Domo Arigato Mr. Roboto

[da]

Imagine writing software that, say, needs to be able to identify whether a photo has Elvis Presley in it. Or whether the voice in a sound-clip is angry. Or does text recognition on scrawled handwriting. Amazon has released an API to make this possible via a web-service. So, your program calls this API (using whatever inputs you want, such as a question, or a picture) and it returns a value answer within some specified time period.

Cleverly, they named it "Mechanical Turk". It does precisely the same thing as the old-time pre-robotic chess-player, which was a fake, done with a small person hidden inside the console. When I came across this, I was convinced it was a practical joke. Like Google's pigeon-ranking algorithm from 1 April 2002. But no, this seems to be legit. People post a bounty (seems typically a penny) for others to claim by performing a "Human Intelligence Task" (as opposed to an artificial intelligence task). The requester can require particular qualifications, and they have the right to refuse bad answers.

Right now most of the questions are surveys paying a penny, though there are some transcription tasks that pay approximately $10/hr. This has fairly exciting (but scary) possibilities for outsourcing work across the globe. On the one hand, it literally puts people into an inhuman job, doing work for computers. At the same time, people have done menial jobs forever; and companies have been doing similar outsourcing over the internet for a while. This just brings it down to smaller companies, or anybody with a bit of cash and a computer program.

This development doesn't surprise me much; my ex-business-partner and I wrote a similar service in 1996, which we gave up on because we didn't have the means to make it big enough to matter. But at the same time, even though it's an obvious development, it creeps me out a bit that a big company is actually doing it.

It's also funny to me that this could directly be used to countervent "are you a human" tests.

Comments

[melted-snowball.livejournal.com]

As I mentioned the last time you talked about Captchas, I really think it's funny that this is the only one of Manuel Blum's brilliant ideas that was de-fanged by globalism, of all things.

[dr-tectonic.livejournal.com]

Is it really such an inhuman task, though? It's an abstract, dissociated task, sure, but frankly, it sounds about a zillion times more interesting than washing dishes or running a gas station.

I had a job editing voice files for phone-mail systems. So, so boring. It was incredibly repetitive. At least working for Mechanical Turk would give you a whole lot of variety...

[da-lj.livejournal.com]

Yeah, if they had enough variety, it could be interesting. "Find Waldo." "read this note." "which tie looks better on me?" At least until the ten-thousandth one. On the first day. :)

I have this weird fascination/repulsion with repetitive tasks. Sometimes, I can really get into it, and wash close to an hour's worth of dishes thinking about nothing and feeling great at the end. Probably has something to do with my history..

I worked at my parents' farm from about age 12, packing maple-syrup into small jugs. It, too, was mind-numbing. I totally lost count of the number of different ways I re-designed the process in my head to be more efficient; though all depended on more machinery, which my dad would never buy (or let me make).

As a kid, voluntarily, I turned myself into a data-entry droid, typing in dozens of Commodore-64 programs in assembly language. I still can't believe I did so many of those.

...I enjoyed the video/audio editing work I did as a summer job at school, but that's because it was a research project and involved spoken and written Arabic, which was just super-cool.

[dr-tectonic.livejournal.com]

Yeah, if they had enough variety, it could be interesting. At least until the ten-thousandth one. On the first day. :)

Sure, but the thing is, the selection process is entirely voluntary, right? You don't get assigned tasks, you pick one. So it's only as much drudgery as you're willing to put up with, no?

And it opens up an interesting free market on the tasks; those that are inherently interesting will get dealt with quickly, for low cost. People who have very dull tasks will need to either pay well or put up with very long response times...

[da-lj.livejournal.com]

On the free-market angle, I'm thinking of a cube-farm where people get paid like phone-jockeys, on the order of $100 a day. That's more or less what you could make if you answered 1-cent ones for 8 hours... I wouldn't be surprised if that's the direction the market went, like it has for phone-answerers...

*shudder*

[dr-tectonic.livejournal.com]

Right, but why would you do it in a cube-farm when you could do it freelance from home and not pay any overhead to The Man? Sure, you have to have a computer and a net connection, but if it pays well enough, that's a small capital investment.

Maybe I'm missing something about how the API works, but it sounds like Amazon will pay anyone in the world who answers the question, so there's no need for a middle man or centralization, if you want the money, you can just DO it.

[da-lj.livejournal.com]

Yeah, I don't want to say that that /won't/ happen; it would even be more efficient, since there's no commuting.

I distrust The Man enough that I woudn't be surprised at that development, though.

[melted-snowball.livejournal.com]

I would also wonder how interested the IRS will be...

[da-lj.livejournal.com]

good point. Though, I suppose I don't know what the tax and labour laws are for hiring wage-slaves on the other side of the world...

[pnijjar.livejournal.com]

I think this is my dream job. Seriously. I hear you about the sweatshop possibilities, but doing Internet scavenger hunts is one of my favourite things, especially when there are puzzles involved. I could sit in my room and eat bon-bons and surf the Internet, and get PAID FOR IT! Eventually AI will get good enough that I will be out of a job, but until then it looks like a sweet deal.

[da-lj.livejournal.com]

Fair enough. I mean, I did think about starting to work on one of the scavenger-hunt-like questions, just out of fun.

Eventually, AI will get good enough...

Haven't they been saying that phrase for the last 30 years? :)

In the meantime, bon-bon away.