The hidden costs of Amazon’s Echo

Schematics for an Amazon Echo Dot (Image credit: Anatomy of an AI System)

What’s the first thing that comes to mind when you think of the Amazon Echo?  

Let me guess: Artisanal cobalt mining in the Congo, shipping containers, the Greek chimera, click-workers workers tagging and labeling data, villagers digging for tin on the Indonesian island of Bangka, Jeff Bezos, and the breast milk of a volcano?

Well, except for Jeff Bezos, maybe not.

But all of these elements and more factor into the Anatomy of an AI System, an essay and map by Kate Crawford and Vladan Joler designed to make people think about the “vast planetary network” that undergirds the sleek, powerfully magical smart speakers. The magic, of course, comes at a cost: Every mundane interaction with the AI-driven device — Alexa, what time is my yoga class?  Alexa, open the garage! — sucks up data, labor, and material resources from around the globe.

We’re used to thinking about the environmental and labor costs of industries like oil, but AI devices and other cloud-based technologies tend to get a pass. But the scope and scale of the supply chain needed to power AI is even vaster. “The production process of one contemporary device could have 100s of different suppliers and then those hundreds of suppliers can have hundreds of sub- sub- sub-suppliers,” Joler told All Turtles. “It is really, really hard to investigate.”  

Another notable difference between AI and previous technologies, as the authors see it, is that AI requires human fuel — “the ingestion, analysis and optimization of vast amounts ofhuman-generatedd images, texts and videos” to train and optimize its systems.

Crawford, co-founder of AI Now, and Joler, professor of new media at the University of Novi Sad, met a year ago at a Mozilla retreat in Italy where they hatched the idea “of mapping the birth, life and death of one AI device,” Joler said. Both researchers have done work on algorithmic black boxes but they wanted to take on the bigger, messier job of trying to map out the invisible infrastructures and processes of AI.  

They chose the Amazon Echo because the device’s streamlined voice-only interface has the effect of rendering the actual workings of its system especially opaque, but they could have analyzed the iPhone or many other AI-driven devices to make the same point.  “The object itself is a very simple extrusion of plastic representing a collection of sensors,” the authors write about the Echo, “it’s real power and complexity lies somewhere else, far out of sight.”

One of the main stories Crawford and Joler want to tell is about the many “extractive processes” that make AI work. Every part of an AI system, from the GPUs to plasma screens to fiber optics to the attractive black plastic cylindrical Echo sitting on your counter, contains elements mined from the earth. Which brings us to that Indonesian island, Bangka. There, workers earn just $7 a day to dredge with bamboo poles for the tin ore used in solder in circuit boards and smartphone transistors and other electronic products. Tin mining has devastated the island’s ecology.

Rare earth metals, which have special optical and magnetic characteristics, are also key ingredients in a vast array of high tech products. Almost all of the metals come from mines in China, which emit massive amounts of CO2 and toxic tailings, acidic water and other pollution.  

“We are generally not aware of the material aspect of technology,” Joler said. “I have this picture in my head that every time I send an email or give Alexa a command, a little hole gets made in the ground.”

Extraction also has an anthropological dimension, as with lithium from the high-altitude salt flats in Bolivia. For the Aymara people, the salt came from a female volcano mourning for her lost child. The authors cite researchers who conclude, therefore, that “your smartphone runs on the tears and breast milk of a volcano.”

Not only is resource extraction surging to meet the demands of the AI network, the products produced become obsolete at an ever faster rate. “From a slow process of elemental development, these elements and materials go through an extraordinarily rapid period of excavation, smelting, mixing, and logistical transport – crossing thousands of kilometers in their transformation,” the authors write. “We are extracting Earth’s history to serve a split second of technological time, in order to build devices that are often designed to be used for no more that a few years.”

The authors also invite us to consider the full array of labor that goes into an AI system — from indentured labor in mines to factory workers making hardware, to low-paid “cognitive workers” labeling training sets, to villagers “recycling” electronic waste. People at the top of the labor pyramid, such as Jeff Bezos, extract value from labor downstream.  In one stark example, the children in the Congo who mine cobalt for lithium batteries make about $1 day to work in hazardous conditions. “In contrast, CEO Jeff Bezos … made an average of $275 million a day,” the authors write, citing an Amnesty International study.

This photo shows Workers dismantle reusable parts from electronic waste at Seelampur in Delhi, India. (Image credit: Getty Images)

Workers dismantle reusable parts from electronic waste at Seelampur in Delhi, India. (Image credit: Getty Images)

Other labor inputs are obscured from view in different ways, such as the armies of people doing the low-paid digital piecework of tagging data to feed the neural networks, the unseen Mechanical Turks still needed to help make the magic happen.   

Users of the Echo and other devices, the authors say, have a multidimensional relationship to the technology.  The authors evoke the Greek chimera, part lion, goat, snake and monster, as a metaphor for the Echo user, who is “a resource, a worker, and a product.”  We purchase the device, but our voice commands are also a resource to train neural networks. As we contribute feedback regarding Alexa’s replies, we also contribute free labor.  “There is a lot of hidden labor in these artificial intelligence systems,” Joler said.

Just like algorithmic black boxes, fully exposing and explaining the infinitely complex supply chain of an AI system is almost impossible.  Crawford and Joler’s map isn’t meant to be definitive — but they do hope to push the transparency conversation into new territory, especially “to consider the relationship between society, technology, and nature,” Joler said.

The dimensions of this relationship are more out in the open in some places than others.  On a recent trip to India, Joler saw families trying to survive by recycling toxic waste from electronics, companies that employ “click-workers,” and even a container ship graveyard.  “In India, you can see a lot of different layers of the stack,” Joler said.

“Usually we see the polished devices, clean interfaces or some beautiful San Francisco office,” he added.  “But as you move away from centers like Silicon Valley you see different realities. Those different realities are part of this map.”

Top image: Schematics for an Amazon Echo Dot (Image credit: Anatomy of an AI System)