Who Trains the Data for European Artificial Intelligence?: Report of the European Microworkers Communication andOutreach Initiative (EnCOre, 2023-2024)

Résumé 0

Executive SummaryMicroworkers: The Forgotten Platform WorkersOn February 22, 2024, a panel of microtask dataworkers (”microworkers” for short), assembledby the EnCOre initiative, addressed grievancesat the Fourth Transnational Forum on Alternatives to Uberization, held at the European Parliament. This event convened platform workers’unions, advocacy organizations, and policymakers to advocate for the adoption of the Platform Workers Directive. Attendees at the Forum expressed unequivocal solidarity with themicroworkers’ representatives, leading to the integration of their specific demands into subsequent public outreach efforts and documentation.Microwork refers to low-paid occupations,such as recording one’s voice, translating shortbits of text, and classifying contents in an imageor webpage. These tasks are essential functionsin the development of machine learning and AIbecause they prepare and correct algorithmic solutions. The Covid-19 pandemic, marked bylockdowns and job losses, along with technicalphenomena like the advent of generative AI, hasled to a significant rise in microwork across Europe.Microwork in Europe: What WeKnow So FarBased on our previous survey of more than 4,000small and medium-sized businesses in two European countries, up to three out of four companiesare aware of these platforms. Microwork platforms are increasingly utilized by companies inEurope to recruit, without committing to longterm contracts, a contingent workforce specializing in data-related tasks. Estimates of theglobal population of microworkers vary, rangingfrom overinflated platform reports claiming tensof millions of workers each, to more conservativebut still alarming estimates by the World Bank,which suggest that this phenomenon accountsfor between 4.4% and 12.5% of the global laborforce.Estimating the growth of microwork in Europe is challenging due to high turnover, evolving corporate practices, and trade secrets. Recent studies show a general increase in this typeof platform work. Although larger EU countries have more microworkers in absolute terms,smaller countries exhibit a higher density. Additionally, significant offshoring to regions likeAfrica, South-East Asia, and South Americamay have shifted demand dynamics for European microworkers.Recent data shows that while the majorityof European businesses are aware of microworkplatforms and data work deep labor networks,only a minority, mainly small businesses, admitto using them. Small businesses engage in microwork indirectly through purchased digital orAI services, with direct microwork clients beingprimarily technology suppliers.The EnCOre InitiativeThe EnCOre project, has been engaging platform users, workers of business process outsourcing (BPOs) firms, and technology companiesthrough interviews, focus groups, and surveysin nine European countries: Austria, Belgium,France, Germany, Ireland, Italy, Luxembourg,Portugal, and Spain. Between March 2023 anApril 2024, we have conducted exploratory interviews (N=7), focus groups (N=18), and fieldedonline questionnaires (N=577).ResultsOur survey reveals distinct demographic trendsamong microworkers, showing a significant overrepresentation of younger people with high education levels. The median age of participants is27, with an average age of 30.87 years, indicatinga younger demographic compared to previousinternational surveys. The significance of thistrend can be seen in the fact that 37% of respondents possess a bachelor’s degree and 27% hold amaster’s degree or equivalent. A high level of education among young microworkers correspondswith the recent increase in the percentage of 25-34 year-olds obtaining higher education degreesthroughout the OECD. Among our respondents,women account for 41% of microworkers. In Europe, there is a relatively smaller gender gap3when it comes to the participation of women inmicrowork.Microwork and UnemploymentPortuguese, Italian, and Spanish microworkersmake up the largest percentages of participantsamong the nine targeted countries, with respectively 33%, 20%, and 14%. Notably, all of thesecountries have higher unemployment rates thanthe average for the Euro area. While 34% ofrespondents hold a primary, permanent salariedjob, 28% are self-employed or in temporary employment. About 25% of those who engage inmicrowork do so because they lack other paidwork. Of these, nearly 15% are unemployed andactively looking for work—double the 6.4% unemployment rate in the Euro area.It is estimated that less than 20% of microworkers depend on platforms as their primarysource of income. However, this small but active group utilizes an average of 2.46 platforms,demonstrating the critical role these platformsplay in their financial ecosystems. Despite microwork being a crucial revenue stream, also forthose who use it as a secondary source, 75% earnless than 50 euros per month, with median earnings ranging from 20 to 25 euros. Moreover, interviews and focus groups revealed widespreaddissatisfaction among workers regarding wagetheft, with payments not only being irregularand unsatisfactory but also being withheld arbitrarily.Microwork and MigrationsThe results of our investigation demonstrate thesignificant participation of migrants in the European microwork market. 21% of the respondentswere born outside of their country of residence,and 18% hold a nationality other than that oftheir country of residence. There is potential evidence to indicate that migrants are engaging inonline labor platforms at higher rates than previously recognized. This is a departure from earlier findings that predominantly described migrants in Europe as mainly involved in locationbased platforms such as transport and deliveriesrather than online freelancing and microwork.59% of the respondents rely on a single platform for their income, suggesting a form of economic dependency, while 41% spread their efforts across an average of 3.42 platforms, indicating a strategy to diversify their income streams.It is noteworthy that almost 90% of these workers also engage in other online earning activities,internet sales, cashback operations, and participation in panels, which do not require specialized skills, and can be conducted on a temporarybasis.The above scenario contrasts with the concern expressed during focus groups that manymigrants have high educational levels but areunable to find jobs or earn good wages in theirnew countries. Many are confined to low-payingmicrowork despite their qualifications. The disconnect raises important questions about the integration and utilization of competencies amongmigrant workers in the digital economy in Europe, where they still seem unable to find highvalue online freelancing jobs requiring computerprogramming, design, and project managementexpertise. Moreover, fewer microworkers are involved in paid content creation and influence,which require significant time investments orspecialized training, emphasizing the difficultythey face in leveraging their educational background into more lucrative online employmentopportunities.The Human Cost of AIThe EnCOre study reveals that despite thetrend towards market polarization where highvalue AI startups are located in the GlobalNorth and data workers in the South, there isstrong development of a microwork ecosystemin Europe. In accordance with our framework,which distinguishes AI preparation, verification,and impersonation microtasks, we provide evidence that European microworkers are heavilyengaged in AI-related tasks. During AI preparation, workers generate or annotate data formachine learning applications, such as recording voice prompts or labeling images. DuringAI verification, workers are required to assessthe accuracy of AI-generated outputs, including audios and transcriptions. AI impersonationtasks, where human workers simulate artificialprocesses, are rare and typically performed internally. In these companies, workers are boundby non-disclosure agreements (NDAs), a practice that is uncommon among respondents to oursurvey recruited through online platforms, butubiquitous among workers in BPO companies.The use of these agreements restricts the freedom of expression of workers, a situation madeworse by the continuous monitoring of their online activities for the purpose of safeguardingproprietary information and assessing their performance.The psychological and emotional effects ofthese microtasks are profound on workers. Theyare often referred to as merely ”data tasks”but entail significant emotional engagement andpose significant socio-psychological risks. Thisphenomenon is particularly evident among microworkers involved in content moderation, whofrequently report adverse effects on their mentalhealth. This type of work has the potential toresult in severe consequences due to its intense4nature. Among them are incidents in whichworkers have fainted, suffered from burnout, experienced psychotic episodes, and, tragically, inat least one instance, committed suicide. Moreover, the isolation experienced by these workersis further exacerbated because of remote workor, in the case of shared workspaces, a generaldistrust among colleagues and corporate management. The combination of high-stress tasksand social isolation presents a challenging environment for microworkers, highlighting the needfor regulatory and support mechanisms in orderto mitigate these impacts.Policy RecommendationsAs policy-oriented initiatives predominantly focus on location-based platforms, microworkersare not adequately represented. Owing to thedispersed nature of their activity, they face psychological rather than physical challenges owingto the social isolation arising from the dispersednature of their activity, and in the case of content moderators, from exposure to all sorts ofonline toxicity. They are also more vulnerableto disclosures of their personal data and privacyviolations. Low remunerations, limited directcommunication with clients and platforms, anddifficulty to challenge payment withholdings arecommon complaints, but the pervasive misconception that microworkers choose this type ofwork voluntarily discourages efforts to improveworking conditions. This report highlights thatmicroworkers are indeed platform workers, andstresses that their distinct needs must be acknowledged in discussions and legislation pertaining to platform work.