January 2023 survey by time The magazine revealed that Kenyan workers earning less than $2 an hour were given the job of ensuring that data used to train AI platform ChatGPT does not contain discriminatory content.
For AI models to learn how to recognize and interact with the human environment, they must be trained with a critical mass of input data. These inputs must be collected, sorted, validated, and formatted. Such time-consuming and undervalued work is typically outsourced by technology companies to armies of precarious workers, usually based in the Global South.
This data work can take several different forms, depending on the goal of the final algorithm. For example, teaching an algorithm how to recognize humans may involve outlining people in images captured by a video camera. Or you might check the output of an automated invoice processing tool and manually correct errors to help the computer do its work.
To explore the identities, roles, and working conditions of these data workers, and to enrich the debate about regulation of the AI sector, we have launched an investigation between Paris and Antananarivo, the capital of Madagascar.
Our research also shows the reality of AI the French way. French technology companies, on the other hand, rely on the hosting services and processing power of the Big Five (Google, Apple, Facebook, Amazon, and Microsoft). Another is the data work performed by workers in former French colonies, particularly Madagascar, confirming the established trend of outsourcing. By the way, there have already been studies comparing the technology sector with mining and textiles.
Research on globalization of AI
Our research project began in March 2021 in Paris. We begin by looking at how the French AI house is involved in data work activities and how they ensure that datasets of sufficient quality are generated for training computer models. We set out to understand what processes are in place. We conducted interviews with 30 founders and employees of 22 of his companies in Paris that are part of the AI ecosystem. From this initial investigation, one discovery quickly became apparent to him. Most of the data work was outsourced to Malagasy contractors.
In the second part of the study, we interviewed 147 workers, managers, and directors from 10 Malagasy companies, first remotely and then on-site in Antananarivo. At the same time, he sent a survey to his 296 data workers based in Madagascar.
Precarious work among highly educated urban youth
Our initial research shows that AI data workers are part of a much wider range of IT services sectors, from call center staff to web content moderators to search engine optimization (SEO) copywriters.
According to survey responses, the majority of workers employed in this sector are male (68%), young (87% under 34 years of age), urban, and educated (75% have some qualifications). higher education). If the job was performed within the formal economy rather than the black or gray economy, respondents were typically full-time employees. The minimal protection provided by Madagascar, in contrast to French employment law, ignorance of workers’ rights, and the weakness of trade unions and worker representation in Malagasy companies increases the precariousness of their position. I let it happen. Their monthly salaries typically range from 96 to 126 euros, with a wide gap between them and the team manager’s salary. Many of the directors are Malagasy and work in the country, but their take-home pay is 8 to 10 times that amount.
Workers on the ground find themselves at the end of a very long outsourcing chain, which partly explains why their wages are so low even by Malagasy standards. The AI production line includes three different players, he said: data hosting services and processing power provided by the Big Five technology companies, a French company selling AI models, and a company providing data annotation services provided by workers in Madagascar. Involved. Each level plays a role.
The companies interrogating the data generally rely heavily on French customers, who manage outsourced workforces in a near-direct manner, and whose middle managers manage their outsourced workforces with the interests of Parisian startups in mind. It forces you to work. The monopoly of these roles by foreigners employed by French client companies or working in Antananarivo is a serious impediment to career advancement for workers who are disgracefully stuck at the bottom of the value chain. There is.
Benefiting from post-colonial France-Madagascar connections
The AI sector benefits from a specific policy called the “duty-free zone” created in 1989 for the textile industry. Since the early 1990s, we have been setting up satellites in Madagascar for French companies, especially in the digital publishing industry. Many other developing countries have similar special zones, which attract investment by offering very attractive tax exemptions.
Currently, of the 48 companies providing digital services in the duty-free zone, only nine are Malagasy-owned and 26 are French-owned. Apart from the situation of formally established enterprises, the sector has developed a practice of hierarchical subcontracting, with gray economy enterprises and entrepreneurs at the bottom of the pecking order, receiving poor treatment and Called into action when labor is in short supply elsewhere.
In addition to cheap labor, this outsourcing industry also benefits from a highly educated workforce. Most have gone on to university and speak fluent French, learned in school, online or in classes at Institut Français. Founded in 1883, this latter institution of introduction to the French language and culture was initially intended to extend imperial power over colonized populations through language.
This scenario is consistent with what researcher Jean Padios has termed “colonial reminiscence.” Former colonies with linguistic and cultural ties to once influential countries now provide business services.
Visualize your AI workers to better understand how they work
The recent surge in commercialized AI projects in the Global North has been driven by a growing number of data workers. The recent controversy over “intelligent surveillance cameras” at the Paris Olympics focused primarily on the ethics of full surveillance. We need to better account for the critical element of human labor involved in training AI models. Not least because it raises new questions about working conditions and rights to a private life.
Visualizing the role of these workers requires asking probing questions about globalized production chains. These are well known in the manufacturing industry, but they are also characteristic of the digital sector. These workers are essential to the functioning of our digital infrastructure and are the invisible cogs of our digital lives.
It also provides visibility into the impact of your work on the AI model. Part of algorithmic bias lies in the nature of how data processing is performed, but this reality is largely kept secret by AI companies. Therefore, a truly ethical AI must set ethical standards for working conditions in the AI sector.
Le projet The human source behind smart technology – French National Research Agency (ANR), HUSH EST soutenu for French research agency funding. We devote our mission and drive to educational research and final examination for dialogue between science and society, and to strengthening dialogue between science and society. Add some fun and check out ANR’s site.
Source link
