Inside the AI factory: The humans that make tech seem human

A several-second blip of footage took eight hours to annotate, for which Joe was paid about $10. [...] Joe could make four times as much running an annotation boot camp [...]

$10/8hours is $1.25/hour - minimum wage in Kenya is 15,201 Kenyan shillings per month equal to $108.42/month, or 160 hours --> $0.68/hr. At four times as much, $5/hr, this job pays 7x the minimum wage, which in my neck of the woods would equal $117/hr. I would do that "task". Caveat follows.

Mechanical Turk

I remember signing up for that and being appalled by how insultingly little it paid. I remember thinking it was for something like training data, but more imagined that it was used to validate other human actions rather than machine learning.

In 2018, an Uber self-driving test car killed a woman because, [...] it didn’t know what to make of someone walking a bike across the street.

reminds me of when "Marines fooled a DARPA robot by hiding in a cardboard box while giggling and pretending to be trees"

categorizing the emotions of people on video calls

I'm sure they have hours of me sounding "irate", but hopefully not "hysterical"

Annotators are warned repeatedly not to tell anyone about their jobs, not even their friends and co-workers

Are there ethical standards to what these humans are tasked to label? If they can't talk to anyone, is anyone going to be able to tally the mental and moral harm done to worked told to "count the number of [atrocities] happening in this video"?

Consequently, there are no granular estimates of the number of people who work in annotation, but it is a lot, and it is growing. A recent Google Research paper gave an order-of-magnitude figure of “millions” with the potential to become “billions.”

pardon, is that billions of human beings or number of tasks? They're talking about a huge percentage of the world's humans doing this isolating and meaningless "non-work"?

To access a paying task, I first had to complete an associated (unpaid) intro course.

Also, how dare they use our unwilling hours of doing captcha for profitable machine learning.

After an embarrassing amount of trial and error, I made it to the actual work, only to make the horrifying discovery that the instructions I’d been struggling to follow had been updated and clarified so many times that they were now a full 43 printed pages of directives

And I thought Papers, Please was dystopian. At least the border agent "only" had several pages of arbitrary and daily updated instructions. The time required to sit and read and internalize through those 43 pages were unpaid, no doubt.

Every question must be answered, and a wrong guess could get you banned and booted to a new, totally different task with its own baffling rules.

horrifying abolition of man. Reminds me of the "everything is just slightly weird", humanity destroying room described in C S Lewis' That Hideous Strength:

To sit in tie room was the first step towards what Frost called objectivity — the process whereby all
specifically human reactions were killed in a man so that he might become fit for the fastidious society of the
Macrobes. [...] the very same initiation through which they themselves had passed and which had divided them from humanity, distending and dissipating Wither into a shapeless ruin while it condensed and sharpened Frost into the hard, bright, little needle that he now was. [...] He saw no possibility of leaving Belbury alive unless he allowed himself to be made into a de-humanised servant of the Macrobes.

By the beginning of this year, pay for the Kenyan annotators I spoke with had dropped to between $1 and $3 per hour. [...] “The question is, Who bears the cost for these fluctuations?” said Jindal of Partnership on AI. “Because right now, it’s the workers.”

Here's the caveat. Once they no longer need you, they just keep lowering the price until the human figures out he/she has been cheated of hours of their lives for nothing. If one doesn't figure it out, the algorithm is happy to keep getting data for free.

“It’s a culture we have developed of helping each other because we know when on your own, you can’t know all the tricks,” he said.

Ahh the beauty of humanity shines through the trappings of the de-humanizing room.

“I remember that someone posted that we will be remembered in the future,” he said. “And somebody else replied, ‘We are being treated worse than foot soldiers. We will be remembered nowhere in the future.’ I remember that very well. Nobody will recognize the work we did or the effort we put in.”

Just like the Human Computers! Even if we remember them in abstraction, that's not at all the same as remembering them as human members of society who (hopefully) made all our lives better with AI, and not just made billionaires even richer.

Last year, Surge relabeled Google’s dataset classifying Reddit posts by emotion.

OH yeah that's not good. I think I deleted my comments too late. And obviously they're going to be mining Tildes as well for "insightful" high quality human content.

one AI training the other.

I love the way this article ended. It's so cunning! A beautiful kiss * chef's kiss * At the same time when American lawyers are caught using AI to fake their hundreds-of-dollars-per-hour work, we've got the lowest tier workers faking their human pennies work. This is how we will triumph: together, and by circumnavigating around those who would exploit us.