The Workers at the Frontlines of the AI Revolution
Rafael Rodríguez Deustúa made his career by drawing on demand. An illustrator who advertises on the freelance artist platform 99designs, Deustúa’s recent works are eclectic — they include a vintage-style logo for a champagne brand, a poster for an international fertilizer convention, a tango album cover, and a handful of custom tattoo designs. He sells his illustrations, which feature his signature cartoon style, under the pseudonym Fafarhd Deustúa.
Deustúa is based in Guadalajara, Mexico. Most of his clients are startups and small businesses in the U.S., with a few loyal customers in Europe and Australia. “I earn more than I would with local clients,” he told Rest of World. In exchange, his clients “get some decent art at a fair price.”
But recently, Deustúa, like many creative workers globally, has encountered a new kind of competition: generative artificial intelligence.
He first noticed AI-powered submissions on 99designs’ popular “contest” feature, which allows clients to post an open call for designs before selecting their favorite. In recent months, he has seen people submitting entries created using AI image generators such as Midjourney, Stable Diffusion, and Dall-E.
“Someone who doesn’t have that talent or [hasn’t] invested years of practice in getting the skills to just ‘win’ a contest with an illustration he or she didn’t really do — that bothers me,” said Deustúa, who refuses to use AI tools in his work.
Since the blockbuster launch of ChatGPT at the end of 2022, future-of-work pontificators, AI ethicists, and Silicon Valley developers have been fiercely debating how generative AI will impact the way we work. Some six months later, one global labor force is at the frontline of the generative AI revolution: offshore outsourced workers.
These include workers hired per commission or on a contractual basis, such as freelance copywriters, artists, and software developers, as well as more formal offshore workforces like customer service agents. As generative AI tools present a new model for cost cutting, pressure is quickly mounting for these outsourced workers to adapt or risk losing work.
Rest of World spoke to outsourced workers across different industries and regions, including call center operators in Manila, programmers in Lahore, and designers in Cairo. Many said they are already seeing generative AI change the demand for their work and the stability of their income. But while some brace for layoffs or diminishing commissions, others have embraced generative AI tools in an effort to keep ahead of the curve. If generative AI represents a tectonic shift in the way we work, offshore outsourced workers are at the fault lines.
Abisoye Otusanya-Azzan
A COPYWRITER LIVING IN LAGOS, NIGERIA
When ChatGPT launched last year, Abisoye Otusanya-Azzan, a copywriter based in Lagos, Nigeria, couldn’t ignore the hype. “The vibe around ChatGPT was crazy,” she told Rest of World. “I started using it as soon as it was made available.
Otusanya-Azzan, 29, started writing for a living during lockdown in 2020, after her income from fashion design dried up. She mostly writes marketing material for U.S. and U.K.-based clients in the personal development, fitness, and home improvement sectors.
She struggled with ChatGPT at first, but bought an e-book and joined a Facebook group for copywriters to pick up tips. “I learned how to give better prompts and the results were a bit better,” she said.
Otusanya-Azzan mainly uses ChatGPT to do research, overcome writer’s block, and craft posts for her own blog. She makes sure to verify its output. “It does make mistakes,” she said. “I don’t completely trust it.”
While AI tools may ease the work of copywriters, they could also reduce demand for their skills. According to researchers at OpenAI, the organization behind ChatGPT, writing is among the occupations most exposed to text-based generative AI.
“I think that a lot of clients are already using AI tools instead of hiring copywriters,” Otusanya-Azzan said. The technology may, however, bring different opportunities. “I’ve also seen some companies looking to hire writers who are also prompt engineers, so that might be the next line of demand,” she said. “I’m curious to see how things turn out.”
In search of lower wages, less oversight, and faster turnaround times, businesses in Western nations have long turned to workers in countries with cheaper labor. Freelancers on online gig work platforms like Fiverr, Upwork, and 99designs may not fit the conventional mold of outsourcing, but they largely map onto the same flows of work: clients in the Global North hiring workers in the Global South.
India, Pakistan, and Bangladesh were among the largest hubs for freelancers on online gig work platforms in 2020, according to a study of 11 such platforms by the Online Labour Observatory. Overwhelmingly, the clients on these sites were located in the U.S., Canada, Australia, and the U.K.
“This is very clearly a new way of offshoring services today,” Uma Rani, senior economist at the International Labour Organization (ILO), a United Nations agency, told Rest of World. “Employers want workers in these countries because they can pay less wages. But on the other hand, the amount of [platform] fees workers have to pay — it’s almost double the exploitation that these workers face.”
The precarity built into the offshore model makes these workers particularly susceptible to AI-fuelled shifts in the labor market.
“This is very clearly a new way of offshoring services today.”
Rest of World conducted a survey of over 31,000 employees in tech, finance, and media via the platform Blind, an app that allows verified employees to post anonymously in industry forums. The largest share of respondents were employed by major Silicon Valley tech firms in the U.S., India, the EU, and Canada, including Amazon, Microsoft, Salesforce, and Meta.
Globally, 76% of respondents said they believed generative AI would allow their company to cut costs, and 63% predicted their company would hire fewer external contractors due to generative AI tools.
Despite predicting less work for contractors, most respondents did not fear changes in their own employment. While 56% said they currently use generative AI in their work, only 24% said they were either “slightly concerned” or “very concerned” that generative AI would render their own jobs redundant.
With the threat of less contract work to go around, many on online gig work platforms are trying to ride the wave of interest in generative AI by offering related services — and undercutting other contractors.
Need someone to churn out an illustration for your online retail store? Rather than hiring a traditional illustrator, you can use a dedicated “AI artist” portal where creators post their rates for custom images made using Midjourney, Stable Diffusion, or Dall-E. Prices regularly run from $15 to $60 apiece. Need someone to write blog posts that top Google search rankings? Instead of finding a copywriter, you can hire Fiverr workers selling bulk orders of AI-generated text. One worker in Bangladesh allows clients to buy as many as 1,000 articles on a given topic at a time. An order of 120 GPT-3-generated articles costs $90, and comes with claims of a plagiarism-free guarantee.
In January 2023, Fiverr said it had seen a 1,400% increase in searches for AI-related job ads on its platform over the previous six months. The spike in demand pushed the company to create entirely new job categories, under an umbrella it calls “AI services.” For workers in this space, the popularity of generative AI has been a boon. Several of the creative workers Rest of World spoke to said it had helped them take on more clients and produce work faster, even if at lower price points.
Shweta Bose, based in Delhi, India, has long offered Adobe Photoshop services on Upwork. The release of “generative fill,” a new Photoshop feature launched in May, allows her to fill in, extend, and retouch images for her clients using simple text prompts. Real estate agents in the U.S. have hired her to retouch photos for listings, and families have hired her to convert multiple photographs of family members into a single digital portrait.
“I think [generative AI] will create more opportunities for freelancers, but drive the price of certain services down,” Seif Percle, an Upwork contractor based in Cairo, told Rest of World. Percle’s recent commissions include creating illustrations for children’s books using Midjourney and Stable Diffusion.
In some sectors, such as programming and engineering, workers have found whole new services they can offer specifically around the rise of generative AI, as companies scramble to adopt the technology.
“I think [generative AI] will create more opportunities for freelancers, but drive the price of certain services down.”
“My Upwork has been going crazy with invites from different-sized businesses trying to improve or build out new stuff for their businesses with AI,” Rehand Haider, a developer based in Lahore, Pakistan, told Rest of World.
Haider, who joined Upwork five years ago, estimates the demand for his AI implementation services has grown by two-thirds over the past six months. One of his recent commissions used ChatGPT’s application programming interface (API) to create an app that automatically generates descriptions for Airbnb listings. Another used ChatGPT to make a Chrome browser extension that automatically fills in job applications.
Ali Zeeshan, a data scientist in Islamabad, Pakistan, told Rest of World the demand for generative AI on Upwork has been a major boost to his profile. “I am getting more views, jobs, and top-notch clients,” he said. His commissions grew threefold once he started offering AI services this year.
Mohammad Maruf Ahmed, based in Sylhet, Bangladesh, used to offer digital marketing services — helping brands in the U.S. grow their footprint on Instagram and YouTube. He switched gears when he received his first AI-related commission in December 2022. Now, he creates custom scripts for his clients to integrate ChatGPT into websites, Google Sheets, and Google Docs. Next, he sees himself using AI to help clients build brand-customized chatbots. “The use case is changing on a daily basis,” he told Rest of World.
It is not just programmers pivoting towards AI. Workers who, in the past, advertised copywriting skills are now offering to copy edit or fact-check articles written by ChatGPT, while others write prompts for AI tools.
But while some early adopters are able to take on more work thanks to AI assistance, there may be less to go around for others. “There’s a lot of competition that happens. Workers undercut one another from developing countries because it’s the only way you can get any access to work,” Rani, the ILO economist, said. She explained that even before AI tools were available, workers on these platforms would take on free assignments or accept pay cuts to stay ahead of other contractors.
There is particular pressure on outsourced workers to offer cheaper rates and shorter deadlines, she said, as they are often employed commission to commission, with little financial security or legal recourse when wages go unpaid.
“I feel a bit secured that I was able to grasp the technology faster and earlier than many,” Faiz Ehsan, an AI artist in Patna, India, told Rest of World. In March, Ehsan switched from designing video game plots and writing quest narrations to creating book designs and illustrations, made largely using tools like Midjourney. The pivot is working — he wakes up to five to seven new messages from potential clients on Upwork every morning. But he knows it may not last. “The challenges are always there. Since every day a new trend happens, I have to know, practice, and understand that trend to be able to sell it,” he said.
Jessica Tarriela
A VIRTUAL ASSISTANT LIVING IN TAGUIG, PHILIPPINES
When ChatGPT launched in late 2022, Jessica Tarriela thought it could be a useful tool. A virtual assistant based in the Philippines, her work involves writing emails and conducting research for a content marketing client in the U.S. “I can confidently say that I’m not a good writer,” Tarriela told Rest of World. “So I use ChatGPT to rephrase or write or summarize things.”
Tarriela, 26, works nights to compensate for the 12-hour difference between Georgia, where her client is located, and her home in Taguig. Her client knows that she uses generative AI in her work — in fact, he encouraged it. “He asked me if I would be open to using AI, and I responded that I already am,” she said.
Part of Tarriela’s role is to create content such as blog posts and case studies for software businesses, and she uses generative AI to polish these texts. Her go-to tool is ChatGPT plugin AIPRM — it offers preset prompts for writing or editing texts, with names like “Complete Proofreading in One Click” or “Human Written [100% Unique] SEO Optimized article.” She said generative AI doesn’t save her time, but it improves the quality of her work.
It’s difficult to know how many Filipinos work as virtual assistants, but payment platforms estimate that at least 1.3 million people work in the country’s business process outsourcing sector. Many are concerned AI tools will make their jobs redundant, Mylene Cabalona, president of the Business Process Outsourcing Industry Employees’ Network, told Rest of World. This is partly because the sector is unregulated, with few protections for workers. “There’s no employer-employee relations; their clients are completely offshore,” she said.
But Tarriela believes being an early adopter will protect her position. “You shouldn’t be afraid of AI, you should be afraid of people who use AI to make their work even better,” she said.
Online gig workers may be at the forefront of generative AI adoption and displacement, but there are clear signs that AI is also quickly coming for the more traditional outsourcing industry.
Call centers have long been the face of offshore outsourcing, and waves of technological innovation have been key to their evolution. In the 1980s, automated phone trees with voice recognition sped up customer service call times. The first chatbots, which blended prewritten prompts with humans behind the curtain, began filtering out basic inquiries a decade ago. Now, conversational AI tools may upend the industry once again.
A recent study by researchers from the Massachusetts Institute of Technology and Stanford University, based on data from more than 5,000 customer support agents largely in the Philippines, found that using generative AI tools increased worker productivity by 14% — a number that rose to 35% for new hires and those who were previously least productive.
Mylene Cabalona is one of the roughly 1.6 million people in the Philippines who work in business processing outsourcing (BPO) — jargon for the umbrella industry that comprises offshore call centers, content moderation hubs, and IT support hotlines. Taken together, the BPO sector is one of the largest employers in the Philippines.
“We’re not against AI, we’re not against any software development,” Cabalona told Rest of World. “AI should be a mechanism to help the workers. But it should not be used against the workers. It should not be used to replace them.”
Cabalona serves as president of the BPO Industry Employees Network (BIEN), a nationwide association of BPO workers organizing in the notoriously anti-union sector. Cabalona said she and her colleagues are bracing for layoffs.
In a letter to shareholders and analysts in May, Daniel Julien, the chairman and CEO of BPO company Teleperformance, wrote that “ChatGPT’s launch has now made AI a boardroom and CEO agenda item, rather than just being an operational back-room topic.” Teleperformance is a powerhouse in the BPO world, with a long roster of Silicon Valley clients including Amazon, Netflix, Apple, Meta, and Microsoft.
“Whether we look at cotton looms or the assembly belts or industrial robots or AI now, people are always worried.”
By plugging ChatGPT into its operator scripts, the company estimates it will reduce the length of each customer call by almost 40%, according to a slide deck shared with investors. Teleperformance has already planned or implemented 110 client projects using generative AI. Like other BPO companies, it maintains that “human empathy” will remain essential to its operations, especially in handling more complex customer calls. But the company estimates that by 2026, conversational AI could handle as much as 20% to 30% of its call volume.
That kind of shift would have an impact on BPO workers in the Philippines, who often handle the front-line customer service calls that Teleperformance has deemed “easier to automate.” BIEN estimates 54% of all BPO workers in the Philippines are front-line call center workers. “We are anticipating AI and most of the affected would be the front lines — and most of these [workers] are women,” said Cabalona. More than 50% of workers in the BPO sector in the Philippines are women, but they tend to be concentrated in lower-paying, “lower-skilled” roles, according to the ILO.
Teleperformance did not respond to Rest of World’s requests for comment.
Skeptics remain, however, over the extent to which generative AI is actually being implemented in the BPO industry. “There is a huge incentive for [BPO companies] to promote themselves as being ahead of the game,” said Christy Hoffman, general secretary of UNI Global Union, which represents over 20 million service sector workers around the world. Hoffman suggested vendors have reason to sensationalize the potential benefits of AI, noting that Teleperformance’s stock value had dropped in the months before the company released its letter to investors. “This is PR, this is not at the level of the workplace and we expect that it will become so, but over time,” she said.
Milagros Miceli, a sociologist and computer scientist researching labor conditions in outsourced data work, told Rest of World the threat of automation can also be used as a tool to get workers to do more for less. “If there’s always the threat of having machines replace you as a worker, that serves to discipline you, for you to be afraid to ask for a raise, for you to be afraid of even getting sick,” she said. “You’ve been warned.”
Labor researchers note that automation predictions are notoriously unreliable. “Whether we look at cotton looms or the assembly belts or industrial robots or AI now, people are always worried — once machines do tasks that humans usually do — that machines might run the labor market one day,” Fabian Stephany, a research lecturer in AI and work at the Oxford Internet Institute, told Rest of World.
In recent history, experts have failed to produce reliable studies on how much of the global labor force is at risk of technological unemployment — estimates in the past 10 years range wildly from 9% to 54%, even though these papers are based largely on the same data sets. Stephany shies away from predictions, instead positing AI will both end and create occupations. “Economic theory and history tells us that AI is not going to automate human work away,” he said.
Wu Dayu
A MARKETING DESIGNER LIVING IN SHENZHEN, CHINA
It used to take at least a week for Wu Dayu’s Shenzhen-based design studio to create promotional materials for online fashion stores. A photo-shoot required a model, makeup artist, photographer, and venue. Producing a catalog of images featuring six sets of clothes would cost around $3,500, he told Rest of World.
But since Wu, 35, switched to using generative AI in March, the same work can be completed in a day, by just two people, and for only $140.
Now, Wu uses a mannequin or asks an employee to model clothes so he can take some basic photos. He feeds the images into a Stable Diffusion-powered AI program that he has customized for this specific purpose. He adjusts the images of the model and background by using text prompts, such as “a girl, wearing sneakers, street background, with green trees on the two sides of the street.”
This approach is becoming more widespread in China’s sprawling e-commerce industry, even if some consumers have criticized the resulting images as fake and sexist. “Some high-end brands might prefer human models,” Wu said. “For small and mid-sized sellers, AI models will save them a lot of money and time.” In April, Wu laid off 60% of his staff.
Shaeman Liu, a 26-year-old fashion photographer in Hangzhou, said he doubts AI will replace all photography. He believes products advertised with real-life models are more appealing to shoppers.
Model Charline Xu, 24, told Rest of World she currently makes about 1,500 yuan ($210) an hour, and that she’s worried clients using generative AI could cut into her income. She’s prepared to compete. “If [designers using AI] charge 800 yuan, I’ll do 600. If they charge 600, I’ll get down to 500,” Xu joked. “There’s no other way out. I’ll fight till the end.”
The impact of generative AI on outsourced workers is likely to similarly be more nuanced than a sudden mass displacement.
For now at least, many labor researchers resist the idea that these tools bring “full automation” — often defined as the use of machines to complete tasks without human intervention. Rather, the current state of generative AI, by and large, falls under what experts call augmentation. Generative AI’s first wave of adopters are mostly working alongside these technologies — treating them as tools, not substitutes.
Even the development of generative AI requires human intervention, a role that often also falls to offshore outsourced workers. Each training data set used to develop a tool like Dall-E began with a human data worker matching an image to an associated word. And content moderators help refine the model that powers ChatGPT — for example, by flagging hateful or toxic speech. “[These workers] are fundamentally necessary to the developments that we see today,” said Miceli. “Large [AI] models — they wouldn’t exist without this workforce.”
Many of these data workers congregate on sites like Microworkers, Remotasks, and Amazon Mechanical Turk, which provide a constant stream of low-paid tasks, sometimes referred to as “click work.” Others are employed through a web of BPO outsourcing subcontractors, including Teleperformance.
“Instead of workers displaced by AI, the future may look like more workers filing into micro work projects,” Rafael Grohmann, principal investigator for Fairwork Brazil, a digital labor and AI research project, told Rest of World. “In this stage now, we need many quality data workers, more people training and annotating data.”
While some offshore outsourced workers have already pivoted to AI-related work, others are committed to holding the line against the rising tide of automation. Deustúa, the illustrator from Guadalajara, said he’s counting on clients who want stronger relationships and attention to detail to keep his business viable.
“A lot of clients have trouble explaining to us, humans, what they want and we help them to explain themselves,” he said. One client recently admitted they hired Deustúa after trying and failing to use an AI image generator to create a workable design themselves. Ultimately, the client decided they still wanted a professional’s human touch. “It’s the difference between buying a snack at a vending machine or eating it at your neighborhood cafe, chatting with the barista,” Deustúa said. “I still love technology, but I bet on humans.”
It is a position informed by life experience. Deustúa wasn’t always an illustrator. For over two decades, he worked as a journalist in Mexico — first as an arts and culture reporter, then editing tech and science stories. Following the rise of digital news sites and social media in the 2010s, he found there were fewer jobs in print media and bet on freelance illustration as a new career. Now, seeing generative AI enter the market for illustrators is a “bitter experience.”
“I was already displaced once because of technology,” he said.