Mark Zuckerberg was being interviewed about the early days of Facebook, and I — metaphorically speaking — had my ear to the door, listening to every word. I don’t know where this interview took place, when it happened, or how it got into the hands of a company called Speechpad.
That didn’t really matter. My job at the moment was to transcribe every word of this interview.
It was all part of an elaborate testing process for Speechpad, which finds people to transcribe audio files through Amazon’s Mechanical Turk program. I spent a week performing tasks for Mechanical Turk, which is a way of crowd-sourcing labor for menial tasks.
The type of quick tasks Turkers are asked to do range from taking consumer or social science surveys, to transcribing audio files and building simple spreadsheets. The pay is terrible: some tasks pay as little as a dime or even a penny.
Some requests amounted to online sleuthing, such as the person who wanted someone to find the addresses of short-term rentals just based on the listing. It’s even been used to try and find missing persons, by asking people to comb through blurry satellite images to look for evidence.
The highest-paying tasks I found were for transcription services, which paid over $50 — but for audio files that were well over an hour long. (Meaning it probably took three or more hours to complete.)
Amazon stumbled upon the idea for its Mechanical Turk program in 2005. It started as a way for Amazon to pay people to find duplicate web pages on its site. At the time, computer programs weren’t very efficient at finding these extra pages. But humans were.
Since then, the Mechanical Turk program has exploded in popularity. A 2015 World Bank report estimated that the Mechanical Turk marketplace has about 500,000 registered users worldwide, though not all of these users are active.It’s ideal for anyone conducting research who needs a large sample size. It’s also used by companies who need small tasks, such as tagging photos, performed.
The term “Mechanical Turk” refers to an 18th-century chess hoax. People were tricked into believing they were playing chess with a mechanical automaton, but in reality a person was making all the chess moves.
Of course, there has been sharp criticism since the program began over the types of wages being paid to Turkers. They were called “click-slaves” by one author of a book on crowdsourcing. During my week as a Mechanical Turk, I only amassed 95 cents from taking two online surveys.
One survey took about 20 minutes to complete and asked me lots of questions about secrets — when I was told them, how I felt about them and what kind of secrets they were. I’m guessing it was for a university researcher.
I found it surprisingly difficult to get my Turk hobby going. The people who needed tasks done prefer to hire people who had more experience in the Mechanical Turk system. I was rejected from taking several surveys for reasons I didn’t understand.
And I was choosy about tasks, avoiding ones that seemed too time-intensive for too little pay.
Given that I regularly transcribe interviews for my job, I thought I would be a shoo-in for transcription work. But after spending nearly an hour perfecting a transcription of that three-and-a-half-minute Zuckerberg interview, Speechpad told me I had failed their transcription test.
So who would do this kind of work? You’d be surprised.
New York University Professor Panos Ipeirotis developed a website that displays demographic information about Mechanical Turkers that is updated every hour.
On September 13, about 65 percent of Turkers were from the United States and 25 percent came from India.
They aren’t all destitute, either. Data from September 13 shows that 27 percent of the Mechanical Turkers had household income of between $25,000 and $39,999. Nearly a third made more than $40,000 a year.
Reporters or researchers who have interviewed Turkers found that some people do it as a hobby of sorts, a way to earn extra income. When NPR’s Planet Money team interviewed Mechanical Turk workers, they found people who performed these small tasks as their primary income because they had disorders, such as extreme anxiety, that prevented them from holding down a more traditional full-time job.
But here’s the bad news. As much as this program has been criticized for its low pay, there might be a worse alternative: artificial intelligence.
In the years since this program was developed, artificial intelligence has matured in leaps and bounds. For instance, Austin startup Alegion has developed a crowd-sourcing automation platform that works with the Mechanical Turk program.
When workers sign up to perform tasks, Alegion’s software builds data sets so that work can become automated in the future.
Just think about my attempt at transcribing the Zuckerberg interview. Right now, that’s a task that is still performed best by humans. I would love to outsource this part of my job to a computer. It’s boring, painstaking work.
But I’m not sure how I feel about artificial intelligence being used to make the boring bits of my job easier - by putting a member of Amazon’s invisible Worker Army out of work.
As for me, my time as a Mechanical Turk has come to an end. The highest-paid work was too difficult to get, and the easier tasks paid so low it was hard to justify the time spent doing them.