Being on the bleeding edge of progress means you see new technologies come out all the freaking time. Some of them are truly worthless and can be safely ignored. Most of them will be intriguing but ultimately what you would expect. A very few of them have the potential to surprise the hell out of you and those are the ones worth keeping an eye on.

About 20 years ago, the surprising thing of the day was using commodity hardware to build supercomputers. Before that point, the way to make supercomputers better was to utilize every hertz of processing power through custom hardware and clever software. The revolution of commodity hardware was not in the engineering, it was in the shift in thinking. The new way to solve hard problems was to just design simple, less efficient algorithms and throw more hardware at the problem. That shift changed not only the types of applications that could be built but also the way we think about building apps. The reason Mechanical Turk is worth keeping an eye on is because its about to do something similar for labor.

When Amazon released its iPhone app my entire understanding of what was possible changed. You load up the app, snap a picture of an object, Amazon will use Mechanical Turk to find the closest Amazon equivalent and, within about 5 minutes, you can buy it for one click. The application itself was a beautiful usage of Mechanical Turk but more interesting is how a shift in thinking had to occur before it could even be imagined. That Amazon is releasing this app for free but paying for human labor means  their business model relies on human labor being cheap enough to hide in the margins. At the same time, the user experience is only compelling because the search results come back before you’ve left the store so Amazon needs to assumes the pool of available labor as essentially infinite to deliver that experience.

Once you’re able to get over that hump of believing that human labor can only be an expensive, limited resource, an entire vista of compelling applications open up. Here’s one I came up with today in a conversation with a collegue: Calorie tracking sucks because of the data entry problem. You need to manually enter in every single thing you ate and that requires far more organization than most people have. Why not just snap a photo with your iPhone and let a Mechanical Turker figure out what you ate? How do you solve the reliability problem? Have every picture looked at by at least three Turkers and only accept it if at least two agree. When labor becomes that cheap, its smarter to be dumb and throw more human hardware at the problem.

Does that mean Mechanical Turk will do to human labor what the commodity hardware & cloud computing did to server farms? Of course not, the analogy is instructive, not a direct mapping. What it does mean is that we as a society are going to experience several “everything we knew was wrong” type moments and that the labor market of 2039 will look as different from today as supercomputers did in 1979 and those who are the first to recognise this change will be the ones who have the best chance of exploiting it.

  • David Jones

    Your Calorie tracking example illustrates that Mechanical Turk relies heavily on tremendous inequities in wealth.

    The rich, fat, lazy person snaps a picture of his lavish meal with a fancy iPhone, and rather than take a brief moment to ponder the calories, … uploads it for, not just one, but *three* other people to ponder and estimate the calories and report back. The most dissimilar answer is rejected; that person is not compensated.

    This only works if there is a very large pool of very poor people who are willing to do these mundane tasks for (fractions of) pennies. These “workers” certainly can’t afford an iPhone; they probably can’t even afford the meal they are looking at.

  • http://Website David Jones

    Your Calorie tracking example illustrates that Mechanical Turk relies heavily on tremendous inequities in wealth.

    The rich, fat, lazy person snaps a picture of his lavish meal with a fancy iPhone, and rather than take a brief moment to ponder the calories, … uploads it for, not just one, but *three* other people to ponder and estimate the calories and report back. The most dissimilar answer is rejected; that person is not compensated.

    This only works if there is a very large pool of very poor people who are willing to do these mundane tasks for (fractions of) pennies. These “workers” certainly can’t afford an iPhone; they probably can’t even afford the meal they are looking at.

  • Hang

    There’s a common perception that Mechanical Turk thrives on exploiting 3rd world labor. This doesn’t appear to be the case.

    From:
    http://behind-the-enemy-lines.blogspot.com/2008/03/mechanical-turk-demographics.html and
    http://asc-parc.blogspot.com/2008/07/mechanical-turk-demographics.html
    82% of turkers are from the US, Canada or the UK and over 75% have a Bachelor’s degree or higher.

  • http://www.bumblebeelabs.com Hang

    There’s a common perception that Mechanical Turk thrives on exploiting 3rd world labor. This doesn’t appear to be the case.

    From:
    http://behind-the-enemy-lines.blogspot.com/2008/03/mechanical-turk-demographics.html and
    http://asc-parc.blogspot.com/2008/07/mechanical-turk-demographics.html
    82% of turkers are from the US, Canada or the UK and over 75% have a Bachelor’s degree or higher.

  • FutureWork

    So, is amazon also keeping track of the pics and the classifications for the purposes of training machine learning algorithms in the future?

  • http://Website FutureWork

    So, is amazon also keeping track of the pics and the classifications for the purposes of training machine learning algorithms in the future?

  • Radu Floricica

    @David Jones

    Nope. It simply means the efficiency of counting calories for your meal, while eating it, and counting calories for a living are extremely different.

    I can easily imagine having a calorie-counting job, where I can look at a photo, recognize the ingredients, maybe receive a total portion weight along with the photo and make an estimate, all in less then a minute. Probably a lot less, because of the 80/20 rule (80% the pictures will be common meals, for which I’d know the contents by heart).

    So 1 minute per photo, 60 minutes pe hour – easily $30 per hour. Not exactly you average third world income, is it?

  • http://Website Radu Floricica

    @David Jones

    Nope. It simply means the efficiency of counting calories for your meal, while eating it, and counting calories for a living are extremely different.

    I can easily imagine having a calorie-counting job, where I can look at a photo, recognize the ingredients, maybe receive a total portion weight along with the photo and make an estimate, all in less then a minute. Probably a lot less, because of the 80/20 rule (80% the pictures will be common meals, for which I’d know the contents by heart).

    So 1 minute per photo, 60 minutes pe hour – easily $30 per hour. Not exactly you average third world income, is it?

  • Hang

    @FutureWork: I’m sure they store the images but I don’t think they have any immediate plans to provide any machine learning. It’s a hard enough problem and MTurkers do a good enough job that there’s no immediate incentive.

  • Pingback: links for 2009-04-23 « Blarney Fellow

  • http://www.bumblebeelabs.com Hang

    @FutureWork: I’m sure they store the images but I don’t think they have any immediate plans to provide any machine learning. It’s a hard enough problem and MTurkers do a good enough job that there’s no immediate incentive.

  • Janet

    You should see Lilly Irani’s work on Mechanical Turk at http://www.differenceengines.com/ and http://turkwork.differenceengines.com/blog/
    Especially check out the Haikus.

  • http://Website Janet

    You should see Lilly Irani’s work on Mechanical Turk at http://www.differenceengines.com/ and http://turkwork.differenceengines.com/blog/
    Especially check out the Haikus.

  • http://brynnevans.com/ Brynn Evans

    @David Jones

    Paying a Turker for their labor and accepting their work for your own purposes are completely separate (to me). I pay everyone for their work (except for blatent scams), even if their work wasn’t quite what I was looking for.

    Also, in my many adventures on MTurk — I can attest to their caliber. Not only are they not 3rd world (I’ve paid IRS agents, United Nations employees, scientists, etc.), they are quite considerate and diligent.

  • http://brynnevans.com Brynn Evans

    @David Jones

    Paying a Turker for their labor and accepting their work for your own purposes are completely separate (to me). I pay everyone for their work (except for blatent scams), even if their work wasn’t quite what I was looking for.

    Also, in my many adventures on MTurk — I can attest to their caliber. Not only are they not 3rd world (I’ve paid IRS agents, United Nations employees, scientists, etc.), they are quite considerate and diligent.

  • Pingback: links for 2009-04-24 « My Weblog

  • http://postlinearity.com gregorylent

    eye-opening post, thanks

  • http://postlinearity.com/ gregorylent

    eye-opening post, thanks

  • Pingback: Josh Mather » Bookmarks for April 23rd through April 28th

  • http://ageofglobalization.blogspot.com/ Derrick

    This is my first introduction into the MT. Having looked at it, I would mention that if the service were more widely used-in relation to say, the number of people who shop at Amazon-then it might be more difficult to find people who are able/willing to work as MTs. As the service scales up, its demographics, and how it works, will change. That said, it makes sense that most people would be in the US/Canada, we have VASTLY more people on computers than most other places, are at the cutting edge of services like this. Also, there is a big difference between MTs for products, I would imagine, than MTs for food simply because there is more access to profit through products by sellers becoming/hiring MTs.

    All that said, there is a good chance that this could provide some insight into the future organization of the labor market, although I would assume it will end up being something like a MT version of call centers.

  • http://ageofglobalization.blogspot.com Derrick

    This is my first introduction into the MT. Having looked at it, I would mention that if the service were more widely used-in relation to say, the number of people who shop at Amazon-then it might be more difficult to find people who are able/willing to work as MTs. As the service scales up, its demographics, and how it works, will change. That said, it makes sense that most people would be in the US/Canada, we have VASTLY more people on computers than most other places, are at the cutting edge of services like this. Also, there is a big difference between MTs for products, I would imagine, than MTs for food simply because there is more access to profit through products by sellers becoming/hiring MTs.

    All that said, there is a good chance that this could provide some insight into the future organization of the labor market, although I would assume it will end up being something like a MT version of call centers.

  • jhuff

    Great analysis of an emerging decentralized workforce. This site has some inspiring stories and windows into the lives of people who Turk for a living.

    http://mechanicalturkdiaries.com/

    The Mechanical Turk Diaries – the voice of Amazon's anonymous workforce. Unedited memoirs from Turkers.

  • staffing1

    Great analysis of an emerging decentralized workforce. This site has some inspiring stories and windows into the lives of people who Turk for a living.

  • http://www.usedtransmission.org Used Transmission

    This is cool! And so interested! Are u have more posts like this? Plese tell me, thanks

  • http://www.700r4transmission.org 700R4 Transmission

    This is cool! And so interested! Are u have more posts like this? Plese tell me, thanks

  • http://www.used-auto-transmissions.com/ Used Transmissions

    Nice……………………..

  • Pingback: Once you’re able to get over that hump of believing that human labor can only be an expensive, limited resource, an entire vista of compelling applications open up. In 2002 I flew to Shanghai and sat beside a Chinese entrepreneur. He told me that if you

  • Pingback: Mechanical Turk changes how we understand labor | Kareem Mayan Blog