A popular assumption of the digital world is that it is democratically built and run, largely due to its mass participatory nature. An invention of modern society, however, it is bound to replicate the asymmetries, biases, and conflicts of the physical world. This is particularly true of the systems underlying the construction of digital infrastructure and devices.
In Algorithms of Oppression, Safiya Umoja explored how data discrimination led to biased algorithms that discriminated against women of colour, and privileged whiteness. In Weapons of Math Destruction, mathematician Cathy O’Neil explored how the use of big data tools in fields like policing and insurance perpetuated inequality and prejudice.
What these instances show is that the digital space is not inherently democratic. And what is now emerging is that the unseen processes at the back-end that develop these systems are also tainted by inequality and bias. In late-January this year, Time magazine ran an exposé on OpenAI, an artificial intelligence (AI) research company responsible for the well-known AI chatbot ChatGPT and art generator DALLE-E. The exposé detailed a 200-strong force of underpaid and exploited content moderators in Kenya who were used to label harmful texts for the ChatGPT software. This news affirms one truth, that the global division of labour has been co-opted by the forces of the digital revolution and have, once again, left the African continent as a site for labour exploitation for the benefit of Northern capital interests.
It Takes a Village to Raise a Chatbot
GPT3 is an all-natural language generator that has been used for many functions, including powering the Philosopher AI chatbot and teaching Mandarin through GPT-3 Grandmother. By sourcing linguistic data from the worldwide web, OpenAI built GPT3, ChatGPT’s predecessor.
As the new vastly unregulated public square, the internet, unsurprisingly, duplicates the same toxic characteristics of the physical world: it is, after all, people who inhabit it and regurgitate their opinions, beliefs, and biases onto chat rooms, open forums, Twitter and other similar platforms. Using the language data found on all corners of the web, the AI chatbot would frequently spew out obscene, hateful, sexist and racist responses to prompts.
In comes Sama, an AI company that signed three contracts worth $200,000 with OpenAI over three years to label textual descriptions of hate speech, sexual abuse, and violence in 2021. Sama, a San Francisan firm with an outpost in Kenya, employs locals as well as Ugandan and Indian workers to label data for Meta, Google, and other Silicon Valley tech-giants.
The labelling process introduced a tool that would be built into the AI bot, helping it to classify and filter harmful texts to minimise the production of harmful content. This was an important step in the functionality and success of OpenAI’s next generation chatbot, ChatGPT, which went on to gain 100 million users within its second month online.
Undoubtedly a breakthrough innovation acknowledged by OpenAI’s spokesperson as a “necessary” tool in combating hate speech in AI systems, it appears that its utility didn’t extend to adequate compensation for the labour behind its functionality. According to the Time investigation, Sama workers were paid wages ranging between $1.32 and $2 per hour “depending on seniority and performance”.
According to a 2018 BBC report on Sama, approximately 75% of the corporation’s employees are from the Nairobi informal settlement of Kibra. Known as one of East Africa’s largest slums, Kibra has a 50% unemployment rate and appalling living conditions characterised by a lack of clean water and sanitation facilities. With more than 10% of the city’s population unemployed, the need for work – any work – in Nairobi is high.
Sama’s employees were contractually required to label 70 text samples in nine-hour shifts. Many disclosed routinely labelling between 150 and 250 texts per shift. Due to the explicit nature of filtered text, interviewed employees suffered from mental scarring, torture, and nightmares.
Sama ended its contract with OpenAI, resulting in dozens of demotions and job losses. Since then, Sama has announced that it would be terminating all its sensitive content related work by March 2023. The consequences have been similar to that of the closure of the OpenAI contract Sama had with Meta.
Global Division of Labour and the Silent Labour of the Continent
This outsourcing of labour to circumvent high wage expenditure is not new to the digital revolution. Just last year, Meta was sued in Kenya for union-busting and contracting the same company, Sama, which subjected workers to inhumane working conditions. And Congo’s cobalt rush has pushed child labourers into the mining sector en masse for years now.
In response to the Time article, Sama responded by claiming to pay “ between Sh26,600 and Sh40,000 ($210 to $323) per month”. If this is true, it would equate to an average of $12 per day, while the comparative rate in the US would be around $30 to $40 per hour. The wage expenditure for Sama and its clients is significantly reduced through this form of outsourced sub-contracting. The rewards are high and the risks are exponentially low for the suits in Silicon Valley and Silicon Harbour.
What ChatGPT exposes is the persistence of an established global division of labour. In doing so, it also contradicts bold claims about how the digital revolution is reshaping the global workforce. Of course, the digital space offers opportunities for the upscaling of skills and the re-distribution of wealth particularly through the global gig economy.
However, the space for these systems to be replicated lies in the invisible labour which props up digital infrastructure. When your chatbot functions properly, there is no thought of what sort of labour enabled that functionality. There is no thought of what an unfunctional bot might even look like. Similarly, when your smartphone battery runs smoothly, you have no worry of its internal make up, let alone where those materials are sourced from. In an era where one global crisis or another is one search bar away, it is a rarity to find an area of society that is not under the scrutiny of the normative gaze. When forms of labour are organised to exist outside the panopticon, the reason is usually to allow abuses to occur.
Challenges and Opportunities
In my perspective, this entire situation does not mean that Africa must be relegated to its perceived status of subject rather than agent on the international stage. If anything, we must recognise the plethora of opportunities that the digital revolution has available for each and every nation.
Labour laws must, firstly, be strengthened and adjusted to the times. If there is one thing that multinational corporations are good at, it is identifying the perfect cracks in the system that can become soil for its ever expansive roots. Anachronistic laws serve no purpose but to lay this foundation.
Secondly, the national auditing of the activities of multinational corporations must also be done diligently and with the best interest of local labourers at heart. If investment is considered to be a major benefit for international entrepreneurial cooperation, they must invest in – not exploit – the labourers who strengthen and add value to their operations.
The biggest challenge to either of these options is, however, perhaps the modernisation of international trade law to encompass the digital economy. The liberalisation of digital goods and service trade has been globalised, limiting barriers to digital trade. It has also enabled easy and low-cost data flow across the world, encouraging the outsourcing of the production and trade of these goods and expanding global supply chains.
The shift towards a more liberal international trade order has obscured the role of the state in regulating, controlling, and shaping global data flows. All this while the regulatory frameworks governing these flows and the subsequent integration of new labour markets in the global map of said data flows remain underdeveloped. In this inchoate policy space, the ability of states to protect workers against the inherent tyranny of trade agreements that privilege the right of capital to migrate unfettered must be reinforced.
How can Africa gird itself from the emerging digital tyranny? One way might be to revive its old penchant for continental alliance-making. Borrowing from Howard French‘s thinking after the US-Africa Summit, where he implored Africans to speak with one voice, or at least fewer voices, there has never been a more urgent moment for the pooling of collective African power.
While presenting a united front, especially during trade negotiations, may be the surest way to avoid reproducing the gross imbalances of the current labour and trade systems, this will be tested and likely undermined by the usual attempts to balkanise the continent along the usual neo-colonial lines of regionalism, colonial linguas and aid inducements. Transcending these traps might sound utopian, but doing so could grant Africa immense headway in using the rules-based international order to increase its gains during the digital revolution.
At the end of the day, the shift from one arena (physical) of living to another (digital) is not the panacea to centuries-old systems of division and oppression. New technologies develop their own axes of exploitation that are systematically incorporated into social, geographical and economic relationships. Vigilance is always required.