What the Rise of Artificial Intelligence in the Global South Tells Us About Our Future
 |
“A lot of AI innovation is going to come out of the Global South.”
I had only half paid attention to this woman’s presentation on software that uses artificial intelligence (AI) via cell phone camera to scan and digitize handwriting from paper health forms in Malawi, when I was suddenly transported back to an undergraduate anthropology class and a book titled Theory from the South (2012).
This book, written by anthropologists Jean and (now-disgraced) John Comaroff, in part posits that the Global North (Europe and America) is actually evolving to be more like the Global South (Latin America, Africa, parts of Asia). The authors go on to argue that the Global South has been treated like a “laboratory” of capitalism – where new configurations of industry, labor, and regulatory environments are tested and honed before being exported to the Global North.
This can be seen in the rise of “the gig economy” in the US, or microlending via “buy now, pay later” apps , and the rapidly widening wealth gap. This is why we in the Global North, need to be more conscious of the technologies that emerge in the Global South, because “it is the Global South [that] affords privileged insight into the workings of the world at large” (1). In particular, we collectively need to turn our attention towards is the way that the testing and training of AI in the Global South prefigures how AI is going to manifest in the Global North.
First, this requires we acknowledge that the “use” of AI in the Global South is exploitative and that the Global South is not the region that is going to benefit the most from this technology -despite a big portion of the technology is being developed there. Tech companies are purposefully using the Global South to train their models due to a regulatory and labor landscape that isn’t present in the Global North and that would both prevent the work from happening and would increase the cost of developing AI systems.
AI requires a massive amount of largely invisible human labor to seem fully autonomous – anthropologist Mary L. Gray calls this “ghost work”. Much of this takes the form of data annotation, where workers label data so that the computer can learn what different texts or images are (like those CAPTCHAs that ask you to click on every square that contains a bicycle - the computer needs a person to label those images first so it can learn what a bicycle looks like).
In the Global South, AI data annotators are in the equivalent of a digital sweatshop - doing repetitive piecework under constant surveillance by their employers, who push them to complete tasks at a rapid pace.
A TIME Magazine investigation found that OpenAI outsourced the worst data annotation tasks – using human labor to flag toxic language from the training sets used by ChatGPT – to a Kenyan company and paid their workers less than $2 an hour to quickly sift through pages and pages of content that contained graphic descriptions of bestiality, sexual abuse, violence, and torture.
There are also allegations that OpenAI was paying that same Kenyan company to collect images that are illegal in the US to better train ChatGPT.
There is no federal minimum wage in Kenya, and though some cities have minimum wages, the TIME journalists found that lowest level data annotation workers were still making less than the minimum wage for a receptionist in Nairobi.
Even the technology that was presented at the conference is exploitative, even if it could be beneficial to public health efforts. It was trained on “local handwriting samples” and tested locally meaning that this private company is extracting value from the local people – positioning them as “unprocessed data” from which they can squeeze every bit of value they can as they develop a product to sell in wealthier countries (1).
There seems to be this belief that those in the Global South only deserve the experiments, the pilot programs and never the finished products. That providing the people living in the Global South any technology at all is good enough. Especially when viewed against the backdrop of what the histories of colonization have done to these areas by the nations developing these technologies, the continued exploitation becomes clear.
This is especially maddening when this paper form scanner is not a technology that would be implemented in the Global North until it is extensively tested, and all the kinks are worked out. There are still too many unanswered questions about cyber security, the risk of misdiagnosis, and data governance that are unanswered but are somehow acceptable for use with the most disenfranchised.
There are many health clinics in the United States that are reliant on paper forms, have the internet connection required, and would benefit immensely from this technology. So why did this company choose Malawi for their training and piloting?
---
 |
| Photo by Johanneke Kroesbergen-Kamps on Unsplash |
During her talk the presenter described how it was one person’s entire job to transport the completed medical forms by motorcycle to a central location where another person would spend their day entering that data into the computer. When she went on to explain how this new technology could do all that data entry and free up those employees for other tasks, I couldn’t help but wonder if these workers will be kept on the payroll once their primary responsibilities are essentially eliminated.
Outsourcing of labor to cheaper sources is a key feature of neoliberal capitalism – typically understood as the offshoring of factories from the US to the Global South in search of cheaper labor. When those factories moved, we also saw the Global North’s labor shift to be more like the Global South via the reduction in union power – part of what drew the factories there in the first place. Now we are seeing how companies from the Global North are turning towards the cheapest labor possible – robots – who don’t require breaks, don’t take time off, don’t get pregnant, don’t get sick, don’t miss work, don’t unionize – who aren’t protected by labor laws.
There are already inklings of how this is going to manifest in the Global North, where AI is going to replace tasks performed by higher-earning, white-collar workers like data analysis, financial reporting, and software development. Already, repetitive administrative tasks are all on the chopping block as more and more workers use AI to take meeting notes, send email reminders, and setting up recurring meetings.
In the Global North there are already rumblings among white collar workers that we either learn to use AI, or we risk liquidation, but that doesn’t address the fact that, for some positions, there likely isn’t a way the worker can “upskill” enough to prevent liquidation.
In epidemiology, there are many who are enthusiastic about the use of AI in our field. Some of us have been using a type of machine learning for outbreak detection since before ChatGPT took over the popular imagination. Syndromic surveillance, at its most basic, is a passive surveillance system of emergency department visits that is monitored for spikes in visits with similar symptoms that could signal an outbreak. Some platforms, like the ESSENCE platform from the CDC, encourage users to utilize algorithms to determine if there are more visits than usual with similar symptoms. However, the algorithms are not terribly useful, and their creator refuses to share information regarding how these algorithms work, as the patent for these algorithms continue to languish in patent application limbo.
Still other epidemiologists think we should use AI for things like selecting data analysis approaches, choosing packages in analysis software. Any epidemiologist or data scientist who has used AI also understands (or learned the hard way) that at this moment in time, generative AI should not be used to write code. Even the simplest “if – then” statements produced by AI either doesn’t work, doesn’t work as intended, or is wildly inefficient.
But one thing all epidemiologists tacitly understand is that we should not be feeding health data into AI, especially if that data has identifiers like names, dates of birth, or (God forbid) social security numbers. Clearly there is no such compunction among the tech companies piloting these systems in the Global South.
As tech companies set their sights on developing AI that can produce working code, they are going to need data annotators who have the necessary technical background to recognize when code produces the intended results or not. This is where the Global North turns toward the South.
I predict that soon there will be an expansion of ghost work into the Global North as white-collar tech workers find side gigs (or main gigs) annotating data, especially if the minimum wages in the Global North continue to fail to increase with inflation (the federal minimum wage in the US as of writing has been $7.25 an hour since 2009) and as more and more people try (with mixed results) to find lucrative careers in the tech industry amidst tech layoffs. Already there are data annotation companies in the US specifically searching for annotators with coding backgrounds.
In this gig economy, where the younger generations are saddled with immense amounts of student debt, unaffordable housing, and a general increase in the cost of living, this sort of casual, piecemeal work could be an attractive supplement to stagnating monthly incomes. And with the general negative attitudes in the US towards workers, it is likely we will begin to see more and more ghost work.
-----
 |
| Photo by Igor Omilaev on Unsplash |
AI feels like such an avatar of millennial capitalism – and I don’t mean capitalism propped up by skinny jeans, avocado toast, and crippling student debt. I mean what the Comaroff’s describe as “capitalism invested with salvific force, with an intense faith in its capacity, if it rightly harnessed, wholly to transform the universe, including the lot of the most marginalized, immiserated, and disempowered” (159).
When the AI craze began in earnest with the roll-out of ChatGPT, there were more than a few technophiles claiming that this was the beginning of an era, where AI was going to eliminate all forms of prejudice, lead to economic growth, improve the quality of life and justice for all who reside on this rock orbiting the sun. We’ve heard this about microlending, social media, iPhones, nuclear power, and genetically engineered organisms, yet all have failed to deliver.
However, there is a way forward. As this Noema Magazine article states, “solidarity between highly-paid tech workers and their lower-paid counterparts — who vastly outnumber them — is a tech CEO’s nightmare.” And cross-sector organizing has worked before.
Organizing by Amazon employees started with two user experience employees and extended from the corporate office in Seattle to warehouse workers across the globe. I encourage those of us who work in fields that are going to be involved with AI in the near future to be sure we are including in our organizing efforts the very people who made the AI possible and demand more equity within AI – from unbiasing the technology to fair wages for the data annotators. AI is not a technology that is going to save the world, but we can demand that it makes the world and little fairer for everyone.
Comments
Post a Comment