As AI Data Annotation Democratizes Remote Work, Freelancers Become Commodified

"I am so looking forward to the next batch!" More than a hundred similarly phrased messages lit up the Slack channel as faces from around the world chimed in with anticipation. I was in a powwow of AI annotation "experts," a collection of more than 300 freelancers, remotely hired and based, to work through a project evaluating the quality of a Large Language Model (LLM)'s different responses to given prompts. Over a little more than two days, the group of people who have (and will likely never) met in person powered through what is likely thousands of tasks.

Data annotation platforms are the shovel providers in the modern-day gold rush that is LLM development among major artificial intelligence (AI) companies. When millions of users around the world are impressed by the speed with which the likes of ChatGPT pump out eloquent answers to any question they ask, many are unaware of the human effort that goes into these chatbots' learning processes. Billionaires have been minted in companies that bring together humans who tediously label and correct machine-produced data so that LLMs can learn what is good and bad in their answers.

Today, I became one of those human laborers fiddling with the fruits of AI. In an age when outputs have increasingly become refined, annotators have to become more and more discerning. The many answers provided to the same question may sound the same, except for a missing disclaimer here or a subtle phrasing that might trigger negative emotions there. Even a thirty-minute session reviewing these similar responses can fry the mind. The wall of AI-generated text just feels all so the same in their linguistic usage and flow.

Yet, there is no denying that this tedious work could be a great economic equalizer. Remote work of the past has too often been associated with professional privilege. Those with unparalleled technical skills, like programmers, can take their laptops to the beach. Those with valuable personal connections can make phone calls from their yachts, connecting buyers and sellers to seemingly easy money from commissions. For "normal" people who are not smart or well-known enough, these are dreams to be had but cannot be fulfilled. 

Labelling AI-generated data points seems to give the average Joe and Jane a way to live the digital nomad dream, too. Yes, evaluating a response to a prompt requires good reading comprehension skills in a particular language, but nothing beyond the means of someone who attended a few years of college. With languages such as English, French, and Arabic spoken in rich and poor countries alike, the need for a massive labor force in the short term, combined with the deep pockets of major AI firms, can finally put developed-world wages in the hands of developing-country folks with internet connections.

It is no wonder that the Slack channel fills with happy messages even when the work dries up, and no one knows whether the end client will pay for the next batch. The participants are high on the reality of chasing a remote-work dream, of a corporate democracy in which a farmer's son in Nigeria and someone in Silicon Valley can truly be treated as equals, even if equality only exists in the confines of a Slack channel. Labelling data might be tedious, but its operators participate in the bleeding edge of modern technology, despite having little technical education themselves.

But as the dream expands with the vision of borderless, tech-based meritocracy, can the dream chasers stop even briefly to consider the consequences? If annotating data is so easy, and the work is accessible to anyone anywhere with internet, why not hire the cheapest laborers out there? As "freelancing" becomes normalized, and people become enamored with the idea of productive work as clicking a few buttons while sitting at home, who'd still want to break a sweat commuting to the office and picking up a hoe on the farm? 

The desirability of remote work, then, speeds up the commodification of labor. In the Slack channel of 300+ people, everyone is more or less nameless and replaceable. If one person is kicked out or separated, no one will know or care. But everyone knows that the channel is part of their dream of working from home forever. As AI becomes better and more people come to know about these opportunities, those in the channel will need to fight harder to stay in themselves and keep others out. When that day comes, looking forward to the next batch will sound more desperate than passionate.