Meta Just Bought Half of the Biggest AI Global Sweatshop
Their $15 billion stake in Scale AI is a bet that more global misery will enrich the AI oligarchs
Thereās a lot of fear out their that AI will eliminate jobs. The good news is AI has created 100,000 annual jobs for contractors hired by Scale AI, one of the companies that helps big AI companies ātuneā their AI algorithms. The bad news is those jobs are for the most desperate workers in the world in places like Kenya and Venezuela that pay on the order of $200 per MONTH in earnings.
Metaās Zuckerberg likes what he sees in Scale AIās sweatshop business model and last week dropped $15 billion buying a 49% share in the company, co-founded by 28-year old Alexandr Wang.
So what does Scale AI do?
AI companies have built their models ingesting massive amounts of public domain data - or some data that's questionably available to them legally - and then ātrainingā the programs on what answers to give using that data. Many people assume that those magical answers produced by prompts given to ChatGPT and other AI companies are the results of sophisticated programming skill by highly trained software engineers. And some of thatās true.
But much of the training is actually done by the contract workers hired by companies like Scale AI in developing nations who are fed results from the AI models and laboriously, line after line, tell the models whether the answer is appropriate or not. These contractors also look at raw data ahead of time and, for example with self-driving car AI models, characterize visual clues to help train the AI on which images are stop signs and which ones are children the car should avoid killing.
Other companies pioneered this kind of global cognitive outsourcing in earlier iterations of machine learning work, but Scale AI became dominant in the industry through one major innovation: figuring out how to pay exploited workers even less than the competition.
Karen Hao, author of the wonderful new book Empire of AI, has been following the industry for years, including as a reporter for MITās Technology Review, and summarizes Scale AIās entry into the industry this way:
Scale AI from the beginning followed a strategy that rested in part on its emphasis for providing specialized, quality services at a low priceā¦
The startupās worker-scouting teams searched for the areas in each country that struck the very same balance of factors that would converge in Venezuela: a high density of people with good education and good internet yet who were poor and thus willing to work hard for very little moneyā¦To support its expanding and diversifying client needs, it entered countries with large populations facing financial duress and who could also speak the most economically valuable languages.
Scale AI was notorious for offering what appeared to be attractive pay, relative to the disastrous alternatives these contractors faced, but then reducing the pay of contractors hired by more than a third within months. And they harshly punished any workers who challenged the rules Scale AI established:
When [a group of workers Hao interviewed] attempted to organize against the changes, the company threatened to ban anyone engaging in ārevolutions and protests.ā Nearly all who spoke to me were booted off the platform.
And as AI oligarchs like Sam Altman bought palatial estates in Hawaii, the sweatshop workers powering their AI models were living in complete squalor. Again Hao:
[W]orkers dropped me location pins on WhatsApp to specify where to find them in the thick of corrugated-tin neighborhoods. One worker named Oliver lived with his sister in a space no larger than one hundred square feet, paying for internet through his phone on a minute by minute basis.
Some may justify this kind of squalor and horrendous pay as what the market will bear in countries where alternatives may pay even less, but in an industry where multi-billionaires have been minted in a matter of a few years based on that global sweatshop labor, itās obscene.
And while the image is of those AI billionaires benefiting from genius programming skills or, at worse, plundering the intellectual property of the world, it is just as much built on this darker story of pure labor exploitation of workers in the poorest nations on earth.
While public data on its internals are only sketchily available, the best estimate is that Scale AI has hired something on the order of 500 million hours of labor on its remote platforms, with those contractors making pennies per task, adding up to often little more than a dollar an hour- and sometimes far less than a dollar per hour - and averaging on the order of $200 of income monthly for the 100,000 contractors managed by the company. (*See source below)
For AI companies, this means they can pay their privileged first world employees massive amounts, while the bill for this global sweatshop becomes a pittance in their balance sheets. OpenAI for example pays its 1500 regular employees $1.5 billion annually which is six times as much as the total of $225 million that Scale AIās workforce of 100,000 contractors are estimated to have collectively received in 2024.
With OpenAI also paying an estimated $4 billion for computing power in 2024, it could double or triple the pay of the global contractors it uses and barely notice the additional costs. But it chooses to hire the bottom-feeding Scale AI to exploit workers globally on its behalf.
Scale AI - now minted as a $30 billion company enriching its original shareholders like Alexandr Wang - is clearly one of the biggest scumbag sweatshop companies in the world. Meta is now half owner, making it also one of the biggest scumbag sweatshop operators in the world. But all of the AI companies whose AI models were built on the hundreds of millions of hours of work by Scale AI contractors should be recognized as the sweatshop employers they are as well.
Sam Altman, CEO of OpenAI, styles himself a philosopher and altruist, and just six days ago published an essay on his personal blog where he wrote:
AI will contribute to the world in many ways, but the gains to quality of life from AI driving faster scientific progress and increased productivity will be enormous; the future can be vastly better than the present.
Sam Altman is notably a promoter of the idea of a Universal Basic Income, a guaranteed income for every person, yet he canāt manage to pay the very contract employees who make his own wealth possible anything approaching decent pay even by developing nation standards.
You read Altmanās words and there is zero acknowledgement anywhere on his blog of the hundreds of millions of hours of sweat and labor by those contract workers in making his high-flown visions of AI innovations a reality.
I happen to think the technology behind current LLM AI models is exciting and worthwhile - despite many problems that need to be addressed.
But the current economic structure of AI oligarchic ownership of the technology built on global sweatshop labor exploitation and theft from the public domain promises not some bright future but a world of ever increasing economic inequality.
Mark Zuckerberg seems to agree with me and has placed a $15 billion bet than owning a large piece of Scale AI, the prime engine of global sweatshop misery, is a key to profiting from this AI future of oligarchic global exploitation.
NOTE on the source of estimates of Scale AIās contractor hours worked and pay. While I did some of my own estimates based on publicly available sources, I decided to ask OpenAIās ChatGPT for its analysis- and have used its estimates of pay and contractor hours worked throughout this article. These line up largely with my own calculations but for both āobjectivityā and the ironic value, Iāve used the ChatGPT numbers here.
Work done for Scale ID and its competitors to power AI models could be routine or occasionally even interesting, but chunk of it meant to help moderate āunsafeā content delivered the sewers of the Internet - and the hallucinations of AI programs themselves - describing āincest, bestiality, rape, sex trafficking, and sexual slaveryā in mind-numbing repetition to the point of notoriously destroying the sanity of many contract workers.