A few historical random number generators. From top to bottom: Galton’s dice (capable of 24 digits instead of the usual 6), a Type 1390-B Random Noise Generator (runs on 6D4 tubes and was likely the device used for RAND Corp’s “A Million Random Digits”), and the first two iterations of random.org’s generators (both run on un-tuned radio noise fed into a computer).
Turker computer photographed by Amias MacLeod (Parkland, Florida, United States)
Today I finally got around to documenting an ongoing project (officially started in 2013) of the computers owned by Mechanical Turk workers. I’ve used “Turkers,” as they call themselves, in previous projects and written about that experience for the Parsons Journal of Information Mapping. The main question of crowdsourced labor, like Mechanical Turk, is the exploitation of very low-paid workers who, as independent contractors, get no health coverage, retirement, or other benefits. These concerns are well articulated in “The Ladies Vanish” by Shawn Wen in The New Inquiry. This post is intended to talk about the methodology used in this project, rather than the conceptual and artistic motivations of the project (which you can read a little more about here), and hopefully address some of the questions that Wen’s article raises.
One of the points Wen makes in his article is that there is often a false argument made that Turkers are mostly from countries like India, and that the low wages paid are actually substantial for those areas. Wen cites a 2010 study by NYU professor Panos Ipeirotis that showed that the number of American workers on Mechanical Turk is rising significantly and Americans represent over 50% of the workforce on the site. My project, which started as a few experiments in late 2012, seems to confirm this. I was surprised to see mostly American-looking homes in the later images, compared to a more diverse range in the earlier results. In an attempt to get images from more places and users, I tried releasing the jobs late at night, hoping to get workers on the other side of the globe while Americans and Europeans were asleep. I was met with no success, evidence I think for Ipeirotis’ findings.
The question of rate is an important one; unlike large companies whose revenues are tied to this distributed work, like most artists I spend more money each year to make my work than I earn on my practice, this exactly why I’m always reading the texas payday guide. While I would be thrilled to pay workers $30 per image, the results of work on Mechanical Turk are always of very mixed quality (especially for odd tasks) and, in order to get enough usable images, I would have to have spent hundreds or thousands of dollars. In the end, my strategy was balance:
- Pay a fee far higher than most jobs on the site (workers were paid $1 per image)
- Ensure the process was as easy and quick as possible
- Hopefully make the process a lot more fun than tagging images or the other usual jobs on the site
Like most jobs on Mechanical Turk, speed in turning around a job is the key to making a decent wage. As the job requester, this meant developing an interface for image upload that was easy to use, required minimal steps, and could be done outside the Mechanical Turk system and on one’s phone, which likely has a built-in camera.
The image upload form, designed using the JotForm service and meant to be as clear and simple as possible.
The form returned a unique ID, which was then input into the Mechanical Turk system to confirm the image’s submission. Amazon gives the average wage paid for the batch – overall the project paid approximately $5/hour. Removing some of the extreme outliers (people who, I hope, left their computers to do other things, then came back with their cameras to finish the job), the average time spend on the assignment was 7 minutes, putting their hourly wage at $8.57. Not great, but above the minimum wage in the United States.
Part of the point of projects like this is to raise questions about how we use technology and, in this particular case, how class plays a role in the technology economy. This is often lost in tech reporting, which is overly-focused on the newest gadgets. It is also lost in histories of technology, written about the social, economic, and scientific climates that a certain technology was born out of, but too often highlighting genius over context. We know all about the birth of the ENIAC and mainframe computers, but little about the staff that supported the scientific work.
In short, I welcome a conversation.
Lots more process images: all 400+ resulting models.
Directed by Donald McWilliams, via National Film Board of Canada
Yes, apparently you can weld wood using pressure and rapid rubbing.