might be using Rekognition. It took a year from when Orlando’s police department began using Rekognition before the ACLU found out through a public-records request.
This doesn’t even account for agencies like the CIA and NSA, which presumably would remain unreported if they were to use AWS technologies.
Given the NSA’s $10.8 billion budget in 2013 (its usually confidential budget was leaked for the year of 2013) and its primary function of surveillance, we might expect that it has either contracted Rekognition technology or built its own even better surveillance algorithms. And we might expect the deployment of similar technology by the CIA, with its leaked 2013 budget of $14.7 billion.
Where does this leave us? Are we doomed to a world dominated by a ubiquitous surveillance state that until a few years ago was beyond our wildest fears? Is privacy dead?
Machine learning has gone through astronomical growth over the last few years. From Google’s self-driving cars to medical research, machine learning has the potential for spectacular good. And while it remains largely a buzzword in society — to the extent it is even discussed — tech companies large and small are investing billions in it.
Interviewing people in the tech community, many tech workers were convinced that this technology is inevitable. If Amazon, Google, Facebook and others weren’t pushing it forward, other companies or the NSA would be. But this doesn’t justify the resignation that society will inevitably have indiscriminate surveillance.
Under capitalism, the government relies on the military-industrial complex to refine the means to exploit people at home and abroad. While the government would presumably continue to work on this technology in any case, contracting the work to private-sector corporations helps to secure the support of the richest people in our society, who ultimately have massive political influence over the government because of their wealth.
Therefore, examples such as Google employees’ victory in forcing the company not to renew its Maven contract with the defense department for drone technology should be viewed as a substantial victory. Similarly, boycott, divestment and sanction (BDS) movements against Israel (and similar historical movements against South Africa) have the potential to destabilize oppressive states.
But this can be taken a step further. Military and security organizations don’t exist in a bubble. They depend on hundreds of thousands of workers. From electricians who wire up CIA buildings to coders who build surveillance software, we find laborers capable of literally shutting down the system when organized collectively.
And we’ve also seen individual attempts from the inside to reveal and destabilize the state’s undemocratic surveillance by the likes of Edward Snowden. While these actions are admirable and at least brought the issue to public light, they haven’t measurably slowed the trend towards increasing surveillance.
All this provides some insight into how workers at Amazon could respond to Rekognition.
Beyond achieving the temporary victory of stopping this one corporation’s licensing of surveillance technology (and thereby slowing down its spread), such struggles broaden awareness of the extent to which this technology is being applied and help to develop a working-class consciousness more aware of its own power in fighting back.
In doing so, tech workers’ struggles can contribute to a culture that is safer for government workers to fight back (just as protests in the US against the war in Vietnam preceded soldiers’ resistance against officers in Vietnam).
Unfortunately, the tech sector has scant history of labor organizing, which has only begun to change recently. Earlier this year, tech workers at Lanetix surprised many by voting to unionize. The company responded by firing its entire engineering staff. At other tech companies, this idea of unionizing is only a distant hope of labor organizers.
Part of the difficulty in organizing tech workers lies in the fact that most of society and a sizeable portion of the left do not view engineers as part of the working class.
But like the rest of the working class, tech workers sell their labor for a wage, and in exchange, produce a commodity, which their employers sells for a profit — i.e., more than the employer had to lay out for wages, raw materials and other costs — in contrast to those who receive a salary for managing others or who passively acquire income due to their existing ownership of wealth.
As in other industries, we observe that technology firms constantly seek to extract as much labor from employees as possible (by getting workers to put in as many hours as possible as efficiently as possible) while making costs as low as possible (by pushing down wages, benefits and working conditions).
While the level of education required and the scarcity of those seeking tech work means firms are compelled to offer higher salaries, tech workers are, like other workers, exploited in the production process.
There’s also a drive to de-skill tech work — just as has happened in other industries. Examples of this de-skilling include the simplification of coding languages, the standardization of coding practices, and the automation of software maintenance.
Although software coders may earn higher wages than many other workers, they still have the same relationship to the means of production, which implies a shared interest in fighting exploitation that will only become more pronounced as software production becomes further deskilled over time.
Thus, it’s reasonable to think that tech workers, given their incentive to fight back against their exploitation under capitalism, might even embrace the strike weapon. At this time, though, the strike seems distant from the current consciousness in tech.
Furthermore, generating labor struggle in response to Rekognition presents the added challenge of relating to a broader working-class and social concern, rather than a workplace issue. Broader community organizing, therefore, seems essential in sparking opposition to Amazon’s licensing of Rekognition.
Fortunately, this has already begun — witness the press conference about Rekognition outside of Amazon led by ACLU of Washington, Council on American-Islamic Relations WA, and several others. Community organizing in Washington County, Oregon, and Orlando against their police departments’ use of this technology could add helpful pressure.
Despite these challenges, some aspects of tech offer unique organizing advantages. Many tech companies originated in an economic situation that promised high profit margins, given the newness of products and the automation of services done via manual labor. These high profit margins implied a flexibility that allowed companies to achieve immense success without the bad PR that accompanies cutting corners.
This is emphasized by Google’s moto “Don’t be evil,” which it not-so-coincidentally abandoned in 2016. The idea of making lots of money while making the world a better place fused the counter-culture of the 1960s with a frontier business ethos that has been labeled “the Californian Ideology.”
However, as competition between tech companies increases and their profit margins decrease, the contradictions of this philosophy have come to the surface. Companies are being forced to choose between profits and morals, a choice that unsurprisingly favors the former.
Google’s don’t-be-evil mantra, though formally abandoned, was still enough to give employees leverage to challenge the military drone development in Project Maven. Google workers organized thousands of employees to express their anger at the project, and a dozen employees resigned over the issue.
In doing so, they were able to create enough bad PR for Google to force it to come to terms with this contradiction. Ultimately, Google had to cancel the drone contract. While this PR may seem insignificant — it seems unlikely that people are going to stop using Google search based on this — Google’s management has other concerns: in particular, employee recruitment and retention.
Many tech workers have invested themselves in the idea that their work should make the world a better place, and Google worried about losing talent if they didn’t address the outcry.
Analyzing the value of Google’s brand from an employee recruitment perspective is difficult to measure, but given that the Maven contract only promised Google $9 million per year, it is no surprise that they chose to abandon it rather than risk losing more employees over this issue.
The prospect for Amazon employees to scuttle Rekognition may not be as rosy. Google promoted itself as an ethical corporation; Amazon has never held itself up to such standards.
It has had its share of negative media coverage — from the New York Times coverage of its cutthroat corporate culture in corporate jobs to poor working conditions at Amazon warehouses. Amazon has similarly demonstrated steadfast resistance to union drives.
Related to this is the fact that a majority of Amazon’s 566,000 employees are not in corporate or software jobs, but instead working in warehouses and logistics. So the subset of Amazon employees who bought into this “Californian Ideology” is smaller than at Google. Lastly, in the retail market, the likes of Walmart set a low bar for Amazon to match in terms of labor standards.
Another source of added difficulty at Amazon is the sheer value of surveillance contracts with government agencies. The Rekognition website details prices ranging from $.0004 to $.001 per image processed (it is cheaper for larger contracts), and $.10 per minute of video processed for facial recognition purposes.
Considering the billions of images and billions of minutes of video that governments might eventually license for, the numbers quickly add up. Remembering the NSA’s multibillion-dollar budget, surveillance is big business. Forcing Amazon to cancel these contracts will require far more than it took to get Google to cancel their $9 million-per-year drone contract.
Finally, there’s the issue that Rekognition still isn’t familiar to many at Amazon. When I asked Amazon workers what they thought of Amazon’s licensing of Rekognition to police, the majority of Amazon employees said they had not heard of the service.
In an effort to spread knowledge of Rekognition and its risks, while simultaneously putting pressure on Amazon, employees began internally circulating a letter to Jeff Bezos modeled on the Google employees’ letter about Project Maven.
It remains to be seen how much traction this letter might get, but the fact that it’s now public knowledge that workers at Amazon have begun such a discussion is an encouraging sign about the future prospects for resistance by tech workers.