The African workers driving the AI revolution, for about a dollar an hour
theguardian.com
theguardian.com
Training AI is also way better than undermining Western democracy... [2]
[1] https://en.wikipedia.org/wiki/Dual-sector_model
[2] https://www.theguardian.com/technology/2020/mar/13/facebook-...
Much of this work does not suit any living human.
https://www.ft.com/content/ef42e78f-e578-450b-9e43-36fbd1e20...
https://time.com/6247678/openai-chatgpt-kenya-workers/
https://www.theguardian.com/technology/2023/aug/02/ai-chatbo...
Since these workers are not indentured serfs, we should generally rely on their ability to switch jobs. To me, this looks like some reporters hounding in on anything that smacks of big-bad-wolf AI.
As for it being work that doesn't suit any living human, it seems a bit hyperbolic. Maybe you've been less exposed to reports of what life is like in the developing world than I have? Here's something that was on PBS the other day: https://www.hrw.org/news/2015/09/29/philippines-children-ris... Of course it's not a company in the developed world that commissions this, but everyone knows whose necklaces the gold is ending up in.
I think we should avoid knee-jerk reactions against AI, which arguably has more potential to help the developing world than anything since electricity, and maybe look more into who benefits if AI is held back.
I don't find it hard to believe that watching mutilations, rapes and and murders all day long will take its toll. It sounds like you do. American moderators have suffered too. https://www.theverge.com/2020/5/12/21255870/facebook-content...
> Since these workers are not indentured serfs, we should generally rely on their ability to switch jobs.
So they can be replaced by those with fewer career choices?
> Maybe you've been less exposed to reports of what life is like in the developing world than I have? Here's something that was on PBS the other day
So because children are dying while mining for gold in the Philippines, we should be quiet about a traumatized AI workforce in Africa? I do not buy this reasoning.
> I think we should avoid knee-jerk reactions against AI
It's not just knee jerks though. I'd wager PSTD is real. The insane energy use is definitely real and the hallucinations are too real. The predictions for constant future improvements are however highly questionable. Critique where critique is due.
The valley hype machine could possibly be racing towards the same cliff as it did with the metaverse and NFTs not long ago.
By those less squeamish.
> we should be quiet
This is a strawman since that's not what I said.
> traumatized AI workforce
We only have data points on a few plaintiffs and plaintiff-adjacent workers, hardly enough to assume they're necessarily more than a drop in the workforce bucket.
Here's a show full of shittier US jobs like noodling: https://go.discovery.com/show/dirty-jobs-discovery-atve-us
> hallucinations are too real
> metaverse and NFTs
It sounds like you have an axe to grind with the Valley and tech in general.
https://cc.bingj.com/cache.aspx?d=5059020920596126&w=O_9GpTf...
$1 per hour is around market rate for this kind of a role (some kind of college education).
This kind of BPO work is how Kenya (and Uganda - they're using the same model) graduate up the services ladder.
Edit: these are employees in a Tier 4 city in Uganda. $200/mo is a great salary in a rural town like Gulu.
[0] - https://reall.net/wp-content/uploads/2022/12/Understanding-H...
AI being trained for content moderation for social media...
Social media seems to be downstream from training AI to recognize and categorize content.
Meanwhile the only connection to AI is that it might provide a potential reduction in the need for this.
So what is odd to me is positioning this as the purpose being for feeding the ever hungry data needs of evil algorithms, because as far as I can tell without those algorithms these jobs would still exist and would be worse.
Writing about training a content moderation AI by humans marking metadata on the content for AI seems to be throughout the article.
“Might” provide a potential reduction feels like a stretch. LLMs literally are decent at receiving input and prompting “tell me about this”.
There is a less than desirable human price being paid to encode content to train an algorithm. Maybe if it was done by someone with tech skills they could reduce it.. instead, for now, it appears to be outsourced to traumatize people who don’t have the health support and salaries to build the AI to begin with.
Someone is choosing to send this work far, far, away to a much more vulnerable population, and maybe in time it doesn’t have to be that way.
If any group is capable to make a difference in reducing any questions about this approach it’s the most capable in the tech, not the least.
Content recognition is something AI done before the current wave of AI pretty well. If anything this might boost things like OpenCV.
Training a content moderation AI is meant to do what social media hasn’t been able to do, moderate content in large ways other than using humans.
AI a popular word right now in zeitgeist, it doesn’t meant it’s beyond being connected to everything that already exists, when it’s based on existing knowledge and practices.
The actually objectionable bit is that these jobs traumatize people over time, and desensitize them to the disturbing content they watch as a part of the job. Would I choose this over spending my day out in the heat, trying to sell food, snacks, or other wares, subject to the risk of pedestrian accidents or violent theft? Maybe?
Ideally there could just be better jobs available all around, but some people who live there might consider this at least somewhat better than some of the alternative jobs they'd otherwise be working.
Our current way of building models requires a large training data set that is already tagged/classified by humans. Unless the technology somehow improves so this is no longer necessary, these jobs are necessary. One bright spot is that once these models are trained, then humans can be out of the loop, which is certainly better than not having the models at all, and having to employ people, in perpetuity, to watch all of this content and directly make moderation decisions.
But perhaps the question we should be asking is: should we just completely do without, and never subject people to these kinds of jobs, no matter what the purpose? Maybe video, photo, etc. sharing on this scale just shouldn't exist if we have to put humans through miserable work in order to do moderation. But I don't really see this changing. So if we continue to employ people to train our AI models, can we get to a point where our future models can be used to bootstrap/train more sophisticated models, without the need for further human classification of the training data?
20 eight hour days at 1 dollar is not much higher than Kenya’s minimum wage (around 120 a month)
Not terrible I guess but also not great even by 3rd world local standards
To make it count as decent you’d need to go somewhere poorer than Kenya
- Queues with gore and violations pay more than regular queue. You can be paid $X/hour for marking traffic lanes or roughly $1.5-1.8X for looking at somebody getting killed. This is personal choice and some people handle it better than others. If you have high empathy, you should absolutely not do it. On the other hand 5% of population are psychopaths with inhibited empathy and they do not care. This is similar to real life jobs - what you see in queues is not that different from what police see day to day in downtown SF or South Chicago.
- Same goes for sexually-explicit/porn queues. If you are 18 year old male you should absolutely not do them - it will fuck up your sexuality (similar to watching porn 8 hours/day). You will also burn out in a couple weeks anyway. The most success we saw was with 50+ year old women working from home - some worked on those queues for years.
- From the article you will notice that Africans worked on African countries queues/moderation - so it's not even imperialism or whatever where they worked on rating rich-countries content. They rated their fellow Africans.
- Africa is not that big in content moderation. Reliable coachable people are hard to find and infrastructure is not great. Salaries are also pretty high - I am not sure where they find workers for $1/hour but we paid significantly more. India does way more moderation, people are more reliable and overall it's much easier to work with Indian BPO.
Do you want psychopaths to be the ones to determine what content is too violent?
Acceptable/Unacceptable violence is clearly defined in guidelines with examples - no moral judgement from rater is required.
On raters - they are doing pretty mechanical work and we generally do robust quality checks. Some % of tasks are randomly sent to two raters and if there is disagreement, task is sent to auditor. Raters with high disagreement rate with auditor are coached and if disagreement rate contienues to be high, they are fired.
Overall quality is controlled by using "golden set" - tasks that have "right" answer decided by HQ. We look at disagreement between raters and right answer.
There are other ways to control for quality - this is all pretty standard and everyone is doing the same things. Does not really matter if rater is phychopath or not if he can follow guidelines.
This does happen. It is also rare, but mostly unavoidable because it can't be predicted. The best that can be offered is support for traumatized people. The support the article says was offered was clearly insufficient, but after a sensationalized opening such as that, I end up doubting the credibility of the article as a whole.
Worker welfare is an important issue, world-wide. I feel like books like the one this article comes from are just finding a new way to exploit these people.
They know their market.
Maybe it's surreal in part because it's so far removed from the end users receiving the benefits.
An uncle of mine died in a car crash, in the late 80s / early 90s. The first thing my father did was drive for hours, to the place where the crash happened, and film. He then went and filmed the car wreck with its blood stains and grass stuck between the tires and the rims. A 230 km/h // 140 mph+ crash: driver lost control.
And although he could barely speak while filming he warned us in the video that this was what would happen to us if we were to act like idiots on the road.
And he, of course, forced us to watch that vid. Immediately and then a few more times at regular intervals months/years later. Best road safety lesson we ever got.
Every single person I ever crossed on the road in my life can thank my father for filming and forcing me to watch that vid.
At times there are very smart families who do allow the publication of videos of crashes where a family member died and say it's to act as a warning as to what may happen when you drive drunk/intoxicated/crazy.
It should be outlawed that families may decide that these vids are not publishable. It happened on public roads: it should be public material, as a warning to the public.
That these are not allowed on social media is criminal for these vids are the best lessons kids / teenagers can ever get when it comes to road safety.
And it is delusional to think that kids/teenagers watch that for morbid reasons: they watch such vids because their survival instinct want to make sense of what's happening.
"If you act like that, you'll kill people and/or die" is quite the lesson.
What maddens me about social media is that they're OK hosting videos of their users driving recklessly on public roads and getting away with it. It normalizes dangerous driving, even glorifies it since these idiots attract millions of followers. Instagram for example doesn't even have a category to report this content.
Most don't, but there's apparently a very small subgroup that feeds on this kind of content. Algorithms can detect that of course and feed them even more of these videos, getting them even deeper into the rabbit hole.
You can't blame it all on the algorithms either, if you build an algorithm that finds relationships like "a likes x, b likes x and y, recommend y to a", this is what you get in some pathological cases, along with pedophiles seeing family videos of kids in swimming pools.
And this is the kicker: it's all very well for us to wring our hands about how their conditions could be better, but people are voluntarily choosing to do jobs like this because it's still better than the alternatives. Staring at video footage for 8 hours a day to get paid a dollar an hour sucks, but it sucks less than walking around in a traffic jam in the African sun, trying to hawk bags of popcorn to drivers and constantly risking getting run over with no certainty of getting paid.
Pity or what have you aside this is a massive L no matter how you take it
If developed countries all refused to outsource work like this to poorer countries, do you think that would help the condition of the people there?
Don't get me wrong, I'm all for better working conditions, but ultimately that is primarily the responsibility of the government(s) of that country.
Maybe colonialism has something to do with it, hasn’t it?
'The exit came as the trio shifted away from former colonial ruler France, with Tiani calling for the new bloc to become a "community far removed from the stranglehold of foreign powers." '
https://www.voanews.com/a/sahel-military-chiefs-form-confede...
You do realize a great deal of the US meddling was prompted by USSR/Russian meddling, right? I’m not giving the US a pass by saying that, but the Cold War and threat of nuclear annihilation is a backdrop to many of these events.
I'm also not going to ignore the issues the countries within Africa deal with that have nothing to do with their past of imperialism or colonialism. I could point to India as a country that reformed into a modern successful nation post colonialism. Nor ought we ignore the vast amounts of financial aid given so often to these countries. It’s a tangled difficult problem.
I suspect you have more ire than foundation.
https://www.britannica.com/place/Kenya/Kenya-colony
What about the last 60 years free of colonial rule connects to the current state today? Do you think these corporate interests stem from Kenya’s Building Bridges Initiative?
I ask because I recently listened to a podcast about China’s Belt and Road Initiative winding down in Africa. I was surprised to hear this because they had committed to a large infusion of capital with some future trade partnerships (tens of billions at least). It was set to be a massive boon for much of Africa after the US reduced its economic aid in the region. Their reasoning was (economic woes in China aside), they found too much was unstable, specifically legislatively and socially. Permits and licenses drug out where most of the time it was difficult to know which official was in charge of which responsibilities. Also, they would build a facility that would be repeatedly attacked and destroyed by warlords in the area. I believe they said it was common to lose workers to independent gangs.
Just wondering if you could expand on what you said so I can connect the same dots. I’m not understanding your critique.
Nobody said anything about savages. Please don’t bring negativity into a conversation.
> alienate themselves
Sounds like a first-world problem to me.