Wow: Anatomy Of An Ai Organisation (Everything That Goes Into Amazon's Echo)

This is a stunning slice of research.
note: this version of the map is via the Verge, HT upwardly fron to Professor Crawford's twitter feed which points upwardly a giant wall-sized re-create at the V too A.

From NYU's Kate Crawford:

The Amazon Echo equally an anatomical map of human labor, information too planetary resources

By Kate Crawford ¹ too Vladan Joler ²
(2018)

Download map inwards pdf format here

I

H5N1 cylinder sits inwards a room. It is impassive, smooth, uncomplicated too small. It stands 14.8cm high, with a unmarried blue-green round down lite that traces about its upper rim. It is silently attending. H5N1 adult woman walks into the room, carrying a sleeping youngster inwards her arms, too she addresses the cylinder.
‘Alexa, plow on the hall lights’

The cylinder springs into life. ‘OK.’ The room lights up. The adult woman makes a faint nodding gesture, too carries the youngster upstairs.

This is an interaction with Amazon’s Echo device. ³ H5N1 brief command too a response is the most mutual shape of appointment with this consumer voice-enabled AI device. But inwards this fleeting 2nd of interaction, a vast matrix of capacities is invoked: interlaced chains of resources extraction, human project too algorithmic processing across networks of mining, logistics, distribution, prediction too optimization. The scale of this organization is almost beyond human imagining. How tin give the sack nosotros laid out to encounter it, to grasp its immensity too complexity equally a connected form? We start with an outline: an exploded persuasion of a planetary organization across iii stages of birth, life too death, accompanied past times an essay inwards 21 parts. Together, this becomes an anatomical map of a unmarried AI system.

Amazon Echo Dot (schematics)

II

The scene of the adult woman talking to Alexa is drawn from a 2017 promotional video advertising the latest version of the Amazon Echo. The video begins, “Say howdy to the all-new Echo” too explains that the Echo volition connect to Alexa (the artificial tidings agent) inwards gild to “play music, telephone yell upwardly friends too family, command smart abode devices, too more.” The device contains 7 directional microphones, thus the user tin give the sack last heard at all times fifty-fifty when music is playing. The device comes inwards several styles, such equally gunmetal grayness or a basic beige, designed to either “blend inwards or stand upwardly out.” But fifty-fifty the shiny pattern options keep a sort of blankness: zip volition alarm the possessor to the vast network that subtends too drives its interactive capacities. The promotional video just states that the attain of things you lot tin give the sack inquire Alexa to do is ever expanding. “Because Alexa is inwards the cloud, she is ever getting smarter too adding novel features.”  

How does this happen? Alexa is a disembodied vocalism that represents the human-AI interaction interface for an extraordinarily complex gear upwardly of information processing layers. These layers are fed past times constant tides: the flows of human voices existence translated into text questions, which are used to query databases of potential answers, too the corresponding ebb of Alexa’s replies. For each response that Alexa gives, its effectiveness is inferred past times what happens next:
Is the same query uttered again? (Did the user experience heard?)

Was the query reworded? (Did the user experience the query was understood?)

Was in that location an activity next the question? (Did the interaction outcome inwards a tracked response: a lite turned on, a production purchased, a rail played?)
With each interaction, Alexa is preparation to listen better, to interpret to a greater extent than precisely, to trigger actions that map to the user’s commands to a greater extent than accurately, too to construct a to a greater extent than consummate model of their preferences, habits too desires. What is required to brand this possible? Put simply: each pocket-size 2nd of convenience – last it answering a question, turning on a light, or playing a vocal – requires a vast planetary network, fueled past times the extraction of non-renewable materials, labor, too data. The scale of resources required is many magnitudes greater than the unloose energy too project it would guide hold a human to operate a household appliance or motion painting a switch. H5N1 total accounting for these costs is almost impossible, but it is increasingly of import that nosotros grasp the scale too compass if nosotros are to empathize too principle the technical infrastructures that thread through our lives. 

III

The Salar, the world's largest apartment surface, is located inwards southwest Republic of Bolivia at an height of 3,656 meters inwards a higher house bounding main level. It is a high plateau, covered past times a few meters of tabular array salt crust which are exceptionally rich inwards lithium, containing 50% to 70% of the world's lithium reserves. ⁴ The Salar, amongst the neighboring Atacama regions inwards Republic of Chile too Argentina, are major sites for lithium extraction. This soft, silvery metallic is currently used to ability mobile connected devices, equally a crucial cloth used for the production of lithium-Ion batteries. It is known equally ‘grey gold.’ Smartphone batteries, for example, normally guide hold less than 8 grams of this material. ⁵ Each Tesla motorcar needs some 7 kilograms of lithium for its battery pack. ⁶ All these batteries guide hold a express lifespan, too 1 time consumed they are thrown away equally waste. Amazon reminds users that they cannot opened upwardly up too repair their Echo, because this volition void the warranty. The Amazon Echo is wall-powered, too also has a mobile battery base. This also has a express lifespan too and thus must last thrown away equally waste.

According to the Aymara legends nearly the creation of Bolivia, the volcanic mountains of the Andean plateau were creations of tragedy. ⁷ Long ago, when the volcanos were last too roaming the plains freely, Tunupa - the solely woman someone volcano – gave nascency to a baby. Stricken past times jealousy, the virile someone volcanos stole her infant too banished it to a distant location. The gods punished the volcanos past times pinning them all to the Earth. Grieving for the youngster that she could no longer reach, Tunupa wept deeply. Her tears too thorax milk combined to create a giant tabular array salt lake: Salar de Uyuni. As Liam Young too Kate Davies observe, “your smart-phone runs on the tears too thorax milk of a volcano. This landscape is connected to everywhere on the planet via the phones inwards our pockets; linked to each of us past times invisible threads of commerce, science, politics too power.” ⁸

IV

Our exploded persuasion diagram combines too visualizes iii central, extractive processes that are required to run a large-scale artificial tidings system: cloth resources, human labor, too data. We consider these iii elements across fourth dimension – represented equally a visual description of the birth, life too decease of a unmarried Amazon Echo unit. It’s necessary to displace beyond a uncomplicated analysis of the human relationship betwixt an private human, their data, too whatever unmarried engineering scientific discipline companionship inwards gild to debate with with the really planetary scale of extraction. Vincent Mosco has shown how the ethereal metaphor of ‘the cloud’ for offsite information management too processing is inwards consummate contradiction with the physical realities of the extraction of minerals from the Earth’s crust too dispossession of human populations that sustain its existence. ⁹ Sandro Mezzadra too Brett Nielson locomote the term ‘extractivism’ to hollo the human relationship betwixt dissimilar forms of extractive operations inwards contemporary capitalism, which nosotros encounter repeated inwards the context of the AI industry. ¹⁰ There are deep interconnections betwixt the literal hollowing out of the materials of the footing too biosphere, too the information capture too monetization of human practices of communication too sociality inwards AI. Mezzadra too Nielson banking concern notation that project is key to this extractive relationship, which has repeated throughout history: from the means European imperialism used slave labor, to the forced piece of job crews on condom plantations inwards Malaya, to the Indigenous people of Republic of Bolivia existence driven to extract the silvery that was used inwards the kickoff global currency. Thinking nearly extraction requires thinking nearly labor, resources, too information together. This presents a challenge to critical too pop understandings of artificial intelligence: it is difficult to ‘see’ whatever of these processes individually, permit lonely collectively. Hence the ask for a visualization that tin give the sack convey these connected, but globally dispersed processes into a unmarried map.

V

If you lot read our map from left to right, the story begins too ends with the Earth, too the geological processes of deep time. But read from meridian to bottom, nosotros encounter the story equally it begins too ends with a human. The meridian is the human agent, querying the Echo, too supplying Amazon with the valuable preparation information of verbal questions too responses that they tin give the sack locomote to farther refine their voice-enabled AI systems. At the bottom of the map is some other sort of human resource: the history of human noesis too capacity, which is also used to prepare too optimize artificial tidings systems. This is a key departure betwixt artificial tidings systems too other forms of consumer technology: they rely on the ingestion, analysis too optimization of vast amounts of human generated images, texts too videos.

...MUCH MORE (an incredible amount of piece of job went into this)