The Hundreds of Thousands of Workers in Poor Countries Enabling AI Systems Like ChatGPT (and Why They’re Controversial)

Since its launch at the end of 2022, the ChatGPT artificial intelligence program has not ceased to be a topic of conversation, both among those who admire this technological advancement and those who fear its consequences.

Much of the debate has revolved around the uses this intelligent chatbot could have, which can answer almost any question a user may have and produce texts that They look like they were written by a human being.

Will the students use it to do their homework? And the leaders to write their speeches? Could I write this article you are reading?

In addition to the major concern about whether this artificial intelligence (AI) program will put millions of people out of work today who perform tasks that this machine can perform in seconds, another controversy has to do with copyright.

ChatGPT used Information is mainly obtained from the internet. But generally no sources are cited, leading to allegations of plagiarism that have already led to legal complaints.

But behind the noise this innovation has created and the progress it represents for technologies using AI, lurks another controversy that is much less well known.

It has to do with the hundreds of thousands of workersmany low-income people, without whom AI systems like ChatGPT would not exist.

We’re not talking about the programmers who design the algorithms, who often work in places like Silicon Valley and earn good salaries.

We talk about the “hidden workforce” as the association called it AI partnership (PAI), which brings together academic, social, media and industry organizations involved in AI.

Who form this hidden force? People outsourced by large technology companies, mostly in poor countries in the southern hemisphere, to “train” AI systems.

tags

These men and women perform a tedious task -and potentially detrimental to mental health, as we’ll see later- but which is essential for programs like ChatGPT to work.

Consists of tag millions of data and images to teach the AI ​​how to act.

Take, for example, the chatbot that is causing a stir.

When you ask ChatGPT a question, the program uses some 175 billion “parameters” or variables to decide what to answer.

As we mentioned, this AI system uses information obtained from the internet as its main source. But how do you distinguish the content? Thanks to these references, which have been “taught” by people.

“There is nothing smart about artificial intelligence. He has to learn while being trained,” explains BBC Mundo Enrique García, co-founder and manager of DignifAI, an American company based in Colombia dedicated to hiring these “data recorders”.

These professionals, known as “data labelers” (data labellersin English), identify information, such as text, images, and videos, and tell the program what’s what, so the machine can understand what’s what and learn in what context to use it.

In the tech industry, they call this kind of job “data enrichment”.

But ironically, and despite being essential work for AI development, data enrichers make up for it the worst link in the production chain of large technology companies.

A fact recognized by AI partnership.

“Despite the crucial role these data enrichment professionals play, a growing number of studies show precarious working conditions what these workers are dealing with,” the agency said.

“This could be due to attempts to hide the AI ​​reliance on this large workforce by celebrating the efficiency gains achieved by this technology. Out of sight is also out of mind,” denounced the coalition, which also includes OpenAI, the company that founded ChatGPT.

Less than $2 an hour

A survey by TIME magazine found that many of the data taggers who were outsourced by OpenAI to train their ChatGPT received a salary of between $1.32 and $2 per hour.

According to a report by journalist Billy Perrigo, the technology company, which counts Microsoft among its main investors, has outsourced data enrichment work through a company of outsourcing named Sama, based in San Francisco, who hired workers in Kenya to carry out that project.

In a statement, an OpenAI spokesperson said that the company was responsible for managing the salaries and benefits of the taggers hired to work on the ChatGPT program.

“Our mission is to ensure that AI benefits all of humanity, and we work hard to build safe and usable AI systems that limit bias and harmful content,” the spokesperson said.

Sama, who also hires taggers in other low-income countries like Uganda and India for clients like Google and Meta (owner of Facebook), is touted as an “ethical AIand claims to have lifted more than 50,000 people out of poverty.

However, Martha Dark, director of British activist organization Foxglove, whose aim is to “take on tech giants and governments, for a future where technology is used to benefit everyone, not just the rich and powerful”, he told BBC Mundo that the big tech companies are using the outsourcing to pay these workers much less than it corresponds.

“All of these companies are multi-billion dollar businesses and it’s frankly inappropriate for them to pay $2 an hour to the people who power these platforms,” ​​he said.

But for DignifAI’s Enrique García, the salary controversy is “a matter of perspective”.

In Europe and the United States it is understandable that this is not enough, he notes, but in other countries it can be a good salary.

“Many people criticize our industry for the issue of pay, but at DignifAI our salary limit is US$2.30 per hour, And that is 1.8 times the minimum wage in Colombia.“, he makes clear.

“If the project is more complex and requires expert recorders, such as architects or doctors, the pay can be as high as $25 an hour,” he explains.

While he acknowledges that there are some companies that pay below the minimum wage, he thinks it’s unfair to focus only on this sector.

“There is dynamism of it outsourcing in many industries, not just this one, so it’s also not fair to label us as the digital sweat shops’ (digital clandestine workshops),” he says.

Social impact

García also emphasizes that there are several companies in the sector, like his own, that have a social impact and aim to “raise people’s productivity and dignity”.

In fact, DignifAI’s motto, which is supported by several aid organizations, is: “Outsourcing dignity through artificial intelligence” (“Outsourcing dignity through artificial intelligence”)

The company is based in Cúcuta, on the border between Colombia and Venezuela, and its mission is employing Venezuelan migrants crossing into the neighboring country and also internally displaced Colombians.

“Many of them were making $4 or $5 a day before working with us. For this vulnerable population that has no options in the labor market, earning 1.8 of the Colombian minimum wage is quite attractive,” he says.

Ingrid, a 42-year-old Venezuelan who arrived in Colombia in late 2018realize it.

The education graduate, who preferred not to give her last name, told BBC Mundo that she was unable to teach, as she did in her home country, because she has not yet been able to validate her title, saying that working as a scorer for DignifAI. This allowed him to earn a living and also train in another profession.

“You work four hours a day and I was able to spend the remaining time taking a graphic design course,” he said of his next work project.

Although she no longer works as a note-taker, having been promoted to project supervisor, I have no hesitation in recommending this job..

“It is more profitable, less tiring and better paid being a waitress, assistant part time or doing physical chores,” she notes, noting that most of her classmates are housewives, street vendors, or college students.

Mental health

Besides pay, another controversial issue surrounding data taggers is the effect of the job on their mental health.

It is not the tediousness of the task that worries some experts – although there is another criticism of this work – sino the toxic material to which some annotators are exposed.

And it is that part of the function of these trainers is to teach the AI ​​program what information is not suitable for publication.

To do this requires some – not Ingrid’s case – to dig into the darkest corners of the internet and catalog all the vast wealth of violent, sinister and perverse material that resides there, to show the machine the rotten side of the large network of networks.

But according to Foxglove’s Martha Dark, doing this job is vital,”can cause post-traumatic stress disorder and other psychological problems among many employees.”

His body represents a former employee of Sama who worked as a Facebook moderator in Kenya and in 2022 took both Sama and Meta, owner of the social network, to court for the psychological damage he suffered, a case that is still under investigation. discussed in Nairobi court.

“These jobs come at a cost to mental health of those who make them and should receive adequate psychiatric care in addition to a fairer wage,” Dark told BBC Mundo.

According to the activist, the big tech companies have enough financial resources to provide this kind of help, but they don’t because “they put profit before safety of its employees”.

Enrique García acknowledges that large companies could invest more in hiring taggers, but he affirms that if they demand too much from them, they may decide to look elsewhere for annotators.

“It could be that Big Tech could pay more, but we are very grateful for the opportunities“, say.

“Either we get defensive about the stinginess of the customer, or we accept the opportunities that are there, which pay above the legal minimum,” he says.

“At least we’re bringing monetization opportunities here where there wouldn’t be any without this alternative.”