Hungry ghosts: Appen’s woes threaten the low-paid workers’ livelihoods
Former local tech darling Appen has just endured a horror year. Now with the global technology sector slipping into a rates-induced funk, its travails also threaten the livelihoods of its legion of low-paid “ghost workers”.
Appen, which provides the data used to train machines to recognise basic images and speech, landed on the ASX in 2015. It soon shot up the ranks in the Australian technology scene, growing rich providing the grunt work crucial to the development of artificial intelligence (AI)technology.
With its shares trading as high as $43.50 in 2020, Appen’s fall from grace has been chastening, its shares are currently trading at $3.37. And its headaches boil down to one factor – its reliance on the likes of Microsoft, Google and Facebook for work.
Weakness in the digital advertising market forced Appen’s customers to pull the brakes and subsequently crush Appen’s blue-sky growth story – and investors. Appen’s value has been shredded, from $5 billion barely two years ago to currently less than $500 million. Some market analysts think its true value may now be under $400 million.
Appen sold itself as a guaranteed bet on the rise of AI. Every time you use a search engine, social media, or make a request to Apple’s Siri, chances are you are benefiting from the work Appen does. But the model involves using cheap labour to manually sort and tag videos and train AI algorithms.
While the tech giants, which account for 80 per cent of Appen’s revenue, were spending up all was rosy, but things turned sour as the pandemic took hold. Appen rolled out a series of downgrades and investors piled out of the stock.
Company chief executive Mark Brayan said at the time that while the tech giants had cooled their heels temporarily the ‘megatrends around AI’ remain intact, and continue to benefit Appen.
However, his predecessor – former Appen CEO Lisa Braden-Harder – says while AI is here to stay Appen’s financial heath hinges on a tiny customer base.
According to Braden-Harder, the bulk of Appen’s business is with two core customers.
“We’re basically talking about Facebook and Google here,” she says. And now these businesses are taking a hit as the global economy slows digital advertising growth, and Apple’s new privacy features also hit Facebook ad revenue.
In simple terms, Appen’s major customers sneezed and Appen caught a cold.
Analysts covering the company have responded brutally, and none more so than JP Morgan’s Bob Chen who slashed its valuation of Appen late last month to just $3. Appen hasn’t traded at those levels since 2017.
“Its largest customers are starting to feel the impacts of weaker macro [economic conditions] and have started reducing investment spend, and this has led to a significant decline in Appen’s core revenues and we have limited visibility on when this could improve,” Chen says.
Meanwhile, Macquarie analysts have cited further potential downside risk from competitive pricing pressure as well as the risk of big tech reducing its reliance on outsourcers like Appen.
And Appen’s woes obscure another crucial issue about its future: the ethics of the crowdsourcing it engages in.
The issue was brought to a head early this year when the company featured prominently in a MIT Technology Review series. The series explored the idea that the AI sector is creating a new colonial world order with crowdsourcing platforms in a race to the bottom to find – and exploit – low paid workers across the globe. It was titled: How the AI industry profits from catastrophe.
It focused on the data labelling platforms like Appen, and the millions they crowdsource for this work – the so-called ‘ghost workers’. These workers label data for the tech giants via small chunks of work that earn equally modest payments. The viability of Appen and rival platforms rely on their ability to assign and pay for this work with as little human intervention as possible.
It pits Appen against the workers for a share of every dollar earned. For fiscal 2021, Appen generated revenue totalling $US447.3 million ($671.2 million). It paid out $US268.4 million for crowd labelling services but the average wage for its one million-plus workers that year was around $US268 ($391).
Appen is also up against other data labelling platforms which scour the globe for the cheapest workforce. If it is generic work that can be done anywhere then “you can definitely do a race to the bottom,” former Appen boss Braden-Harder says.
It was one reason she left Appen soon after the IPO with the rising pressure to maximise investor returns.
“I sort of knew it was going to get bad. There had been pressure already,” she says.
“I knew that with this business model, there weren’t too many options for any CEO, in terms of giving the Australian investors what they seemed to be looking for.”
The MIT series looked at how these platforms descended on Venezuela following the collapse of its economy, which has driven its middle class into poverty and driven the demand for any source of employment. Venezuela’s economic collapse produced the magic combination of a desperate, but educated workforce, and internet connectivity.
Oskarina Fuentes Anaya was one of many forced to turn to Appen as her only source of work. She fled Venezuela for Colombia. Her situation was exacerbated by a chronic illness that limited her work options, but Fuentes soon learned what it was like to have her life governed by the platform algorithms that ensured the most economic dispatch of work to Appen’s million-plus workforce.
“We all help each other out,” Fuentes told MIT of the support these workers gave each other to share what little work was available.
The MIT story chronicles the cuts to pay, the desperation to grab the dwindling work available, and account suspensions – which also triggered pay suspensions with limited recourse to a human operator from the platforms.
“What began in Venezuela set an expectation among players in the AI industry for how little they should have to pay for such services, and it created a playbook for how to meet the prices that clients have come to rely on,” says the MIT story.
While the data labelling has provided a lifeline to workers like Anaya, it has also exposed them to a Darwinian scale of exploitation as platforms lowered their pay, and suspend accounts – and livelihoods – in an ongoing race to the bottom.
The perils include harsh reviews by clients which can lead to account suspension and ambiguous tasks and administrative errors that can see an account suspended for months.
Julian Posada, an associate professor at Yale who has studied these crowdsourcing services in South America, says there is a huge power imbalance that favours the platforms which have the power to set their own rules. They can literally scour the globe for cheap labour to perform these menial tasks.
But Venezuela’s educated population, great infrastructure from before its oil economy collapsed – provided a rare combination of ingredients that made it perfect for these outsourcers, Posada says.
“So on the one hand, you have the infrastructure for work. On the other hand, you have people who are in a crisis with the highest levels of inflation, so you can pay them as low as you can,” Posada says.
At the start, it was good work.
To build up a viable network of contributors, these platforms offered bonuses, and in one case even paid these outsourced workers an hourly rate. But once they reached critical mass, a lot of these payments disappeared and pay rates dropped.
In one case, a platform Posada studied accidentally left its payments data for thousands of workers on a public Google spreadsheet.
He says it provided a clear picture of the relationship between rising crowd numbers and declining pay.
“The more people were joining, the less people were earning,” he says.
As the situation slowly improves in Venezuela, with higher oil prices, the trick will be to find the next low-cost labour market with enough people desperate for work.
“The next time there is a country in crisis, they will probably be there, as long as there are computers and desperate people,” Posada says.
After the MIT story, Appen has started highlighting its treatment of its crowdsourced workforce which includes the company’s code of ethics.
It cited an internal survey of 7,000 workers from late last year indicating that 17 per cent were long-term unemployed before joining Appen, 16 per cent were living below the global poverty line. Sixty-three per cent were using Appen earnings to support their household or pay for education.
But another figure was telling. In its annual report, Appen reported the survey showing 67 per cent identified Appen as their primary source of income.
In response to queries, Appen said: “We are committed to fair pay and ethical treatment for our Crowd. Our Crowd code of ethics explicitly states that our goal is to pay our crowd above the minimum wage in every market around the world in which we operate. To help guide our customers we have a fair pay feature available on our platform.”
Appen also adjusts its pay per task to the minimum wage of the worker’s locale. This means workers from a poor country get paid less for doing the same task as someone from a wealthier country. In the MIT story Appen said it had seen an uptick in fraud where users had used VPNs to access higher pay on offer in other countries.
Braden-Harder, for one, is not impressed with the talk of minimum wage which is set by individual states in the US and tends to be really low.
“You can pay the legal minimum wage and still be paying poverty wages,” she says.
Posada cited a recent fair work project that looked into the working conditions across all of the crowdsourcing platforms and found that none of them met the minimum standards. But Appen was the best of a bad bunch.
“It’s like, the best of the worst. They have some standards, they have some rules in place,” he says.
Braden-Harder has stepped back from executive roles and is currently an advisory board member at Santa Clara University’s Global Social Benefit Institute.
She helps mentor global start-ups like one run in Kenya by an Australian university graduate who provides school lunches.
“I think all of us, myself included, believe that businesses can do things for good, but you have to have the right business model,” she says.
When it comes to solving the crowdsourcing issue, Braden-Harder says big companies need to change their thinking when it comes to procurement of these services.
“In my experience, procurement is the evil side of any company because the same guy who’s buying toilet paper for big companies is also buying these services.”
Get news and reviews on technology, gadgets and gaming in our Technology newsletter every Friday. Sign up here.
Most Viewed in Business
Source: Thanks smh.com