Kate Crawford is a leading researcher, academic and author who has spent the past decade researching data systems, machine learning and the impact of artificial intelligence on society. She is a Distinguished Research Professor at New York University, a principal investigator at Microsoft Research in New York, and a visiting professor at the MIT Media Lab. Her recent publications cover data bias and fairness, the social impact of ARTIFICIAL intelligence, predictive analytics and due process, and algorithmic accountability and transparency.

The AI Now Institute led by Kate recently published a research report using Amazon Echo as a starting point for analysis, trying to make readers understand the complex knowledge content behind smart products, from mining the materials used to make these smart devices to collecting massive amounts of data to train intelligent dialogue models inside devices.

You’ve probably experienced this in your own life: a round object sits in a corner of a room, small and fluid in shape, and if no one speaks to it, it sits in the corner, indifferent to its surroundings. At that moment, the woman of the room, with a sleeping child in her arms, said to the cylinder, “Alexa, turn on the hall lights,” to which the cylinder replied, “Yes.” Less than a second later, the room lit up. With a slight nod, the hostess led the child upstairs. This is how users interact with Amazon Echo smart devices. Such short commands and quick responses are currently the most common interaction scenarios between smart speaker consumers and voice-enabled AI devices. However, in order to complete the interaction that takes place in this instant, we need a vast matrix of knowledge: interlocking chains of resource extraction, manual and algorithmic processing across mining, logistics, distribution, prediction and optimization across the entire industrial network. The scale of the system is almost beyond human imagination. How can we begin to understand it as an infinitely connected network structure? Let’s start with an anatomy of a single AI system.

"We need to have a deep and complex conversation about building AI at scale." We are all vaguely, uncomfortably aware that our lives are caught up in systems we don't fully understand. The last meal you eat may contain produce from another distant country that has been harvested, processed, packaged, shipped and sold to your table. The phone in our hands is the end product of a more complex chain: from a man in an African mine to an assembly line in China to a designer in an office in San Francisco.Copy the code

Explaining how these systems are connected and their impact on the world is no easy task.

The main artwork in a recent exhibition by Kate Crawford and Professors Vratanjoler is a giant map, 2m high and 5m wide, dedicated to an ARTIFICIAL intelligence gadget that tracks one of the most complex products of modern times: the Amazon Echo.

It’s a jumble of black and white branching lines that looks more like a schematic of a nuclear reactor than an everyday gadget. The panorama is called the Anatomy of ai Systems, but its subtitle explains the scope of its discussion: “The Anatomy of amazon Echo as a product of the combination of human labor, knowledge and data and planetary resources.”

The first part of “Anatomy” shows how Amazon Echo collects data and feedback from human users

In advance of The report's release, The Verge spoke with NYU professor Kate Crawford, co-founder of AI Now, an AI institute that studies The social impact of ARTIFICIAL intelligence. Crawford and her collaborator, Johler, a professor at the University of Novi Sad's School of Art, say the anatomy was created because of a lack of understanding of the structure of modern gadgets, especially artificial intelligence.Copy the code

Founded by Kate Crawford, the AI Now Institute is a research organization testing the impact of ARTIFICIAL intelligence on society

“We need to have a deep and complex conversation about the implications of building AI on a large scale,” Crawford said. “Through ‘anatomy,’ you can see it and begin to understand it as part of something bigger.”

(TheVerge :) first, why was Amazon Echo chosen as the focus of the project? Kate Crawford: I'm really interested in the simple voice-based interactions in these systems. Sitting in your house, Echo looks very simple and compact, but it has big roots and can connect to huge production systems: logistics, mining, data acquisition and AI network training. This is an entire infrastructure stack like you've never seen before. All you need is a simple voice command -- "Alexa, turn on the lights" and it turns the lights on like magic. But trying to investigate and understand how this magic works is the point of the project. From this point of view, Echo is powerful because it provides users with this sense of convenience, but when you take the lid off, you can see the huge cost involved in producing it. Some would argue that technology has always been like this. When you talk about the huge deforestation that occurred in the 19th century because of the need to harvest natural latex to insulate and wrap undersea cables, how is that any different from today?Copy the code

In the 19th century British ships laid cables at the surface of the Atlantic ocean

We've had a lot of technology booms before, while extracting resources to make it happen. This is certainly a trend. But I would say the shift to ARTIFICIAL intelligence is a very, very important reason. First, the level at which it operates is starting to change the way society itself operates, because AI systems are being built in the institutions that matter most to us, from health care to criminal justice, and these systems are really changing the way everyone interacts with the world.Copy the code

Echo is built with a variety of minerals, including lithium harvested from Bolivia’s Salar de Uyuni. Photograph: Dean Mouhtaropoulos/Getty Images

TheVerge: Anatomy itself is divided into three broad systems, each called an "extraction process." One of the material resource extraction process, divided into data and artificial two. Why do you think it is useful to build these systems in such a way as "extract"?Copy the code

Crawford: All of these processes extract value in different ways. When you think of coal mining, for example, you might think of an industry that drives rampant growth, high profits, but that also generates costs that are initially ignored and uncounted in the economic system. The true picture of resource extraction may take decades to emerge. Does data mining have similarly unknown costs beyond our current economic framework?

"Ai systems are extracting residual value from all kinds of human activities." The concept of an algorithmically black box is now well known, thanks to the important work of academics like author Frank Pasquale. Our project is interested in how to communicate with other kinds of black boxes. Echo itself is a very difficult box to examine: users can't see how it works, how it records data, or how it trains its algorithms. Then there is the hidden logistics around how to collect and smelt and assemble the simple components within it, through multiple layers of contractors, distributors and downstream component manufacturers. In the article, Crawford wrote an example of how Intel had been able to get a good understanding of its supply chain over the years to ensure that its microprocessors were free of the Congolese metal tantalum. Imagine a well-resourced company with highly skilled staff, good record keeping and databases, and it takes years to understand its purchasing patterns! This shows how difficult these processes can be to investigate and analyze inside companies, let alone researchers and journalists working outside. But the process of telling the story of production is important, and needs to be: how we begin to see the dizzying complexity of global production of technological goods. In the article, Crawford also talks about the complex chain structure of the world's resource allocation, which will lead to another shift in the way wealth is accumulated. There are billionaires at the top of the system who extract maximum value, and the further you go down the logistics and production chain, the closer you get to raw materials, the bigger the gap becomes. For example, Amazon CEO Jeff Bezos, through a combination of machine learning algorithms and modern e-commerce, saw amazon's sales, profits and stock price soar in 2018, rising 270% in three years and 103% in the past 12 months. Now amazon is closing in on Apple as the world's most valuable company. Bezos has a net worth of nearly $160 billion, making him by far the richest person on the planet.Copy the code

CEO Jeff Bezos earned an average of $275 million a day in the first five months of 2018. That’s obviously a lot, but you can really understand the gap when you contrast it with workers farther downstream.

Amnesty International released an investigation into child Labour in Congo mines, in which cobalt was traced to lithium ion batteries. In this case, one of the cobalt kids would need to work for about 700,000 years to earn the same amount Bezos makes in a day.

This inequality has been repeated throughout industrial history and is not unique to the age of ARTIFICIAL intelligence. But large-scale AI systems require vast amounts of data, infrastructure and maintenance, so few companies are able to build and operate them. If we can manage these systems well, it's important to think about what that means over time.Copy the code