Anatomy of an AI System: The Amazon Echo as an anatomical map of human labor, data and planetary resources

Anatomy of an AI System: The Amazon Echo as an anatomical map of human labor, data and planetary resources is a 2018 essay and large-scale infographic map by Kate Crawford (AI Now Institute, Microsoft Research New York) and Vladan Joler (SHARE Lab, University of Novi Sad), jointly published by the AI Now Institute at NYU and SHARE Lab in September 2018. The work takes a single Amazon Echo device as its anatomical subject and traces the full supply chain of minerals, labour, data, and energy on which it depends — rendering visible the planetary and human infrastructure that commercial AI systems characteristically obscure behind the convenience surface.

The work is dual-form: a large-scale map designed for high-resolution printing (approximately four feet by eight feet), and an analytical essay that develops the argument the map images. The two registers are deliberately distinct — the map performs the systemic, cross-sector scale of the argument in a single image; the essay grounds each layer in specific supply-chain relationships and analytical claims about visibility and power.

Content and argument

The map's analytical structure traces three overlapping supply chains. The materials chain runs from the device's physical components backward through smelters and refineries to extraction sites: coltan from the Democratic Republic of Congo (feeding tantalum for capacitors), lithium from Chilean salt flats, cobalt from Congolese and Indonesian mines, neodymium from Inner Mongolian rare-earth processing, and bauxite for aluminium — each sourced through supply chains the essay documents as entangled with precarious and informal labour, environmental damage, and conditions the device's marketing image actively suppresses. The labour chain traces the human work embedded in the device and its supporting infrastructure: Amazon warehouse workers who pack and ship devices, Mechanical Turk micro-task workers whose labelled training data teaches the speech-recognition model, content moderation and data annotation workers who curate responses, home users whose voice data is retained and used to improve the system, and the device's disposal pathway through electronic-waste processing chains in lower-income countries. The data chain maps how the device's every interaction produces behavioural, temporal, and locational data that Amazon retains and models — a structure in which the purchase of a device also constitutes agreement to become a node in continuous data-collection, a form of labour the labour contract does not name.

The accompanying essay extends the supply-chain tracing into an argument about visibility and power: AI systems are structured to appear as ambient, convenient services while the extraction regimes, labour conditions, and surveillance architectures on which they depend are hidden from users and from the regulatory frameworks that nominally govern them. Crawford and Joler propose the anatomical-map form as a genre capable of holding the systemic, cross-sector, multi-layered nature of AI's material dependencies in a single analytical image — one that case studies of specific harms or technical audits cannot reproduce.

Reception and institutionalisation

The work received the Beazley Design of the Year 2019 at the Design Museum London — the award recognising the year's most innovative design across six categories (architecture, digital, fashion, graphics, product, transport) — placing a research-and-criticism artefact among entrants evaluated in design terms. The map was subsequently acquired by MoMA and entered into the permanent collections of the V&A, the Ars Electronica Center, and the Design Museum London, institutionalising an academic-origin AI-critique artefact in the international museum system at an unusual scale for the genre.

Joler produced the essay through SHARE Lab, the data investigation arm of SHARE Foundation that he founded; the collaboration connects the Serbian digital-rights field's investigative practice (SHARE Lab's program of studies into invisible digital infrastructures) with the AI Now Institute's academic-critique programme at NYU.

Place in the corpus

Anatomy of an AI System is the visual precursor to Atlas of AI (2021) — the map works out in infographic form the same supply-chain-and-extraction argument that the book subsequently develops as a full-length monograph with field-site case studies. Karen Hao's 2021 MIT Technology Review review of Atlas of AI names the essay explicitly as "the visual-essay precursor the book formalises into a single-volume argument." Within the Crawford-Joler collaborative arc, the essay is the first of three large-scale works mapping technology and power — followed by Training Humans with Trevor Paglen (2019) and Calculating Empires: A Genealogy of Technology and Power since 1500 with Joler (2023–2025), which won the Silver Lion at the Venice Architecture Biennale.

Where the corpus's other foundational algorithmic-accountability publications — Weapons of Math Destruction, Automating Inequality, Algorithms of Oppression — analyse harmful outcomes produced by deployed AI in specific sectors and populations, Anatomy of an AI System shifts the analytical register upstream: the subject is the hidden substrate — minerals, workers, data — that makes the system possible, not the discriminatory outcome the system produces. That upstream frame is what the material-AI-critique register of the movement — and Atlas of AI — subsequently carries into policy, journalism, and organising.