In a small (a very small) way, I collect 18th and 19th century company tokens. These symbolise for me the worst of the financial exploitation of the industrial revolution. It was not enough for the industrialists to ruthlessly exploit the excess availability of labour and pay badly the new industrial workforces of their dark satanic mills*. In many cases they also paid not in money that could be spent anywhere, but in scrip – tokens that could only be spent at the company’s own store. Prices there reflected the guaranteed market, so workers were exploited over again. These practices were progressively abolished in England from 1831 onwards by the oddly-named (at least to modern ears) Truck Acts. There is similar legislation to bar such abuses elsewhere in the world – though not everywhere.
As we know, this financial exploitation sat alongside brutal working conditions where injuries and even death were common, accepted outcomes of the industrial process. That was a lack of health and safety gone mad.
We’ve known for a while that the AI revolution depends on a similar exploitation of the health and safety of workers. The stories of the employees of Sama in Kenya, who helped train ChatGPT, are disturbing. The human training of these supposedly ‘artificial’ intelligence systems (ChatGPT is no worse in this regard than its rivals) involves individual people being exposed to the worst things that the draft forms of AI machines produce. As the machines’ training materials are the entire internet, this replicates the biases of the present and prejudices of the past, and includes all the filth that humankind has produced in recent years. The human job is to tell the AI not to produce further paedophilia, repeat racist incitement, and so on – but in order to do that, people need to read and look at truly horrific material.
Sadly, the people who did this work are not treated well. Their mental health disorders are the equivalents of the fingers on the floors of cotton mills. These seem not to trouble those who are making epoch-making amounts of money, and little enters the public discourse so that it has minimal impacts on consumer use of these products.
But it turns out that such physical exploitation of people’s health isn’t all that’s going on in the current technological revolution. Amazon has revived the company scrip model. It pays some of its MTurk workforce in Amazon gift cards, and severely constrains how those gift cards can be spent so that the workers are unable to get full value from them. MTurk – mechanical Turk in full – is the name for the distributed self-employed workers who perform tasks that help test and train much modern IT and so ensure its smooth working. The name aptly reflects the 18th century supposedly mechanical chess board, called the Turk, that toured Europe playing matches. Instead of being an automaton, the Turk actually only worked because in place of a machine there was a skilled human chess player crammed uncomfortably into the space under the board.
In the same way, the human work that is necessary to help train current supposedly ‘artificial’ intelligence technologies suggests there is some artifice in calling them artificial.
The DAIR Institute (Distributed AI Research Institute in full) – the grouping formed by the authors of the Stochastic Parrots paper – have launched a Data Workers’ Inquiry trying to bring forward the stories of the people who are directly involved in facilitating the current technology revolution, and who all too often are its unhappy victims. Consistent with the DAIR philosophy, this includes putting the voices of the individual workers themselves at the heart of the work, and facilitating them in telling their stories in the forms they find most comfortable and appropriate.
One of the stories discussed on the launch webinar, and on which the Inquiry has published a short paper, highlights this issue. Though its author, Alexis Chávez, is from Venezuela, the use of gift cards as payment isn’t restricted to countries where currency or sanctions issues might limit payments in real money: Chávez shows that the practice applies in (at least) Brazil, Colombia, India, Kenya, Mexico, Pakistan, and the Philippines. The paper details the convoluted processes needed for these individuals to gain value from their gift card payments, which mean that they are in effect forced to take discounts of 20-30% in order to extract value. It’s like the mark-up in the company store.
And it’s hard to argue with Chávez: “Even though Amazon does not see them as employees but as independent contractors, it’s our right to be paid fairly and in a useful manner.” We fondly thought the worst of the financial practices of the industrial revolution were far behind us – they should be – but unfairness clearly persists in the very human side of the supposedly ‘artificial’ intelligence business.
The second event in the Data Workers Inquiry happens this week, and Chávez himself is due to speak on August 26th.
* This phrase is from William Blake’s preface to his lengthy 1804 poem in praise of John Milton, words that are now known to us as Jerusalem. Please consider supporting the campaign to save Blake’s cottage in the West Sussex village of Felpham.
I am happy to confirm as ever that the Sense of Fairness blog remains a wholly personal endeavour.
See also: Learning from the Stochastic Parrots
Mental Health and Drug Dependency in Content Moderation, Fasica Berhane Gebrekidan, the Data Workers Inquiry, June 2024
Click Captives: The Unseen Struggle of Data Workers, Wilington Shitawa, the Data Workers Inquiry, June 2024
The African Women of Content Moderation, Botlhokwa Ranta, the Data Workers Inquiry, June 2024
OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic, Billy Perrigo, Time, 18 January 2023
The Distributed AI Research Institute (DAIR Institute)
The Impact of Gift Card Payments on MTurk Workers, Alexis Chávez, the Data Workers Inquiry, June 2024
The Sense of Fairness blog 