The fourth post from my Just Two Things newsletter on Artificial Intelligence—a guest post by Peter Curry

This is maybe the closest that AI researchers will ever come to writing a Norse saga, so strap yourself in and buckle your seatbelts. 

In a long but readable paper, Kate Crawford and Vladan Joler cover every single thing that comprises an Amazon Alexa, from the raw materials, to the research, to the marketing, to the poor quality living conditions of the miners, to the people speculating on the miners.

“Put simply: each small moment of convenience – be it answering a question, turning on a light, or playing a song – requires a vast planetary network, fuelled by the extraction of non-renewable materials, labor, and data.”

An endless string of resources

Crawford and Joler relentlessly document the frangible future of AI, which depends on an endless string of non-renewable resources and exploited labour.

Image: Crawford and Joler

They also produce a map, documenting the Anatomy of an AI. It’s worth downloading a PDF of the map and having a play, because it is information dense.

“If you read our map from left to right, the story begins and ends with the Earth, and the geological processes of deep time. But read from top to bottom, we see the story as it begins and ends with a human. The top is the human agent, querying the Echo, and supplying Amazon with the valuable training data of verbal questions and responses that they can use to further refine their voice-enabled AI systems. 

At the bottom of the map is another kind of human resource: the history of human knowledge and capacity, which is also used to train and optimise artificial intelligence systems.”

Sorting the data

Let’s visit the bottom of the map for a second. These two circles are the source data of AI. If you follow this newsletter closely, you’ll be familiar with some of the problems of AI, but as you go around the circle, the complexity of turning so much data into any sort of cohesive system is made manifest. There are multiple problems with sorting and refining any of those data points into uniform data tranches that are easier for machines to work with. (Here, here, and here, for more on this).

I’m not going to be able to represent even a fragment of the depth and detail that has gone into the construction of this piece, but there are two more things that are worth reviewing.

Complex supply chains

The first is the sheer complexity of the supply chains that create these simple tools. Mines, smelting, component manufacture, transportation, assembly, waste. All of these are inordinately complicated in their own right, and don’t begin to include any of the internet infrastructure, or AWS infrastructure, or decisions about how AI is trained, which would all demand essays in their own right.

“One illustration of the difficulty of investigating and tracking the contemporary production chain process is that it took Intel more than four years to understand its supply line well enough to ensure that no tantalum from the Congo was in its microprocessor products.”

It took Intel more than four years to understand its own supply chain. It is nigh on impossible to track the companies that you’re being supplied by, because they in turn have their own supply chains, which have their own supply chains. It is a nightmare for the company itself, let alone any researcher or journalist. That there is a new academic field of supply chains, as mentioned in this newsletter a few weeks ago, is a positive step, but also a monumental task.

Mechanical Turks

The second is the ‘Mechanical Turk’. One of the main points of the piece is often AI will look big and scary and monolithic from the outside, but will actually require a significant amount of human intervention on the inside. The Mechanical Turk is Amazon’s admission of defeat on the ability of neural networks and machine learning algorithms to fully train themselves. Instead, they pay online workers scarily low wages to help train the machines.

The original story of the Mechanical Turk is a more comical affair. Invented by Wolfgang von Kempelen in 1770, the Turk was the first ever chess “computer.” It was also a big fat lie. It was just a big machine that let a person hide inside and play chess. Sometimes it can feel as if the world is overfilled with complexity and that understanding is impossible. Sometimes you just need to pull back the curtain and ask the Wizard of Oz what’s happening.

Kate Crawford’s new book, Atlas of AI, draws on the work described here. She was interviewed recently by The Guardian about the book.

A version of this post also appeared on my Just Two Things Newsletter.