A Cloud Is Not Just a Cloud

Today, a journalist asked me, “Clearly ICE is doing terrible things, but is providing cloud services to an organisation that does terrible things also terrible?”

If ICE and DHS want to use Bluehost, my webhost that just told me it won’t let me run Ruby on Rails today, they can go ahead and try. But Microsoft is enabling much more than storage and computational cycles like the sort that would run on my laptop or even my own server at UCSD. The Microsoft Azure blog boasted of the cloud architecture’s capacity to support facial recognition tasks (source). Cloud architecture that supports algorithms rapidly jumping across streaming or stored images, real time from cameras all over the world, needs to have state of the art data processing capabilities — that means cutting edge database, disk access algorithms, custom chip designs. A cloud is never simply a cloud.

Google attempted this Clouds-are-General-Purpose argument with Project Maven. They said that the Pentagon would just be using open-source software, which meant the military could still carry out the work without Google’s help. Google employees, the ones who knew “the cloud” and its APIs intimately didn’t agree. Google has even gone so far as to design custom microprocessors — TPUs, or Tensor Processing Units — to speed up “inference” to keep up with real time image, sound, and video flows (source).

Similarly, Microsoft Azure Government dedicates its engineers to developing AI and systems architectures to suit major clients — and clients don’t get much bigger than the government or military. Microsoft’s cloud services have APIs, or application program interfaces, that do face detection and identification, as well as real time translation of voice audio. As a Senior Program Manager at Microsoft put it at an April meetup for government employees and contractors, “With the translation stuff, we’re talking about use cases like real time interviews, border patrol, immigration. I mean, a lot of this they sky’s the limit with your imagination. We’re seeing a lot of interesting scenarios coming up.” (source)

If the government and the military didn’t have access to Google, Amazon, and Microsoft clouds, they would have to invest designing a whole stack of computer systems, from databases, to chips, to storage architecture, on their own. Denying the military and militarized police and border patrol the corporate cloud can forestall automated surveillance and targeting for years. For some, this is precisely why they want the military to have access to private sector innovation. But there are many, including service members I have met who have seen our operations first hand, who are appalled at the destruction wrought by the militarization of our foreign policy, our borders, and our police. Engineers, product managers, and data workers are also citizens and have a right to ethical and moral judgement about how they want to contribute to these processes.

So the cloud is not just a cloud. Clouds are complex infrastructures engineered and optimized for some social needs and not others. We are throwing social resources into engineering clouds for security, for cutting people off of welfare rolls, and for making examples of communities fleeing war.

Microsoft’s, Google’s, and Amazon’s rush to provide cloud translation and face recognition for government, including border patrol and spy agencies (source), alerts us to how society’s computing skills have been directed to technologies of surveillance and control rather than human flourishing.

As algorithms and the cloud services that run them have entered into government work, researchers demonstrate how automated bureacuratic judgement becomes even less accountable to oversight, audit, and appeal. Virgnia Eubanks, for example, shows how automated welfare decisions remove power from social workers and place them in privately-owned algorithms and call centers, removing them from public oversight. Safiya Noble’s Algorithms of Oppression (especially Chapter 4) shows us what happens when citizens lose the right to have their pasts forgotten once they are stored in corporate servers, possibly informing the algorithms by which everyday people are judged. ICE’s use of Microsoft Azure and the military’s use of AI drone targeting is just the tip of the iceberg.