Every computer is built out of rocks, sand, and captured sunlight, refined in some capacity and made to turn lightning into a storehouse of knowledge. That this is a gross reduction of everything from the development of the petroleum industry, the science of circuity, and the electrification of the world does not make it less true. We are comfortable with the electrified-rocks-that-hold-information because we have built a vocabulary for it.
There is a struggle, specifically within the Pentagon, to harness a similar vocabulary for data, and especially the processes by which it will turn that data into algorithmic gruel powering autonomous machines.
I’m Kelsey D. Atherton, reporting from Socorro, New Mexico, and I’m going to talk about a metaphorical problem.
The paucity of language for dealing with data was on display at the Association of the United States Army AI and Autonomy Symposium, held in Detroit November 20-21, 2019. A panel premised around the question of “Is Data the New Oil?” led an array of hosts to state, affirmatively, that while data is a resource, it is unlike oil.
What data is like, however, is a trickier hurdle.
Is data like ice, where a cube melts alone but a bucket stays cool the whole night? That metaphor conveyed the idea of datasets as reinforcing and better in combination, but it fell short of approximating anything of the utility of data, which is all in how its used, and very little in being stockpiled and left alone.
Is data like ore? Harvested raw in an aggregated mass, ore has to be broken down into component parts before it is clear what nuggets of value was contained next to what rubbish. The metaphor captures that harvested data needs refining to be valuable, but it still treats data as a sort of ambient product of nature that humans stumble upon, rather than the active project of a growing array of sensors and tools used across all levels of human functioning.
Data is none of the easy metaphors because data is a created thing, and the data the Pentagon will get in the future depends a great deal on the sensors the Pentagon deploys today and the collection processes it chooses to adopt for those sensors. The biggest obstacles to useful data, from contamination to malicious manipulation to simple deletion or complex classification regimes, are all human processes, highly malleable and subject to change.
If the Pentagon is going to get a grasp on the data it holds and hopes to use in the future, it will need to move away from metaphors plucked out of keggers or board game nights. Building a functional vocabulary for a complex and abstract function may sound daunting, but it’s an essential part of how the military functions. Without clear, useful language to understand data, the Pentagon risks a future of managing AI where data scientists have to convert every algorithmic regression to a football metaphor first.
And it’s not like it hasn’t been done before. This is, after all, the part of society that formalized rules and procedures for how to make a fast rock break bone, and called it all “kinetic effects.”