Arturo & I decided to collaborate on project #1:
“Working with a partner, use a website such as Mechanical Turk to ask a question, make an offer or assign a task.”
Our original conversation started out curious about the workers themselves. We thought of requesting self portraits.
Self portrait as a ____
Self portrait in a ____
We were online last Sunday chatting about last weeks readings, and about project ideas and came to the very efficient idea of requesting the Turk workers do the reading / summarizing for us. A two-fer, as it were. And for a bonus, we’d ask for a quick self-portrait. We looked into using a live HTML drawing tool for workers to draw, but Arturo suggested we go for ASCII portraits, a la his email signature, almost as an afterthought.
We gave ourselves a budget of $50 ($25 each). Arturo scanned the pages, counted the number of paragraphs, and we worked up a price per paragraph scheme: 50¢ per paragraph, 50 words max. We’d get several summaries per paragraph and then pick the best and reconstitute it into a single summary for each of the week’s six readings. They would be a fragmented versions of the texts, with the shorter paragraphs much more likely to be summarized than the longer ones based on the 50¢ flat rate per paragraph. We thought it would be interesting to see if workers would engage a text in its entirety (which is what usually occurs) or if they would get simply go for the highest time/pay ratio. What happens to the meaning of a text when it is summarized in parts rather than as a whole? Would all the paragraphs be summarized or just the easiest cost effective ones? Would we be able to make sense of a assembled text to complete our homework?
The system of Mechanical Turk has a lot of interesting rules that preserve the sense of anonymous community, as well as consideration for labor ethics. Our plan was not going to cut it. First of all, positive reinforcement is used in approval system, just like in animal training: if a worker doesn’t do the job (trick) well you must reject their work and they won’t get the payment (reward). Doing so, “will train a worker who is making a specific mistake.” Setting the payment really low would not only lose the interest of a lot of the workers, the ones who actually do it would be set up to fail due to poor incentive. We didn’t think the added value of art theory education would get any traction.
Our task also had a poor cost effectiveness. We had to scan the documents, isolate the paragraphs in Photoshop, number them, and embed them as individual tasks in the Mechanical Turk interface. Then we had to think of parameters that would aid the workers in being most efficient at their task, ie setting them up to succeed by removing as many speed bumps as possible in the task description. Quickly it clear that this was unwieldy and going to take forever, so we started narrowing down to just a couple of the shorter readings. It was still a ton of paragraphs, so we chose to do just the Tiravanija reading, as we were wondering what people would make of the individual paragraphs in that particular article. While uploading the individual paragraphs into a blog made for just this purpose, and pasting paragraphs into individual Turk HITs, it became apparent that we had to do something else.
The self portrait bonus box just looked so appealing and perfect on its own. We discussed and agreed to ditch the paragraph summaries and go straight for the visual goods. At $1 per portrait, we requested 45, setting aside $4.50 for the administrative fee so as to stay under budget. A total of $49.50 spent.
The hits were created and published at 1:00pm on September 23rd.
All hits were completed by 2:46pm on September 23rd.
Less than 2 hours.
We were amazed to see how quickly they were all taken and completed. Such an incredible feeling that we’d hired this anonymous mass of humans and they were simultaneously churning out their individual mini works of art. I wondered if they were enjoying themselves. According to the batch manager, the initial time spent was closer to 3 or 4 minutes, with an hourly rate of $16-17 ish. But by the end, the average time spent was 5 minutes, with an hourly average of $11.96.
Watching the batch results as they were streaming in, we were extremely disappointed to see that there were no carriage returns in the comments… the portraits showed up as a single line of characters. We’d never be able to recreate the portraits exactly.
But when the batch was done, there was an option to download the .csv file and there, like presents, were the portraits as created by the Turk workers. They were clearly enjoying themselves. Only one appears R-rated & with a caption “Naked Me!” Some are quite creative with use of characters and very personable.
Conclusion: ASCII self-portraits are quite descriptive of personality and looks through a few simple shapes and lines. The low-resolution solution gave people a somewhat level playing field as the technical prowess of a good draughtsperson would not be as helpful without the HTML drawing tablet. There are a limited number of ASCII characters, and we hoped the finite number of possibilities would compel workers to identify expressive shapes with their own looks. We wanted the workers to be aware of themselves, to think of themselves as they worked for [100] pennies. We also wanted to have the signature of each person who worked on it. Indeed, the signature of 45 people is the piece itself. Perhaps a title would be untitled in parenthesis 45 self-portraits in ASCII.

The assignment as the workers would see.

Batch details from the Turk site

Oh no! The portraits appear on a single line... the artistry lost.
And finally, as the workers intended, more or less.

Mechanical Turk Worker Self Portraits