‘Humans Pretending to be Computers Pretending to be Humans’, by Oscar Schwartz


As we’re sure you already know, we here at the Brow are running our second annual Prize for Experimental Non-Fiction with help of RMIT’s non/fictionLab and The Copyright Agency. (If you don’t yet know all about the prize, what are you doing with yourself and your life? Get on over to the Prize page to learn everything there is to know about it!) But what is experimental non-fiction, exactly? Well, you can get a sense of what we’re after by reading some works by the authors we recommend on the prize page, but you can also get a sense of it by reading the winner of the inaugural TLB Experimental Non-Fiction Prize, Oscar Schwartz’s essay ‘Humans Pretending to be Computers Pretending to be Humans’. We’re republishing it on our site not because we want to see more essays like it—what could be less experimental than cribbing notes from the previous winner of a literary prize?—but in the hope that it will inspire you to consider different ways of experimenting with form, voice, style, point-of-view, and so on. We hope you enjoy reading it, and, even more, we hope it gets you fired up to submit something of your own to the prize, which closes on May 29th of this year.

In 1770, Wolfgang von Kempelen stood in front of Empress Maria Theresa at her court in Vienna and proclaimed to have built a mechanical man that could beat humans at chess. The mechanical man—or ‘the Turk’, as von Kempelen named him—was life-sized, carved from maple-wood, dressed in ornate robes and aturban, and sat behind a large cabinet, on top of which was a chess set. Von Kempelen opened the cabinet to reveal a labyrinth of levers, cogs and clockwork machinery. He then closed the cabinet, inserted a large key, wound it up, and after some ticking and whirring the Turk lifted its head, studied the board, took hold of a white pawn and moved it forward two places. News of the Turk spread, and chess masters from across the empire travelled for their opportunity to play the machine; they usually returned home defeated. For the next few decades the Turk toured Europe and America, trouncing some of the most formidable minds of the time – Catherine the Great, Benjamin Franklin, Napoleon. Legend has it that Napoleon tested the Turk by making illegal moves, but the Turk grew fed up, and swiped the board.

A young Edgar Allan Poe, who was obsessed with the machine, suggested that the Turk was a hoax. Poe was right.

The Turk’s success evoked varied responses. While some conceded that humans had actually been surpassed by machines, there were a host of counter theories. One was that the Turk was controlled via magnets from a distance. Others believed that it was operated by a spirit held captive in the machine. A young Edgar Allan Poe, who was obsessed with the machine, suggested that the Turk was a hoax, a brilliantly constructed diversion machine that was controlled by a human hiding in the cabinet. Poe was right. The cabinet on which the Turk sat was constructed to conceal a person, and most of the gears and levers were for show. When the cabinet was open, the human would shuffle quietly among them to avoid detection.

This story, with its cast of famous characters, provides an almost mythological dawn for the age of artificial intelligence, signifying some of the radical adjustments human consciousness has had to make in coming to terms with the fact that intelligence might not be what makes us unique. But when the story is retold, there is a certain character that always remains invisible and voiceless. No one ever talks about the person sitting inside the cabinet controlling the Turk, scurrying among the levers and clogs to avoid detection. The Turk was essentially just an elaborate puppet, and all of its many great achievements were thanks to a real human hiding inside. History has made the Turk into a star, whereas the person inside is a ghost in the machine. We are willing to ignore the human for the romance of a thinking machine.

My name is Beverly and my day started out at 8:45 AM. I hadn’t slept well, and had a late night. I had a lot on my mind from this past week and I was feeling very tired and down, right from the start. I came out into my kitchen (I live in a small two bedroom apartment with my husband and two cats) and decided it was time to make breakfast… I couldn’t decide what I wanted because I was too focused on talking myself out of crawling back into bed. It was probably 9am when I said to myself “I NEED coffee.” Once I had the coffee water boiling I prepped my French press for the job, 2 tablespoons of 8 O’clock original roast coffee. Once that was steeping, I decided to grab a package of pop-tarts to eat for breakfast while I began my day working on the computer. At this point I realized that my husband (Daryl) was already awake and watching TV in the living room, he had the day off of work and I had completely forgot. I went in to greet him and say good morning. It was at this point probably 9:15 AM. After we had our normal “good morning, what are your plans for the day” conversation I finally went and poured my coffee and sat down at the computer to begin my work on Amazon Mechanical Turk.

My name is not actually Beverly. I didn’t wake up at 8:45am, or eat Pop Tarts for breakfast, and I don’t have a husband called Daryl. These experiences belong to someone else. But I did pay for them using Amazon Mechanical Turk—$5 to be precise—and now, in a way, I own them. Amazon Mechanical Turk, or mTurk for short, is a service offered by Amazon that promises a scalable, on-demand workforce for menial, computer-based work that requires human intelligence. This workforce is not made up of Amazon employees, but an internationally dispersed and anonymous group of workers who sign up to mTurk and complete tasks, which usually require no specific skill or training, for small sums of money. mTurk advertises itself as “artificial artificial intelligence.” The tasks that the human workforce complete are rote and repetitive — like transcribing audio or captioning photos — but which computers still find challenging.

Those who use mTurk can be divided in two groups: Requesters, who outsource work; and Workers (affectionately known as Turkers), who complete the tasks that the Requesters set. A Requester logs into mTurk and creates what is called a HIT (Human Intelligence Task). For example, a Requester might put out a HIT asking for Workers to read a receipt, extract all of the purchased items and their price, and then calculate the total. Often, the Requester will put out tens of thousands of these HITS at one time, and pay around ten cents or so for the completion of each one. If a Worker wants to complete this request, they accept the HIT, read the receipt, total the price, and then send the data back to the Requester.

Like the chess-playing mechanical Turk of the eighteenth century, mTurk is designed to make human behind the machine invisible.

Earlier this year I became a Requester on mTurk. The first and only HIT I made was as follows: “Please send me a minimum 2000 word journal entry about your day. Please provide as many concrete details as possible without providing any information that would give up your identity. Please use fake names.” I set the price at $5, an expiry time of twenty-four hours, and total budget of $35, meaning that after the first seven Turkers submitted, the HIT would terminate. I filled this quota within the first four hours of posting the HIT. That evening I sat in my room and read through the private thoughts and private conversations of seven people I had never met. The strangest thing was that this whole process was mediated entirely via the Amazon webpage. There was no personal interaction required between me as Requester and them as Workers. In fact, it feels like the site is set up to discourage any type of personal communication, as the Worker is presented to the Requester not by name, but as serial number. If you’re requesting large scale, impersonalised work—like totalling purchased items on a receipt—the effect of this computer-mediated service is to make the interaction appear not between a human and a human, but a human and software, as if it’s just a matter of executing a program. Like the chess-playing mechanical Turk of the eighteenth century, mTurk is designed to make human behind the machine invisible.

Seeing as though I am self-employed and make my own hours, I was not in a terrible rush to get to work. So, I just sort of hung around the living room with my dogs and mother until I felt I needed to get started on work or I’d be working all day. I unenthusiastically walked back to my room, “my office” as I call it, and sat down on my bed. My room is where I work every day in my pajamas. It’s the coldest room in the house, so I slipped on my slippers and covered my feet and legs with two blankets and a comforter. Before I start actually working, I go through a ritual of checking all of my social media sites including Facebook, Twitter, Tumblr, my email, Youtube, and sometimes eBay if I’m looking to bid on something in a certain timeframe. I spend a little time looking at my notifications, going through messages, etc until about 1:00 p.m. when I actually get down to looking at what I have to do for the day for work. I do transcription from home and usually have an abundance of files that are waiting to be done. I opened up my folder that contained the files that needed to be transcribed and grabbed my mechanical keyboard from under my bed where I keep it in order to get started. I began typing out the day’s work.

John Maynard Keynes pointed out that although new technology can cause short-term unemployment by automating tasks previously done by humans, because automation raises productivity by economising on the use of labour, it ultimately generates new products and services leading to previously unimaginable forms of employment. These new forms of employment, Keynes acknowledged, destabilise the work force, creating what he euphemistically called a “temporary phase of maladjustment”, in which the idea of labour is turned on its head. For example, during the early stages of the Industrial Revolution, manufacturing was done on a small scale at home or in a small workshop. This meant that manufacturers were essentially artisans creating products from start to finish. In the later stages of industrialisation, large-scale manufacturing moved to the factory. Suddenly, workers were given one or a few repetitive tasks, making components of finished products rather than whole pieces. The disconnection between labour and product created disillusionment, and productivity suffered. Bosses subsequently learned to impose tight schedules to maintain productivity. Workers were made to feel expendable, as their labour required hardly any artisanal expertise specific to them as an individual. The machine technology of the Industrial Revolution did not just automate previously human tasks, but reshaped the concept of human labour into the sort of precisely defined mechanisms that the new technology needed–cogs in a much larger system.

Like the steam-powered machines of the nineteenth century, digital technology is now forcing us to re-evaluate the meaning of human labour. Increasing algorithmic sophistication has meant that software is encroaching on fields that previously seemed uniquely human. Think about how seamlessly we accept self-checkout at the supermarket, or how familiar we’ve become with speaking to automated voices on the phone. It’s been suggested that nearly half of the jobs in the developed world could be automated within the next two decades. The question is: if all these jobs are being taken by computers, what and where are those new jobs that Keynes promised follow technological development?

I explained, as usual, that I had done basically nothing with my day other than work and sit around relaxing.

For lunch I found a chicken melt type sandwich that you put in the microwave and placed it on a paper plate. I placed it in the microwave for two minutes until it was sufficiently melted and opened the microwave door. Because my life is so boring, I then returned back to my room which also doubles as an entertainment center where I turned on my TV and started to eat the lukewarm chicken melt that I had just nuked. I’m constantly on my laptop as I do this and went to Facebook where I spend the majority of my days posting statuses about oddball things I come in contact with and sharing memes with the rest of my pack of weirdo followers. I talked online to my friend Bill for a while until around 5 o’clock about what we had been up to for most of the day. I explained, as usual, that I had done basically nothing with my day other than work and sit around relaxing. I turned on some Bob Dylan, muted the TV, and still proceeded to zone out while looking simultaneously for HiTs I could do on mTurk for a bit of extra money. This whole time the only real-life person I had come in contact with today was my mother.

Tech-utopians are hopeful that when computers become sophisticated enough, they will take over all the jobs we don’t like doing, allowing us to focus on more creative and compassionate occupations, inherently unsuited to machines: a workforce of dancers, relationship therapists, and yoga instructors. Realistically, though, the type of labour performed on mTurk—menial, freelance, context-specific, as-yet-to-be-automated-but-seemingly-automated work—seems more representative of the future of human labour. mTurk was launched in 2005 and already its workforce has grown to 500,000 from 190 countries. Luis Von Ahn, a Guatemalan entrepreneur and computer scientist believes that mTurk falls under a broader category of what he calls “human computation”. Human computation, Ahn explains on his blog, “harnesses the combined power of humans and computers to solve problems that would be impossible for either to solve alone.” An example of human computation is reCAPTCHA. The way reCAPTCHA works is much like a normal CAPTCHA – those weird letter/number combinations you have to type into your browser to prove that you’re not a bot. Instead of making the verification code random letters and numbers, reCAPTCHA uses scanned images of a few letters from not-yet-digitised books. This means that when you type in the code you are helping transcribe the book into digital format, one word fragment at a time. Ahn’s reCAPTCHA takes a specifically human skill, and uses digital networks to transform it into massive-scale collaboration. He says that over 750 million different people—more than ten per cent of humanity—have helped transcribe at least one word of a book using reCAPTCHA.

Like reCAPTCHA, mTurk integrates the not-yet-automated and entirely ordinary capacities of human workers located all over the world and organises them via digital networks so as to function with the power of a computational resource. This is why Amazon calls mTurk artificial artificial intelligence: it is powered by humans pretending to be computers pretending to be human. Just as the human became a cog in the wheel during the Industrial Revolution, the human becomes an iteration of code in a piece of software during the digital revolution.

This is why Amazon calls mTurk artificial artificial intelligence: it is powered by humans pretending to be computers pretending to be human.

New types of labour made possible by digital networks have some obvious upside. For Requesters, mTurk provides a cheap, reliable, online flow of replaceable workers. From the perspective of the Worker, the best thing about mTurk-like employment is that it does not discriminate. If the task is completed according to the rules, it makes no difference who is doing it. Consequently, mTurk provides a means of self-employment for those who have previously not had access to the workforce. A study of mTurk demographics conducted in 2010 revealed that many Turkers are US-based single parents, usually female, who use the service as a secondary source of income to support their families. Also significant is the fact that thirty-four per cent of Turkers come from India, where differentials in minimum wage means that working on mTurk can become a primary source of income. mTurk’s relatively frictionless point of entry—you still need internet connection and a computer—coupled with the fact that Turkers control their hours and work from home, means that mTurk is the perfect employer for those who are often excluded from the workforce in developed Western nations, and breaks down national boundaries that limit those in the developing world from accessing commercial benefits from the global network.

At 1:00 pm, I brushed my teeth. I looked for my outfit for the day. I picked out underwear, a bra, and a shirt. I got a towel and placed them all in the bathroom. I took a shower. I first washed my hair with head and shoulders, conditioner my hair, and then I scrubbed myself with Dawn dish wash soap. We are currently out of body wash. I keep meaning to buy some but I keep forgetting. I turned off the water when ever I scrubbed to minimize water usage. I dried myself and put on my clothes. I opened the bathroom door, washed my hands, and left the restroom. I put a layer of almond oil on my skin. It was very heavy. I applied a layer of sunscreen to get rid of the oiliness. It takes forever to all sink in. This is not enough. I then apply Monistat chafing cream as primer. I apply BB cream and wait a while for this to sink it. I apply lip liner and then a bit of lipstick. I apply concealer under my eye and on blemishes. I then used BB powder all over. I apply more on problem areas. I apply a nice pink blush on the apples of my cheek. Next I fill in my eyebrows with Anastasia dip brow with my MAC brow brush. I do not do a good job. I next curled my eyelashes and applied two layers of mascara. I used a finishing powder on my entire face. Next, I sprayed it with finishing spray. I plugged in my blow dryer and dried my hair. I looked for some pants, put them on, looked for socks, put them on, and put on some boots. I found my glasses, lip gloss, and an evaluation form for my supervisor and stuffed it all into a bag.

One of the most troubling side effects of human computation is that it obliterates difference and personal identity in a human workforce. mTurk’s design encourages Requesters to perceive Workers as lines of code, or part of a software-based cloud. This allows Requesters to experiment with human labour like a programmer might experiment with a line of code. If it doesn’t run, you get rid of it, and try again. There is no human accountability or ongoing commitment to those employed in the experiments. Seeing Turkers as human software makes it easier to charge tiny amounts for work that may take a fair amount of labour time; wages on the site average between $2 to $3 an hour. In fact, Requesters can refuse to pay for work if it doesn’t meet their standards with no legal consequences. Recourse for Turkers is limited to contacting the Requester through mTurk’s web interface, but even then, Amazon does not require Requesters to respond. The problem with this model of labour is that any form of human labour that competes with computation must accept the economic conditions of computers. By repackaging human labour as a form of computation, mTurk is implicitly comparing its labour market to that of a slave workforce: a workforce that has no autonomy, that exists to be exploited, that does not receive compensation for labour.

By repackaging human labour as a form of computation, mTurk is implicitly comparing its labour market to that of a slave workforce.

This is not to say that Turkers are slaves. They are not. They can choose whether or not they sign up, and they also choose what work they do and when. But using the metaphor of computation for human labour has psychological implications, particularly for the person completing the work. To begin with, mTurk labour is conditional on the failure of developers to make new programs that could replace you. Like software, your work will quickly become obsolete, and once the update comes, you can be sure that no one will revert back to using the old stuff. Also, the computational nature of HITs on mTurk creates an even larger disconnect between labour and product than in the factory. You don’t know who you’re calculating these shopping lists for, why you’re doing it, and what impact your labour will have on the world. It’s as if the model of the factory has gone wireless and entered people’s homes, creating an ever-shifting, concatenating production line. Finally, unlike machinery, software is incorporeal. Software is idle until it is executed. It exists to work, has nobody extending in space and doesn’t do anything unless instructed.

After my lunch break ended at 12:00 noon, my next class arrived. This was a geometry class, and we were going over graphing slope and solving for y-intercept equations. Some of my students did not fully understand what we were going over together, but I did my best to demonstrate to them how to solve the material. Since I am not their regular teacher, there is only so much that I am able to do in order to help them, and I have to accept the limitations of my role as a substitute teacher. Reluctantly, I let each student know to just do their best and moved on to the next class. The next course started at 1:00 pm, and was another algebra class.This class was extremely routine, and very little out of the ordinary happened. It was a refreshing break, and I enjoyed the ease with which this class progressed. At 2:00 pm, my planning period began. This was an hour-long break period with no students and no course work to teach. During this break period, I logged in to my computer and signed in to Amazon Mechanical Turk in order to complete some assignments and to make a little extra money during my brief period of down time. I am always looking for new ways to earn more money, and Amazon Mechanical Turk provides me with just such an opportunity to supplement my substitute teaching income and keep me productive during my down time. Finally, at 2:00 pm, my final class of the day arrived. This was another geometry class, and we went over the same material, covering slope and y-intercept equations on graphs. The class went well, and I encountered no problems during it. At 3:00 pm, the bell rang, dismissing school for the day. I dismissed my students, packed my belongings, and headed down the stairs to my car in order to leave the school for the day.

Kristy Milland signed up as a Worker on mTurk in November 2005. She was in her late twenties, a mother of one, and running an at home day-care centre in Toronto. At first she used the service to subsidise her income, but now, almost ten years later, she tells me that Turking is her full-time job. Milland, who studied psychology at Ryerson University, has been carefully examining the emotional effects of her long-term engagement with mTurk. Like most jobs, Turking involves significant stress. But this stress is compounded, Milland tells me, by the fact that her ability to earn money on mTurk is less dependent on her work ethic and more on who happens to post work at any given time. “If it’s a good Requester with lots of work, maybe I can end my day early. If there isn’t much to do I may be chained to my desk from sunrise to sunset, and I still might not make my goals. Not making my goals means not buying groceries, paying the rent or utility bills, or purchasing needed prescriptions.”

Her ability to earn money on mTurk is less dependent on her work ethic and more on who happens to post work at any given time.

The novelty of working as a Turker, Milland explains, also implies a certain type of alienation. Her friends that work in IT do not consider Turking part of the industry, because it doesn’t require specific technical training. And her friends that work offline don’t consider Turking to be “real work”. “Anyone I tell about my problems tells me I should thank my lucky stars I can work in my pajamas,” Milland says. “I am an outcast, a black sheep, outside of societal norms of what a worker is, and that leaves me with little leverage to change things for the better. It also leaves me feeling lonely. I can’t tell my friends and family about the troubles or triumphs I face as they just don’t get it. The amount of effort it takes to explain it often leaves me drained and them confused.”

These feelings of alienation prompted Milland to search for other Turkers online. She found Turker Nation, the oldest continuous forum for the discussion of mTurk-related things. Turker Nation, Milland discovered, has huge upside for both Workers and Requesters. The forum allows both halves of the market to talk in a human way in order to figure out how a HIT might be tweaked so that tasks can be completed more quickly and accurately. It allows Workers to voice their grievances with certain Requesters who don’t pay up when a HIT is complete. And it also provides Requesters a forum on which to discuss the growing issue of Workers who automate spam responses to HITs, in the hope that Requesters do not review work, but just pay the five cents automatically. Spam undermines the integrity of mTurk, which is bad for Requesters and Workers; so on Turker Nation, they have worked together to make lists of Workers who are known to spam, and lists of Requesters who are reliable. Beyond the optimisation of the system, there is an entire ‘social area’ on the forum, where Workers share personal anecdotes, and give each other tips on how to avoid red-eye and back pain after long sessions of sitting in front of the monitor.

Milland is now a community moderator on Turker Nation. She is adamant that these forums illustrate that online, crowd-sourced work is a sustainable form of labour, but only if communication between Requesters and Workers remains open, to make sure both feel a sense of accountability towards one another. Rather than falling back on failed examples from the offline work environment, Milland wants to use mTurk as an experiment for new labour relationships: “Online work makes failures obvious, such as wage theft and precarity of employment, so it also stands as a perfect place for experimentation.”

Late last year, Milland started an online campaign called “Dear Jeff Bezos”, which encouraged Turkers to write the Amazon CEO an email about how they feel about their work. Milland also asked that those writing emails forward them to her, so that she could publish them online. The purpose of this exercise, Milland told me, wasn’t to troll Bezos or denigrate Amazon, but to “help change the reputation of workers by showing who we truly are as educated, skilled, intelligent people with a wide variety of talents and abilities.” For instance, a letter from Manish, a Turker from India, begins by thanking Bezos for creating a platform that allows him to “earn a living from the comfort of our homes.” The rest of the email lists a number of grievances Manish has with the system – in particular the fact that Indian workers have no option but to withdraw money from Amazon via mailed cheques that take six weeks to arrive, and are often lost in transit. “If I had to sum up all my demands in one sentence,” Manish writes, “it would be that I want Amazon to start caring for the workers a bit more.”

The power of a letter like Manish’s, Milland explains, is that it breaks open the idea that mTurk workers are an anonymous, computational resource, or that human labour can be used in an algorithmic way. It provides a human voice to the menial task. This is precisely what Milland believes has to happen in order for mTurk to be a viable, ethical form of labour in the future.

I worked on the computer for another few hours desperately scrounging, hunting, and picking up jobs until around just after midnight, when I found this writing hit: “A two thousand word report about your day.” I looked at it and decided why not give it a shot. Yes, it sounds weird to write a report about my day that a complete stranger will see, but somehow it was almost relieving, and liberating to do it. It made me feel invigorated, and was sort of a way for me to vent some of my stress without worrying whether the person I’m telling is listening or not, knowing that I can get it all out before something interrupts me. I honestly enjoyed it in a weird way, and now at 1:50 AM i plan to go finally call it a day and relax and try to get some much needed restful and restorative sleep.

What is the underlying ideology of Amazon that it wants to make us see each other as computers?

Like many new labour technologies, mTurk creates new workforce markets accessible to people previously barred from finding employment, and it also facilitates new methods of exploitation. Kristy Milland is saying that we should try to reduce exploitation as early as possible, and the best way to do that in this case, is to listen carefully to the grievances of the Workers. The difficulty is, like von Kempelen’s Turk, mTurk is designed to disguise the human behind the machine. This design increases exploitation by reducing a sense of accountability or communication between Requesters and Workers. Requesters not paying Workers, or Workers spamming Requesters – these are symptoms of an underlying cause: the reduction of the human worker into source code. Milland doesn’t understand why Amazon advertises Workers as artificial artificial intelligence instead of what it really is: human freelancers. What is the benefit, she wonders? What is the underlying ideology of Amazon that it wants to make us see each other as computers?

For von Kempelen, the success of his Turk required that the human remain well hidden inside the machine, and also required that the person inside was willing to pretend that their very human intelligence was mechanical. But who was this person? It turns out that it was no one in particular, but an ever-changing collection of talented, but not professional, chess players. As the Turk toured around Europe and America, von Kempelen would send his assistant ahead to find a local chess player who needed work badly enough to spend a few weeks playing chess cramped inside a mechanical man. The success of von Kempelen’s Turk required not only that the audience believe that a mechanical chess player was possible, but also on the existence of a workforce of talented yet underpaid chess players willing to merge their identity with the machine for a bit of extra cash.

What an eventful day this was! I hope to have many more days just like it, and that tomorrow will be a wonderful day as well. My life can be very busy, but I enjoy living it, and am grateful for all the wonderful experiences and people that I come into contact with on a daily basis. Let’s hope tomorrow is just as great as today was, and that all days in my future are wonderful as well! Each day is what you make of it, and I hope to make as many days as I can memorable, fun experiences that I can look back on with fondness. In the end, I can thankfully say that I made today count.

Oscar Schwartz is a writer from Melbourne. He is currently writing a PhD on whether computers can write literature. He tweets at @scarschwartz.