After my phone interview with Troy Stringfield, my mind wandered. I envisioned an old Cadillac with massive Texas longhorns adorning the hood meandering along a dusty road. This road was in Egypt, and Stringfield was behind the wheel, sweat glistening on his brow as he hauled a load of freshly-baked sesame seed bagels.
No, I hadn’t been experimenting with some designer hallucinogen. But my conversation with him, as happens with particularly captivating sources, conjured evocative concepts and imagery, the kind of stuff that begs to be illustrated in word pictures.
Thing is, although his latest enterprise encompassed many of the issues I aimed to address in my most recent feature story in MIT Technology Review — such as fair labor in the AI industry, data ethics and the future of work — his background as a former Halliburton executive who became a bagel-making entrepreneur during his time in Cairo as an HR consultant with the oil giant never made it into the story.
As is often the way in long-term reporting and writing projects, some of the most bountiful anecdotes and the juiciest quotes are left on the cutting room floor.
If you haven’t read the MIT story, you should probably start there. This companion piece provides a glimpse into how I reported that story over a period of nearly a year.
A mile-a-minute gabber, Stringfield supplied lots of those memorable nuggets that didn’t quite fit into the final version of the story. Had I been writing for, say, Harper’s (ha!), a generous backstory guiding the reader toward the crux of the piece would have been more fitting. But, alas, this was Technology Review, and I had around 2,000 words to play with.
When I spoke with Stringfield earlier this year, he told me how he had established business links in Egypt during his time there with Halliburton. His natural propensity toward helping the less-fortunate (he made sure to tell me about his early do-gooder days serving spaghetti to the poor) had led him to spot what some might call win-win business opportunities. Ultimately, Stringfield represented a slightly more forthright example of the entrepreneurial crew I’d be writing about in my piece, all people who started businesses in the hopes of making a profit while helping people at the same time.
It was a concept I happened to be exploring while reading Winners Take All by Anand Giridharadas, a book that takes a critical view of the doing well by doing good philosophy.
At some point around 2018 Stringfield hitched his wagon to Alegion, a company in Austin that provides data services for AI tech firms. He spotted a business opportunity, one providing data-related jobs to the poor. You see, AI corporations pay firms like Alegion slivers of a cent per minute to provide data annotation services that prepare information fed into AI systems. The industry needs bodies, people willing to sit at computers and perform often-simple tasks, cleaning and labeling the data necessary to train AI systems to spot patterns, learn and get smarter.
They have conjured up corporatized approaches they believe will help solve intractable problems such as poverty and unemployment.
But while many companies do this kind of work, a small handful of them aim to fulfill a higher purpose. They call themselves “impact sourcing” companies, and like Stringfield and Alegion, they have conjured up corporatized approaches they believe will help solve intractable problems such as poverty and unemployment, particularly in the world’s most impoverished places.
A Reporting Journey Launched from a Cloth-Covered Hotel Conference Table
But it took me months to get to Stringfield. The seeds of this story were sown when I sat down with a colleague of his from Alegion at an AI industry conference in San Francisco in September of 2018. About a month before I had decided to build a reporting beat covering issues related to artificial intelligence ethics. I wanted to understand what ethical AI is as both a concept and a movement, who wanted to influence it, and what issues and dynamics drove the conversation around it.
I figured attending an AI industry conference would help me get a sense of how corporations were addressing ethics subjects from within a cutthroat AI market.
Not surprising, there was a lot of soothsaying involving China superseding the US in the oft-referenced AI Race. Despite a keynote from Meredith Whitaker of AI Ethics advocacy group AI Now Institute and a couple small-room panel discussions involving ethics issues, the impression I walked away with was that the industry was probably too driven to beat China to concern themselves with ethics on any level beyond press releases, lip service and sparsely-attended event panel discussions.
I had to wonder, if China can’t be bothered to build and implement this tectonic-plate-shifting technology in a thoughtful, ethical manner, why would we expect AI tech firms beholden to venture capitalist investors in the US to do so?
Stringfield’s colleague at Alegion and I found our way to a cloth-draped table in the lunch area of the conference where he told me about the data services the company provided. He suggested the firm’s head data scientist would be able to answer some of my more detailed questions involving ethical approaches to AI development and data use.
If China can’t be bothered to build and implement this tectonic-plate-shifting technology in a thoughtful, ethical manner, why would we expect AI tech firms beholden to venture capitalist investors in the US to do so?
Sometime the following month, I got on the phone with Cheryl Martin, Alegion’s chief data science officer. In my time covering the data industry for Advertising Age from 2012-2017, I spoke to a lot of data scientists, nearly all of them men. It was a welcome change of pace to chat with Martin.
We discussed her work and background at NASA, before diving into table-stakes AI ethics topics like algorithmic transparency and bias in AI. I used her as a source in one of the stories I published to launch RedTail in October 2018, Can We Disrobe Our Algorithmic Emperors?
Delivering Bagels… and Data Jobs
A month or so after I spoke with Martin, I read a story in the New York Times about data workers in China, titled How Cheap Labor Drives China’s A.I. Ambitions, and thought of Alegion.
This somewhat simplistic story described the experiences of data labelers and annotators using a manufacturing industry metaphor. Laborers toiled in what were referred to throughout the piece as “data factories,” despite the fact that they were photographed hunched over computers in a carpeted office rather than hunched over car parts on a cement floor. The story implied that Chinese AI firms have the ability to employ a powerful local army of data workers that the US cannot, giving them an advantage.
“It isn’t clear that the A.I. race is a zero sum game, in which the winner gets the spoils,” noted the story, by Li Yuan. “But the ability to tag that data may be China’s true A.I. strength, the only one that the United States may not be able to match.”
Was the implication that this work was only being performed in China representative of what was actually happening, I wondered? After all, wasn’t this the sort of stuff Alegion did? I reached out to Martin, and we set up another call for December.
“It’s definitely not the case that it’s only happening in China,” she confirmed, explaining that outsourcing data jobs and data processing had been going on for a long time across several fields. We chatted awhile and eventually she mentioned what she described as a “fairly new” effort at Alegion to partner with nonprofits and other groups to hire underprivileged people in the US and elsewhere to do data annotation work. This “fairly new” effort, as it turned out, was what Alegion calls its “impact” business.
Enter Stringfield and his Egyptian bagels.
This was the new business they hired him to expand, and I wanted to learn more. Alegion hooked me up by scheduling a call with him the same month. It was a fascinating, whirlwind of an interview. He told me about how while he was working for Halliburton in Cairo, his wife missed the taste of the bagels she used to munch back home in New Jersey. So, they learned how to make them — and voilà — a food service firm was born.
“We fell into this side thing and the next thing you know we’re, like, delivering bagels,” he told me. He named the “side thing” Jared’s Bagels, after his eldest son, a hat-tip to the Egyptian tradition that holds the oldest male child in high esteem. Stringfield wanted the business to benefit impoverished youth in the country. He told me, “We hired orphans. We would bring ‘em in from upper Egypt.”
“We fell into this side thing and the next thing you know we’re, like, delivering bagels…. We hired orphans. We would bring ‘em in from upper Egypt.”
– Troy Stringfield, Alegion
When eventually he connected with Alegion, the serial entrepreneur saw another opportunity. Rather than AI replacing jobs, Stringfield anticipated a growing industry creating them. “I thought about all jobs to be created among the poorest, because they’re the only ones that fit the business model.”
In other words, they fit the business model because the data industry fueling AI technologies relies on cheap labor. “I mean, to train AI you have to not have thousands or tens of thousands or hundreds of thousands but millions of records,” he said.
Today as Alegion’s global impact director, Stringfield has assembled a team of 45 people in Cairo who do data annotation work. His goal is to build an operational structure that can be replicated, franchise-style, at future “impact centers.”
“It’s a total win-win because, you know, as Alegion grows it needs more on the supply side, more and more workers to do work,” he told me.
A Depressing Race to the Bottom
Again, my MIT Tech Review story describes in more detail the data work needed to train AI and why so much of it is done by people overseas earning far less than what US or European workers would require. Suffice to say AI companies are in a race for dominance, which sets up the market in such a way as to deemphasize the very real people behind automated technologies, and their need for fair pay and work conditions.
AI companies are in a race for dominance, which sets up the market in such a way as to deemphasize the people behind automated technologies, and their need for fair pay and work conditions.
At this stage I had an inkling of this concept, but had a lot more reporting to do. I wanted to talk with someone on the ground who understood the challenges of offering fair work opportunities in this demanding industry from a more logistical standpoint.” I spoke with Daniel Kaelin, Alegion’s director of customer success, via Skype in March from his home in Malaysia, where he oversees the bulk of Alegion’s workforce the company calls “data specialists.”
A couple times during my talk with Kaelin, he lamented the fact that the business of providing data annotation labor meant operating in a nickel-and-dimed world of price-undercutting that resulted in lower and lower wages for workers. Twice he referred to that aspect of his work as “depressing.”
“It is really a race to the bottom,” he told me, bluntly.
Daniel Kaelin’s admission was telling in a stark and transparent way most interviews with sources — particularly business executives — never are. Twice the Alegion exec referred to price-undercutting in his business as “depressing.”
Kaelin’s admission was telling in a stark and transparent way most interviews with sources — particularly business executives — never are. But his description of his work as sometimes “depressing” never made it into the final cut of my MIT story.
As a refresher, my interview with Kaelin came six months after I spoke with my original Alegion source, and three months after I first discussed the firm’s impact efforts with Martin. Meanwhile, despite pitching a couple concepts around to editors, I had yet to find a publication interested, in part because I probably hadn’t quite isolated what the story was yet. Still, I knew there was something there.
Uncovering a Burgeoning Industry, and a Wider Story Lens
As I dug into my research, I realized Alegion’s efforts to provide data jobs to people in need were not unique at all, and neither was this term, impact. A handful of other firms including iMerit, CloudFactory and Samasource, had made it their mission to connect data work opportunities to people in need in places like India and Africa.
These companies believe the hidden human labor needed to produce AI translates into job opportunity for farmers in Uganda, moms in Kuala Lumpur or autistic people, military vets or others in the US who are marginalized or have special needs.
The concept of impact sourcing is an offshoot of business process outsourcing — think of call centers or data entry services based in emerging markets. The idea is to hire people who have been unemployed for extended periods or living under the national poverty line.
Stringfield, Alegion, and these other companies believe the hidden human labor needed to prepare data to produce AI translates into job opportunity for farmers in Uganda or moms in Kuala Lumpur. Or, for autistic people, military vets or others in the US who are marginalized or have special needs, this work can serve as a launchpad for something more advanced and permanent.
This impact thing would give my story the wider lens it had been missing. Eventually, I’d speak to representatives of iMerit, Samasource and Daivergent, a company that connects AI firms to autistic data workers such Leon Campbell, the young man I wrote about in my MIT story lede. But the continued research and interviews also surfaced several additional reporting tangents that would require my own self-editing as well as the guidance of some expert editors, which MIT TR provided.
Talking through Ideas to Reveal Kernels of Truth
Around the same time I talked to Kaelin, I stumbled on an Amazon page dedicated to a soon-to-publish book addressing some of the very issues I’d been discussing with sources. I got in touch with Mary Gray, an Indiana University professor, Microsoft Researcher and co-author of the book Ghost Work, which is about the labor issues associated with the gig economy.
When we talked she helped me encapsulate an idea swarming around my reporting thus far: Amid the cutthroat competition of the AI market, the raw materials of machine learning and artificial intelligence – in this case data and the labor of a remote workforce employed to prepare it – are at risk of being undervalued as mere commoditized expenses on a spreadsheet.
Indeed, this is one of the problems that these impact sourcing data services firms could help alleviate. But, as is the case with so many deeply-reported stories, more research led to more questions.
What I realized I needed to understand and address in my piece were the dynamics at play when corporate entities attempt to solve social problems within the profit-driven construct of a market. Can it be done, and who defines success?
What I realized I needed to understand and address in my piece were the dynamics at play when corporate entities attempt to solve social problems within the profit-driven construct of a market. Can it be done, and who defines success?
Despite the nuances I perceived as I fell down the reporting rabbit hole, Stringfield seemed to see it as a black-and-white situation. “The success of Alegion means the success of people in need around the world. The more Alegion succeeds the greater impact they can make,” he told me, plain-and-simple.
The thing is, it’s not that simple, not by a longshot, and I aimed to explain why in my MIT story. That article is the one that’s really worth reading. And, perhaps, if anything, this behind-the-scenes companion piece offers a glimpse into the amount of work that goes into investigative freelance reporting and analysis.
Or, perhaps, it’s only left you with a craving for an everything bagel and a smear, and hey, I can’t blame ya there.
0 comments on “Reporter’s Notebook: Behind the Scenes of a Fair-Trade AI Data Story”