This is yesterday’s issue of Time’s Corner. Subscribe here.
One of the creepiest services I’ve ever come across online is an Amazon service called Mechanical Turk. The service is named after a sideshow curiosity from the 1700s, a turbaned robot that could beat any human at chess. It was touted as a marvel of mechanical engineering. The trick was that there was a small person inside the Turk, controlling its movements.
Amazon’s Mechanical Turk lets you remotely hire people to do menial tasks for miniscule sums of money. Let’s say you have a task so repetitive and boring that your mind gets numb just thinking about it—e.g., changing every “5” to a “6” in an Excel spreadsheet. Through Mechanical Turk, you can hire someone else to do it at a tenth of a penny per change. A Pakistani worker changes a thousand 5’s to 6’s and you pay him a dollar. What looks like an automatic system from the outside is actually a guy frantically clicking and typing on a computer. Unlike their eighteen-century counterparts, Amazon doesn’t try to hide the fact that there are humans doing the work. In fact, the tagline for the service (now removed from the website) is “Artificial Artificial Intelligence.”
Every machine in human history needs to be operated by a human at some level. An ax helps you cut down a tree, but you need to swing it. A BMW gets you quickly from here to there, but a person has to design it, build it, maintain it, and drive it. No matter how automatic or magical a manmade object seems, it always draws on human power to function. The ones that seem the most magical are the ones that keep the human operator most hidden, like the Eternal Engine in Snowpiercer that’s actually powered by small children stuffed between the gears.
An AI tool is an elaborate machine that hides its human human operators so well that it seems to be thinking on its own. This is true on the input side, where thousands of workers in Kenya crawl through the opioid palaces of the internet and flag content that’s deemed “too toxic,” and the output side, where thousands of remote workers review the “choices” of self-driving cars.
It’s incredibly important for us to remember this. The worst part of AI tools is that they absolve people of their wicked deeds, or at least provide them with plausible deniability. Matthew Butterick aptly describes this as “human-behavior laundering”:
If AI companies are allowed to market AI systems that are essentially black boxes, they could become the ultimate ends-justify-the-means devices. Before too long, we will not delegate decisions to AI systems because they perform better. Rather, we will delegate decisions to AI systems because they can get away with everything that we can’t. You’ve heard of money laundering? This is human-behavior laundering. At last—plausible deniability for everything.
What AI really provides is an excuse. We’re not stealing your stuff. AI is. We’re not driving your car into an eighteen wheeler. AI is. We’re not whipping a crowd into a frenzy. It’s artificial intelligence. In other words, no individual person is responsible. It’s bureaucracy at its finest, the Orwellian passive voice writ large.
Christian Leithart 