- by C Bartholomew
- The Guardian
- Issue #2186
Photo: publicdomainpictures.net (CC0)
AI tools and platforms are often touted as bringers of automation. Companies discussing AI focus the conversation on digital elements such as machine learning, data sets, and algorithms. However, AI is more than just fancy software. It is a technology with physical impacts. The land, water, energy, and mineral demands of AI systems are gaining increasing attention in public debate. Another demand which is still largely hidden is the large number of underpaid, mostly African workers vital to processing the data behind AI.
BEHIND THE CURTAIN
AI systems are, in fact, not artificial intelligence at all. These systems do not think, solve problems, or understand data. They are essentially more advanced versions of your phone’s predictive text. AI systems are fed mountains of data, which they then use to produce a plausible output in response to a user’s input. In other words, they take whatever is handed to them and create what it’s most likely that the user wants to see based on patterns. This process of analysing data is not fully automated, but instead reliant upon thousands of human workers.
In countries such as Kenya, data labelling companies, subcontractors to big tech corporations such as Meta, Microsoft, and OpenAI, employ thousands of people to train AI systems. These data labellers, data annotators, or content moderators are paid a cent or two per task to label objects within images, transcribe audio, describe videos frame-by-frame, and perform other tasks essential for AIs to function. These workers are also used to verify responses and perform tasks AI struggles with – like scheduling appointments, and mitigate biases and disturbing content. This last job is why many of these workers are regularly subjected to hour upon hour of graphic violence, child exploitation, and pornographic materials.
PRIVACY
Users of AI systems may unknowingly be involved in these human processes. Some data labellers have reported that, while ‘training’ AI companions, chatbots, and sexbots, they were actually being made to sext with users in real time. Other labellers involved in processing data for Meta’s AI glasses have reported seeing videos direct from users, including videos of people using the bathroom, having sex, or even showing private information such as credit card and banking details. Annotators – who have to verify the answers that Meta’s glasses and other AI systems provide – report reading transcriptions about crime, protests, suicidal thoughts, and sexual fantasies.
Users may be told that they have a choice regarding which data is sent to companies and which is kept private, but the truth is that for the AI systems to work at all data must be sent to company data centres. If the data being sent to third party companies for labelling and training is sourced from these same centres, then it is unclear how user privacy can be preserved at all.
SILENCE, INTIMIDATION
Rather than act responsibly or be held accountable, tech companies simply bully their hidden workforces into silence. Workers are often managed by algorithms, meaning they have no idea which big tech company they are working for. Labellers are made to sign brutal non-disclosure agreements (NDAs), while offices are under heavy surveillance. Michael Geoffrey Asia, secretary general of Kenya’s Data Labellers Association (DLA), has described these NDAs as “a slave tool used to enslave people to not speak about what they’re going through.” The workers’ low pay, disposability, and often poor access to services such as healthcare also encourages silent compliance. Human rights lawyer Mercy Mutemi has reported that when countries such as Kenya seek trade deals with the US they are pressured into providing immunity for big tech companies.
Organisations including the DLA have emerged as data labellers are increasingly speaking out against low pay, for an end to NDAs, and for mental health services for those workers subjected to violent, sexual, and disturbing materials. AI companies rely on the public perception that their systems are fully automated. This lie not only helps attract investment, but it also obscures the reality that these systems depend on human labour.
AI has become the latest episode of imperialism, as tech companies exploit vulnerable workers in countries dependent on Western capital. These workers are the basis of AI companies’ massive profits, yet they receive only the tiniest fraction of that wealth. AI’s mystique is a purposeful façade to keep that reality secret.
