AI's elves live in Kenya and twiddle your captcha
November 5, 2018 7:16 AM   Subscribe

When Artificial Intelligence works as intended, Silicon Valley types often say it's "like magic". But it isn't magic. It's Brenda, a 26-year-old single mother. In her eight-hour shift, she creates training data. Information - images, most often - prepared in a way that computers can understand
posted by infini (30 comments total) 32 users marked this as a favorite
 
That was interesting. The article wasn't really about AI at all-- it was about the impacts of an American company outsourcing work to Africa. The surprise (at least for me) was there seemed to be more positive impact than negative.
posted by gwint at 7:32 AM on November 5, 2018 [2 favorites]


I have a relevant meme!
posted by chappell, ambrose at 7:44 AM on November 5, 2018 [13 favorites]


Kinda like how Stitch Fix hired a bunch of stylists (women) to personally pick out out fashions for people then on the backend used that to train AI. Which is a bunch of men.

Kinda like how computational technology advancements have been taking advantage of women's labor since literally day one.
posted by nikaspark at 7:44 AM on November 5, 2018 [29 favorites]


"Samasource targets those currently earning around $2 a day, or less, in the so-called informal economy of odd - or dangerous - jobs. Samasource instead provides a living wage of around $9 a day. That's an improvement, but still a pittance for Silicon Valley."

"Yes, it's cost effective," Janah said. "But one thing that's critical in our line of work is to not pay wages that would distort local labour markets. If we were to pay people substantially more than that, we would throw everything off. That would have a potentially negative impact on the cost of housing, the cost of food in the communities in which our workers thrive."

Give me a fucking break, what a sick excuse to exploit your workers.

"None of the staff we saw at the office had any kind of acceptable ergonomic support, often crouching over, clicking away furiously, for hours on end - a certain strain to eyes and body. The company has said it would work on that."

If you refuse to pay them fair wages, at least give them a good working space. Fuck this company.
posted by GoblinHoney at 7:45 AM on November 5, 2018 [28 favorites]


And, as my Kenyan tweeps are asking, what happens to Brenda when this project ends? Can she go back to selling veg in the informal economy after this training and this stint?
posted by infini at 7:49 AM on November 5, 2018 [1 favorite]


Wait, the company is paying 4 times what many people were making previously and the worst thing about the job is a lack of ergonomic chairs? And most workers go on to higher education? This is sick exploitation?

I'd really like to hear more directly from the Kenyan perspective on this.
posted by gwint at 7:57 AM on November 5, 2018 [13 favorites]


This kind of manual classification is happening in a lot of cheap labor environments. Mechanical Turk, for one. Offices in India, for another. Justin O'Bierne's latest tour-de-force cartography dissection spends a lot of time trying to figure out if Apple has some huge new image recognition algorithms for making maps, is relying on humans to hand-trace features, or some mix of both. Part of what's nice about machine learning techniques is you can start with hand classified in one area, then train the system to apply the classification globally. There's a smooth path from 100% hand-classified to 100% machine-classified.
posted by Nelson at 8:01 AM on November 5, 2018 [5 favorites]


But much of Africa has leapfrogged the desktop PC era.

Sure, just like much of Kibera has leapfrogged the having clean water era.
posted by sfenders at 8:10 AM on November 5, 2018 [7 favorites]


I'd really like to hear more directly from the Kenyan perspective on this.

"training data sweatshops"
posted by infini at 8:27 AM on November 5, 2018 [6 favorites]


I highly recommend you reach out to any of the people I've linked to, particularly the Kenyan women, and ask them directly. African twitter is way more conversational and willing to interact than developed continent twitter.
posted by infini at 8:29 AM on November 5, 2018 [5 favorites]


Sure, just like much of Kibera has leapfrogged the having clean water era
posted by infini at 8:30 AM on November 5, 2018 [1 favorite]


Oh wait, I'm sorry for thread hogging, I forgot I'm the OP, outta here.
posted by infini at 8:32 AM on November 5, 2018


Wait, the company is paying 4 times what many people were making previously and the worst thing about the job is a lack of ergonomic chairs? And most workers go on to higher education? This is sick exploitation?

Yes, because exploitation is less about the behavior of the exploited, and more about the behavior of the exploiter. If you could easily provide for your workers' well being and you don't because you don't have to, that's exploitation.
posted by NoxAeternum at 8:32 AM on November 5, 2018 [10 favorites]


Fuck this company

It's a nonprofit, and it's effectively moving people out of poverty. A book by the founder, Give Work: Reducing Poverty One Job at a Time, is a good read for those who want to learn more.
posted by pinochiette at 8:35 AM on November 5, 2018 [4 favorites]


I'd really like to hear more directly from the Kenyan perspective on this.

Me too. This is a complex topic. I wonder how many people commenting here have ever been to a "slum" / informal settlement?
posted by JamesBay at 8:37 AM on November 5, 2018


infini, you're certainly not thread hogging! Thanks for those links. Twitter is so terrible in so many ways but the ability to give unfiltered voices to humans half a world away might just be its saving grace.
posted by gwint at 8:41 AM on November 5, 2018 [2 favorites]


The company is a non-profit and posts its financials online, although they have haven't published 2017 yet.

Anyway, in 2016 the company had revenues of US$12M (see pp 16-17); about $1.7M of that was eaten up by management overhead ($880K) and fundingraising ($800K). The rest was devoted to "program services", which presumably is wages for the "elves" around the world.

So, that ratio of revenue to management expenses is not bad at all for a non-profit. Using conventional, back of the napkin HR accounting, that $880K would employ 8 FTEs in the US, although per capita that would be less than a living wage in San Francisco, where the company is located. And they probably have more than 8 US-based staff.

Not bad for a non-profit, and they do seem to be leveraging program funding to help a lot of people.

FWIW, Nairobi has a big tech startup scene; it's the tech hub of eastern Africa and the western Indian Ocean. I've been to tech events in Colombo and the Philippines, and there is always a large contingent of entrepreneurs from Nairobi.

Exciting times, and I think that some of these employees will make the leap into that growing tech community.
posted by JamesBay at 8:52 AM on November 5, 2018 [4 favorites]


pinochiette, with all due respect, its that exact thinking that makes us (out in the boondocks of developing world) cringe.
posted by infini at 8:58 AM on November 5, 2018 [3 favorites]


One of the best people to talk to would be Nanjala - just tell her Granny sent you.
posted by infini at 9:02 AM on November 5, 2018 [1 favorite]


This article reads like it was paid for by Samasource. Which is possible! But also possible it just reflects the author's mostly positive bias on this topic.

It also seems likely, based on limited information available to me, that this is a good job relative to others available, and that we don't know the complex impact on the economy and culture of these jobs.

To me, this kind of assumption is a big stretch: I think that some of these employees will make the leap into that growing tech community. Basically, in the absence of a strong union, structural change, or strong government regulation, higher paying jobs go to people with more money - at least here in the US. I definitely don't know enough about Kenya to extrapolate my experience, but I've learned to be very cautious about US companies and non-profits and their mixed impact on developing and "second world" countries. Not to knee jerk assume this project is negative either. I imagine the impacts will be complex and mixed.

Having said all that, this Kenyan AI training mill would make a great setting for a novel.
posted by latkes at 9:50 AM on November 5, 2018 [1 favorite]




Kinda like how Stitch Fix hired a bunch of stylists (women) to personally pick out out fashions for people then on the backend used that to train AI. Which is a bunch of men.

I'm sitting in a WeWork about 30 feet right this second from a sign that says "Welcome to Stitch Fix", and (women) is only 50% right. Ok maybe it's 60/40 women.
posted by sideshow at 9:57 AM on November 5, 2018 [3 favorites]


I guess I just learned why captchas these days always ask me to identify squares with cars, trucks or traffic lights in them.
posted by snofoam at 2:25 PM on November 5, 2018 [1 favorite]


That is great to know about stitch fix. It’s been a couple of years since where I work was engaging their data science team to talk shop and it was always dudes from the data science team that I encountered.
posted by nikaspark at 3:07 PM on November 5, 2018


And just like Japan, eventually the local standard of living will rise until its priced out of competition with other low wage areas. Which means international wage arbitrage is jut an exploit with a finite pool of labor to exploit.
Then what?
posted by Fupped Duck at 4:01 PM on November 5, 2018 [1 favorite]


I'd really like to hear more directly from the Kenyan perspective on this.

I'm not Kenyan. But I lived in Nairobi for over half a decade. So I have non-Kenyan perspective.

When I got to my office on day 1, I had a list of Kenyan colleagues effectively queued up to offer me services, some of which I wasn't sure I needed. A gal in HR was going to connect me with local real estate agents, she wrote me an email like it was her job (turns out she just kind of made it part of her job and she was senior enough that nobody saw fit to question whether it should have been). A guy on my own team was ready to help me shop for and import a car from Japan, because that's where used cars in Kenya come from (and used cars are easily as much as, if not more in price point in Kenya than they are new in the US, if that gives you any idea what new cars cost in Kenya). One of the office gate guards asked me if I would be needing a driver (I drew the line here and drove myself, FYI). And a colleague on another team wanted to know when she could schedule her cousin to clean my apartment, once I moved in.

Everyone was looking for how they could benefit (or benefit someone in their family) on the side of a new mzungu colleague who made significantly more money than they did. It seemed pretty par for the course. In the case of the real estate agent, she took me around to a couple crappy looking places, then to one that I had found myself and really liked and wanted. When we got to that one she pointed out all kinds of problems (read: they weren't). I realized what was going on, she was looking at not getting her kickback from her realtor buddy who had set up the appointments with the crappier apartments. We didn't end up being the closest of friends.

With the maid thing though, I got a deep dive lesson in how paying someone drastically above market rates ripples out. I don't think I got nearly as deep a dive as I could have, but here's how it went down, in any case.

Christine, the tiny 20-something year old gal, was quiet, understood basic English but spoke it pretty sparingly. She came to my 3br / 3ba apartment 1 day a week and in about half a day she cleaned the whole place, did laundry and ironing, any dirty dishes sitting around, swept, mopped, made the beds, wiped down the windows, you name it. I never did laundry once when I lived in Kenya.

I paid Christine the equivalent of $80 a month for this service. So, if you want to be mad at people for using an excuse to not pay *more* - be more mad at me than at Samasource. I was paying Christine more than a Samasource person makes in a month of full time work, for working 4 half-days a month. I was paying her more than twice as much as a Samasource person makes. I also paid for healthcare for Christine's family. Healthcare is a cash business in Kenya, if you have the cash, you get the care. If you don't, no care. Nobody pays people for healthcare in Kenya, really.

Christine was terrified to tell me she was pregnant. Like, literally trembling and crying. I told her she could work until she was too uncomfortable and that I'd pay her sister to do the cleaning while she was on maternity leave (I had to explain the concept to her) and that I would pay her while she was on leave. Between that, paying for her to have her kid in a hospital (the first in her family to be born such, as I understood it) and a surgery that her kid needed when he was one year old, I maybe laid out another thousand or so bucks at most.

I came to learn that my Indian neighbors HATED me because of what I paid Christine. They paid their maids much less than half of what I paid her, and their maids worked 6 days a week, usually from 8am to 10pm or so. And in addition to that their maids did all the grocery shopping, cooked all the meals, and nannied all the kids in the house. All for way less than $9 a day. And those were GOOD jobs in Kenya, jobs that got you outside of Kibera and into a nice safe neighborhood for the day. The other maids in the building must have also loathed Christine for her high station - I can only imagine.

Christine, however, wasn't learning marketable skills beyond the cleaning she already knew. I set her up a few times with friends' maids who cooked for my friends, because my friends had taught their maids American recipes that smacked of home. So she got a little training in cooking, but she never learned how to use a computer as far as I know.

When we started packing up to leave Kenya, we gave Christine a lot of our stuff. Appliances, dishware, furniture, anything I couldn't be bothered to bring with us. I remember her being appalled by the fact that I wasn't planning on lavishing her with a large sum of cash as a severance payment of some sort. Apparently that's also par for the course, but I felt I had more than taken care of her. From her perspective, I guess, she (and by extension her extensive family) had become quite reliant on me as their source of income. The fact that I had set her up with a maid gig with some newly arrived expat friends didn't even garner much of a thank you, compared to the rage that I wasn't ponying up cash in addition to the new job and many gifts.

The general expectation we had with all of our Kenyan friends was that, for all intents and purposes, they looked at us as an ATM. Sooner or later they were going to come knocking for a withdrawal. Even in our church, with people we had into our home for community groups, we talked pretty openly about how if you could get something that helped you get ahead in life, it didn't really matter what rules you bent or broke to get it. Graft was just a part of the price of doing business, really. The same presuppositions applied to social relationships, not just business ones. I expect at the end of the day, Samasource is just creating unrealistic expectations on the part of their workforce. It's not like there's lots of other companies waiting to gulp these employees up at similar rates once Samasource moves on. I'd wager their employee retention rate is well above 95%.

As far as I know, there isn't any existing data that proves that paying a select group far above market rates has a net positive impact on a country's economy at the end of the day. Samasource could stop doing business tomorrow and 90% of Kenyans would still survive on less than $2 a day, just like they were years ago.

I guess at the end of the day, what would be better: pay many, many more people a lower wage that still is above average? Or pay a select few a wage that's many exponents of the average? Or do something in the middle and try to promote education and economic development in local communities? Which seems to be roughly what Samasource is trying to do? They're at least giving people work that also gets them a transferable skill. We can rage that they aren't paying them 10 times market rate and only 5 times, but at some point you are beginning to skew local economies and being less fair to those you aren't employing than you are to those who you are. They could be paying people twice as much as market rate and employing 3 times as many people, if they had enough work to go around.

It's pretty hard to make everybody happy at once. At least Samasource is doing something.
posted by allkindsoftime at 4:07 PM on November 5, 2018 [21 favorites]


And just like Japan, eventually the local standard of living will rise until its priced out of competition with other low wage areas.

Wages are so low in Japan that some manufacturing is moving back to the country.

On top of that, Japan has dealt with the problem of high labour costs by moving up the value chain and increasing productivity.
posted by JamesBay at 4:52 PM on November 5, 2018 [1 favorite]


somewhat related, the Deep Learning Indaba will be held in Nairobi at Kenyatta University next year.

they got some very, very big names in machine learning to give talks this year, and it looks like heavy sponsorship from Google; hopefully that trend continues.
posted by vogon_poet at 6:32 PM on November 5, 2018 [1 favorite]


Its hard being an ATM or being seen as a walking sack of grain. We think we have relationships, but we're prize goats, who're slaughtered at time of departure.
posted by infini at 2:00 AM on November 6, 2018 [3 favorites]


I recently read an article from China's GQ report on this subject, and the humans working for AI in this case are people who live in the provincial towns.

To quote parts of it:
"回乡后,她开过网吧,做过超市收银员。去年,她应聘了这份“电脑操作员”,每个月收入3~4千,比收银员高,比富士康低。她觉得工作环境不错:坐着用电脑,有空调,下班早,晚上回家还能陪孩子玩一会儿。而且,3个老板都是熟人,“我们村子都挨着,打小就认识”。"
They get paid about 500 dollars a month, which is more than a supermarket cashier's pay but lower than manufacturing job wages at Foxconn. One of the workers interviewed was quite positive about her job conditions: she gets to sit on the job, there's AC, and the work hours are not long. Her bosses are people who grew up from nearby villages (and thus inherently more trustworthy.)

但处于产业链末端的人,不思考这些问题。这离他们太遥远了。北京一位AI数据标注公司老板告诉我,“工人只是打开人家的网页,用人家的软件,在上面把人家的数据,按人家的格式给人家处理好,交给人家,公司接触不了人工智能任何东西,这个数据自己也保留不了。”“我们没有研发能力,纯粹也就是一个(代工的)富士康。”
These 'data marking' companies exist at the end of the product chain. They don't own any data or technology themselves, and are at best a Foxconn for AI.
posted by em at 1:54 PM on November 7, 2018


« Older GORMENGHAST: THIRST TRAP   |   high tech, low practice Newer »


This thread has been archived and is closed to new comments