Todd Lewandowski

So I’ve tried out this thing over my Chinese New Year break that you may have heard of, it’s called the “Amazon Mechanical Turk”.

The phrase “mechanical turk” has an interesting history, it was originally a machine with a life-size doll of a Turkish dude from Turkey (though probably incorporating all other Ottoman and Middle Eastern references from the times) that was built to play chess. Although the doll was obviously fake, it moved by itself and moved pieces to actually play chess! And this was 200 years before computers and mechanized robots! Later it was discovered that it was only a really short guy crammed into the machine who was playing Chess in a very cramped setting. The short dude was trained as a Chess master, and when the machine was “opened” for inspection he would simply slide into another compartment (standard magical misdirection).

The Amazon Mechanical Turk is much the same, it purports to be “artificial intelligence” like a computer but is actually powered by humans. What this means is that businesses, or requesters, have a bunch of small tasks, “microtasks”, that they upload to the system and then when other people finish them they are downloaded from the system back into their website or other part of their business. Ideally each microtask is boring enough that it could be completed by a computer program, but the status of AI research has not evolved enough to solve that problem. Thus the term of “artificial artificial intelligence”. Each microtask is worth $0.10 or so, very small. Amazon takes care of the micropayments (no credit card company would deal with amounts so small) that allow you to cash out whenever you choose. I am sure that Amazon takes their 10% cut or so from the business’ side in order to power the platform.

That’s enough for the explanation. You can read about how it works elsewhere. But now is my opinion on how it actually works in practice. I just started to work, only for a day or two. I’ve submitted about 100 HITs (human intelligence tasks, or microtasks) or so. I have made all of $5.99 so far. When my other pending tasks come in I will probably make around $10 or so. So as far as money goes, it is just not worth it. Minimum wage (as I learned from the State of the Union) is about $9 and I definitely worked around 3 hours or so. I would make more money tutoring here in Taiwan ($20 an hour) or even playing poker online!

The advantage is though, that I can stop/start at anytime. I suppose if this platform were extendable to cell phones I could “work” while on the bus/subway commute to my actual job. It would be the same nervous fidgeting and keeping busy as a game or checking Facebook, but I could actually be making a dollar or two. It is easy enough to start too. So far I don’t have the connections or courage to build out my tutoring network either. I’m still working on that.

The content of the tasks varies. One of the tasks is “Copy information from a business card”. This wouldn’t be so bad but the payout is only $0.02 per card. I tried it and I figured out that I can’t possibly work that fast and that accurately. Another bad one was “Give tags to artwork”. Not so bad until I realized most of them were abstract modern pieces no one really understands. So much for that. Another was “Does this page answer this question?” task. Basically it was trying to help build out a Wikipedia clone or something.

I spent most of my time on “Write a title and description for an adult video clip”, haha. That had a relatively good payout (5 cents per clip), I was able to work fast, and enjoyed it too, lol. I even got a “thanks for the good work!” comment from the reviewer, lol. But even in relative “porn terms” the clips were not that valuable. I mean, that a man would not really get turned on by watching these clips. At best they are useful only for a tease or an advertisement. Maybe that is why they were contracted out, to get better descriptions and move more traffic.

The highest value task you can do is a psychology survey for a university. If you have ever taken these in college, they are much the same. Even online. There is usually some hook or trick, then they ask you a lot of questions about how you feel. A lot of the feeling words are very much the same, so you get a little frustrated at being asked the same question again. There are also questions that say “Fill in Strongly Agree for this answer” just to make sure you are paying attention and giving honest answers. The tests are much the same as psychology/business school tests at Stanford but I remember getting paid more for my time.

I think there are some more transcriptions work that pay a lot more money, but I haven’t tried them yet. The psych surveys I think are the highest paying ones you can do without special qualifications. There are some interesting transcriptions in Czech or French or other languages.

So these are my first impressions. If I’m free or absolutely bored I’ll continue with it. But I feel like it’s just not worth my time. Maybe when this system was first starting there were more interesting or profitable tasks. But now it kinda feels like it’s super-segmented or bargained down, it needs more life more activity to be a true ecosystem. It’s true these tasks are really “bottom of the barrel” work. At least they are somewhere to start from. I found myself moving from 1) adjusting/understanding the work 2) pumping through as many as possible 3) happy to easily earn money 4) really desiring to do more high quality work for more money. All this cycle moved rather quickly then I got stuck at number 4.

I tried to look at odesk.com and elance.com a year or two ago but I didn’t do any work for them. I found the problem there to be something quite different. There you couldn’t get any work unless you had lots of experience with the platform or strong recommendations. So it is chicken and the egg problem “can’t get a job without experience, can’t get experience without a job”. There was no way to ramp up, unless you uploaded your entire life/work history or bid the project cost down to extremely low levels. There are “tests” or qualifications you can take on all these platforms to get higher quality work. But they didn’t help me on odesk/elance. A little bit on mturk.

So so far it seems like the ramp-up problem has not been solved by any of these companies. The metaphor is more like a “cliff” rather than a “ramp” or at the least “stairs”. Ideally if you had enough jobs, requesters, employees, and companies on the system it would have enough for everyone to be perfectly matched with each other. Then it would much the same as the job market in the real world. But as the subset is finite, there are inefficiencies in the system that lead to idle workers and unfulfilled tasks/jobs. As such, the system is not designed well enough to teach companies/employees to overcome these obstacles. The system should not only solve the problem for 80% of the people but also work to overcome the 20% inefficiencies by 1) pressuring companies to pay more and get more applicants 2) pressuring companies to accept lower quality work/workers temporarily and be willing to train them 3) encouraging and making easier education for workers, education that is directly tied to the work needed.

These recommendations seem to apply as much to the unemployment crisis as they do to the microcosm of Amazon Mechanical Turk / Odesk.