For a decade, hundreds of thousands of workers fueled the boom in artificial intelligence by performing repetitive micro-testers: identifying objects on images, moderating content, classifying texts. These activities, carried out online for a few cents per unit, have led to the most powerful AI models in the world. Today, these jobs disappear as models become able to generate their own data. A structural change takes place in the value chain of artificial intelligence, relegating in the background those who were the invisible workers.
An economic model based on the abundance of cheap labor
From 2015, platforms like Appen, Remotasks, Scale AI or Mighty AI prospered thanks to a simple promise: subcontracting workers from countries in repetitive tasks crisis that algorithms did not yet know how to accomplish. Venezuela, ravaged by hyperinflation, became a strategic pool. In 2018, up to 75 % of employees in some of these platforms were Venezuelan. They worked without contract, from home, with the only condition access to the Internet.
The economy of micro-tashes was based on a logic ofextreme outsourcingwhere platforms delegated to precarious workers the training of computer vision models, autonomous vehicles or research tools. The missions were simple, standardized, weakly paid, but constant.
Technological rupture: self-learning and synthesis of data
The arrival of generative and self-learning models upset this balance. By generating their own training data (Synthetic Data), the AI reduce their dependence on human annotation. Formerly crucial, such as classification of images or linguistic correction-are now carried out by the models themselves, with minimal supervision.
The demand moves to rarer, more complex data games, intended for specific uses: advanced logic, coding, scientific knowledge. These tasks can only be entrusted to experts with a solid material infrastructure and high technical skills.
Result: the volume of tasks accessible to low -skilled workers collapses. Appen, one of the world leaders in the sector, has lost 99 % of its market value in three years. Its turnover fell 14 % in 2023. The company, formerly specialized in human annotation, now redirects its strategy towards automatic generation projects.
Exclusion by complexity
The change of nature of tasks creates a technological entrance barrier. Workers located in countries like Venezuela or Colombia must face multiple obstacles: obsolete computer equipment, unstable connections, frequent electricity cuts. Even when they have left their country, their migrant status limits their access to the formal labor market, and therefore to the rise in structured competence.
In a context where platforms offer neither training, progression, nor contractual stability, the transition to higher added value tasks remains illusory for the majority of them. The disengagement of platforms is not negotiated: when the tasks disappear, workers are simply no longer called.
An economy of click under tension
This phenomenon goes beyond annotation tasks. Content professions, such as SEO writing or freelance copywriting, are also affected. Customers now demand that editors use tools like chatgpt to produce more quickly – while drastically reducing prices. On platforms like Upwork or Fiverr, average remuneration for 1,000 words have dropped by more than 70 % since 2022.
This shift introduces a contradiction: the workers must have to follow the pace imposed by AI, but do not have resources to achieve it. Access to Chatgpt, limited in some countries, requires a VPN, an international bank card, or even a monthly subscription. A technological arms race sets up between precarious workers.
A transition without shock absorber
This paradigm change reveals structural fragility: the economy of artificial intelligence has been built on a flexible, global, interchangeable workforce. It is now transformed without support plan, without regulation, without reconversion policy. Millions of micro-travelers who made the possible AI may be the first victims.
The upmarket of the data market lets emerge a social vacuum. Those who master the logic, code or fine evaluation of the models are courted. The others, deprived of income, are relegated to invisibility.
