Why?

“No idea,” he says with a laugh. “Sometimes I wonder why anyone would pay me to do this.”

The reason is AI. Self-learning machine models need vast amounts of data to develop. But not all data is equal.

Today’s generative artificial intelligence struggles on many fronts and needs human input for things such as distinguishing features in an image and making sense of texts.

This has given birth to an industry that contracts millions of people like Delmar (his name and the names of other workers in this article have been changed, since the platform they work for has warned them not to talk to the media).

They are freelancers, working alone on obscure tasks for anonymous employers over impersonal online platforms. Outside that reality, their existence is rarely noticed.

So far, AI’s growing autonomy, intelligence and ability to take over the world has been of greater concern than the legions of workers it depends on to make out the difference between a cat and a dog.

The hunt is on to make more of the new technology and, beyond customer service, report writing and self-driving cars, companies in every imaginable field spar for AI talent to boost productivity.

For the foreseeable future, however – and despite all the “self-learning” PR – all of this will be far from autonomous. To make chatbots seem human, self-driving cars reliable and shopping programmes accurate will require enormous amounts of human labour.

Market research firm Grand View Research predicts that the Asia-Pacific market for annotating data will grow by more than 200 per cent by 2030. This means plenty of job opportunities – but also challenges.

The industry is unregulated and plagued by a culture where workers are seen as disposable. Some call it a digital sweatshop. Aside from being a threat to individuals and local economies, this could also put the quality of future AI at risk.

Delmar opens the door, eyelids sagging. It’s just before lunch. “I love it,” he says, grinning. “I can work whenever I want!”

We step into his world: four walls covered in anime posters. A small, humming refrigerator. An unkempt bed next to the desk with the computer, and, by the opposite wall, a dumbbell.

Since he saved up to buy a computer and router he leaves his home only to go shopping and, once a week, to meet friends. In short, any young man’s dream. Except he also provides for his sister, grandmother and aunt. A hard task when electricity and internet bills often eat up half of his salary.

“My father left when I was 18,” he says, “so since then it’s been up to me.”

Delmar’s shack of a house is squeezed into a tight alley in Agusan, a community along the main road hugging Macajalar Bay on its way east out of the city.

Death on the high seas: the dark side of China’s rise as seafood superpower

Every major downpour floods the area, but the neighbours keep it tidy in between bursts. Graffiti on a concrete wall reads “We’re poor”, in a proud, joking Tagalog turn of words. “What can I say,” says Delmar. “It’s home.”

A scrawny figure, Delmar thanks his lucky stars that he found Remotasks, and he doesn’t have to do manual labour.

“I wouldn’t have been able to work in construction like my dad,” he says. “Now, I’m a tasker, just like all my friends.”

It’s not a coincidence that tasking has become so popular in the Philippines. The country has a large, young, connected and English-speaking population that will work for low wages.

In the past couple of decades, the country has leveraged these factors to become a world leader in business process outsourcing. A fair share of that industry has been attracted to Cagayan de Oro, a regional hub of 700,000 on the southern island of Mindanao.

The city boasts a technological university as well as a lower cost of living and less congestion than Manila. However, it is also a migration magnet in a particularly underdeveloped and conflict-ridden part of the country.

Official figures show the unemployment rate is twice as high as the national average. During the Covid-19 pandemic, it was even worse.

Harsh lockdowns forced many to look for opportunities online. Some found computer games where players could turn virtual currency into real money.

Save for a few, these schemes did not turn out well, often leaving those taking a chance on the gaming apps further in debt. Others, however, found Remotasks: anyone can create an account, take a few tests and start tasking.

Once you’ve proved yourself at some simpler tasks you can progress to work that is more advanced and better paid. A dashboard lets you keep track of the money you’ve made, which will be deposited weekly into your online account.

According to the website, more than 240,000 taskers are currently signed up, and they have earned a collective US$15 million.

A contemporary of Delmar, Mario was on the brink of despair when he first heard of Remotasks. Six long, jobless months had passed since he had earned his degree in computer science, then a friend invited him to an internet cafe for a training session. Paid, no less.

Mario learned how to mark various features in photos, maps and 3D graphics. He was trained in how to label different parts of text in a variety of languages, and how to sort products such as shampoos and liquor.

“In the beginning, the work stuck in my mind,” he says. “I even dreamed of pointing out cars and road signs. It was a nightmare, like living inside a machine. But after a while I got used to it, and focused only on making money.”

And money he made. He progressed in the company hierarchy, and started training and overseeing new taskers.

iPhone design chief enlisted by Jony Ive, Sam Altman to work on AI devices

When the internet cafe was booked out for Remotasks, people fought for seats. They performed tasks with obscure animal names such as Falcon, Gorilla and Flamingo. Because of his rank, Mario knew some of the companies hiding under those pseudonyms.

“Parrot was Lexus, Whale was Waymo [formerly known as Google’s Self-Driving Car Project]. A lot of them were car brands.”

Unlike the other taskers, he also knew the creator behind Remotasks: Scale AI. Founded in San Francisco in 2016, the company has become a tasking giant, valued at more than US$7 billion.

On its website, it boasts it has furnished over 7.7 billion data annotations for companies such as OpenAI, Meta and even the United States Army.

A video brings the viewer into a bright office of wooden beams and exposed brick walls, while 26-year-old founder Alexandr Wang narrates, “What I’m really excited about is assembling a really cohesive, hard-working group of people who are all there to build something great.”

Smiling colleagues pile in over a communal table with free snacks, as Wang continues, “The most impactful thing we can do is to accelerate every single team working on this and lay down the infrastructure to actually make it easier and faster to build machine learning.”

For Mario, it didn’t take long before problems started cropping up and becoming more frequent.

How illegal miners feeding China’s stainless steel demand scar South Africa

Taskers complained about being paid too little, too late or not at all. Some of them were fired for bringing up their grievances in the communal chat forum. Others quit.

During one period, Mario says he replaced half of his labour force every month, most of whom were teenagers, some even under 15.

By now, they were taught online, without compensation and, according to many, without sufficient guidance. And the task of training machines is complex.

Firstly, because you need to think like a machine. A car that is partly covered by another car doesn’t abruptly end, for instance, but continues behind that other car. And you have to make that understood by the AI.

Secondly, instructions for the tasks constantly evolve.

If you don’t keep up to date and get them right you can be shut out of the platform. Mario, who was eventually laid off, blames himself for not helping his team more.

“It was like a Chinese sweatshop,” he says. “We were just pulling people from the streets, as many as we could, training them in a rush and putting them to work.”

In 2021, when Remotasks expanded to Venezuela and India, salaries and task availability in the Philippines dropped.

Since then, well-paying tasks have been fewer and further between. For students this is not a big issue, but others use tasking for their sustenance.

These days, it is common to be paid below the minimum wage (US$7.50 per day in Cagayan de Oro). Internet cafe owner Gerard says taskers sometimes ask him for loans.

“When I heard that companies worth several billion dollars run this operation,” he says, “I thought to myself that there must be many corrupt people in this world.”

Establishing liability is difficult. And for politicians who want to keep the unemployment rate low it might not be a priority.

The involved parties don’t make it easier, either. Scale AI’s Filipino subsidiary, Smart EcoSystem Philippines, has neither a website nor a phone number. No one answers its official email; its local office in Cagayan de Oro is empty and, according to neighbouring businesses, has been so for weeks.

Russel Jallorina, assistant regional director at the Department of Labour and Employment, laments that the employer is invisible: “If they don’t have any physical presence, no office, how are we then going to be able to inspect them?”

Scale AI refers to a blanket statement on its website. There, the company writes that it is a partner in the Global Living Wage Coalition and undertakes quarterly pay analyses.

An anonymous survey it had conducted of 5,000 taskers across the globe showed that 75 per cent rated their satisfaction with Remotasks at four out of five, or higher.

Scale AI also writes that the number of tasks available fluctuates alongside demand, and most workers are part time or use it to supplement their main income.

This contradicts the blurbs on Remotasks’ sign-up page, however. “Earn a living,” it says, with “new projects always coming in & long-term projects you can consistently work on”.

Uma Rani, senior economist at the United Nations labour agency International Labour Organization, argues that local politicians have abdicated their responsibility.

Instead of creating jobs for their skilled labour force (most taskers are educated), they have allowed the online platforms to take over.

For orphan who fell into drugs and saw her son die, hope after the heartbreak

“Sure, it might be better than no job,” she says, “but if you scratch the surface it’s not all that free and flexible. Fierce competition forces you to constantly be on call for new tasks.

“And after five, seven years you’re burnt out, without a future, and without any meaningful development of the local economy.”

She places equal responsibility on the foreign firms who buy the data.

“Whose productivity is being improved by all this work? At what cost? If automobile companies need this annotation, they should hire a division of labourers and give them proper contracts.”

After The Washington Post published an article on the issue last August, a host of progressive American politicians, including former presidential candidates Bernie Sanders and Elizabeth Warren, wrote an open letter to the US’ leading tech companies, including Scale AI.

They demanded that the companies explain how they use taskers and what they are doing to improve their work situation.

Danila Petrelli, senior data manager at AI Sweden, the Swedish centre for AI development, says that besides the ethics, there are other reasons why improving labour conditions should be a priority for tech companies.

“If we stop for a second, we see that this is not paving the way for any state-of-the-art help to improve AI. We get a high quantity of data, yes. But I’m concerned about the quality of data these workers produce.”

She mentions the need for supervision, clearly defined tasks, a maximum amount of work hours, decent pay and psychological support to allow taskers to be able to work well.

“The view of these labourers as interchangeable makes the industry miss the expertise they build up, which will be necessary to develop the work of tasking further,” Petrelli says.

China unveils new AI rules for scientists, banning use in funding applications

Wednesday is payday, which also makes it Delmar’s night out. This is when he usually goes for a beer or a Jollibee dinner with his tasker friends. It’s a time to unwind, but also to share information about the latest tasks and updates at work.

Recently, however, there have not been any beers. The tasks have dried up, especially the ones that pay a decent wage. For a few weeks now, Delmar has had to take out loans to make ends meet (another app industry fraught with abuse and unaccountability).

Asked what the future holds, he is stumped. Sitting in the internet cafe, he smiles a little unassuredly, turns back to his computer screen and clicks his mouse. The screen displays the message “Your task queue is currently empty.”

“It’s very bad,” he says, rubbing his eyes. “No work, no rice.”