TLDR: A former gig worker, Johanna Knox, shares her 18-month experience training generative AI for major tech companies, revealing a landscape of precarious employment, declining wages, and constant pressure. She argues that the real concern isn’t AI’s inherent danger, but the exploitative practices and immense power wielded by the billionaires who control its development and deployment, contributing to a widening wealth gap and misdirection about AI’s true societal impact.
Johanna Knox, a guest writer for The Spinoff, has provided a candid first-person account of her 18 months as a gig worker, or ‘tasker,’ involved in training generative AI for some of the world’s largest tech companies. Her journey, which concluded in September of this year, sheds light on the often-unseen human labor underpinning advanced AI systems and raises critical questions about corporate ethics and the future of work.
Knox initially embarked on this role in early 2024, driven by a ‘know-your-enemy impulse’ amidst her growing apprehension and fascination with generative AI. She cited concerns over its ‘queasy-making attempts at art,’ the ‘hollow positivity of chatbots,’ and its ‘hungry disregard for copyright, privacy and data sovereignty.’ The prevailing media narratives of AI ‘coming for our jobs’ and ‘doom-filled prophecies of super-intelligent AI escaping human control’ also fueled her decision to ‘poke the beast and see what happens.’
Her recruitment began with a LinkedIn message promising ‘well-paid flexible work’ for postgraduates to train ‘leading-edge AI models.’ After a series of onboarding processes, including ID verification, videos, Slack channels, and extensive Google Docs detailing how to rate AI models on criteria like accuracy, instruction following, writing style, and safety, Knox completed an online assessment. For this initial seven to eight hours of onboarding, she was compensated US $350 (NZ $550–$600), a rate comparable to an average freelance editor.
Knox’s tasks primarily involved sourcing public-domain images from the internet and crafting prompts that required chatbots to interpret and reason about them. Examples included asking for statistical comparisons from timelines or graphs, requesting step-by-step solutions to puzzles, or identifying relationships within royal family trees. Following the prompt, she would then write an ‘ideal chatbot response.’ These tasks, either single-turn or multi-turn, were rigorously reviewed and audited before being used by anonymous clients to train their AI models.
However, the working conditions quickly deteriorated. Taskers faced constant pressure from strict timers, with the risk of losing hours of unpaid work due to platform glitches if they timed out before submission. Simultaneously, rushing tasks led to ‘low quality’ scores and threats of removal from projects. Knox herself was informed that despite high quality, her speed was ‘looked on unfavourably,’ without clear guidance on expected metrics.
Compounding the stress, hourly rates plummeted from US $40 to $35. The time allotted for tasks at the full rate was shortened, with any additional time compensated at ‘barely minimum wage.’ Unpaid training for new projects became increasingly common, sometimes consuming hours of work. Instructions were subject to ‘constant change – sometimes drastically, always without warning,’ reflecting a ‘vertical hierarchy, with blindsides all the way down.’
The author highlighted the human cost through the experience of ‘Dana’s squad,’ a team led by an empathetic supervisor who was often as ‘in the dark’ as her subordinates. Dana was later ‘furloughed’ without warning, a common occurrence where workers were let go abruptly despite promises of project longevity. Knox’s own pay steadily decreased in $5 increments, eventually reaching $25 an hour for a set period, followed by approximately $14 an hour before timing out. She noted that company expectations often forced workers to continue at lower rates, perform unpaid preparation, or submit low-quality work.
Knox eventually reached a breaking point during another unpaid training session, finding herself unable to even read the instructions. She ceased working, acknowledging this was a ‘luxury not everyone has.’ While she still feels the loss of income, she expresses relief from the stress and the knowledge of ‘directly contributing to the way the biggest AI companies are causing harm.’
Her overarching conclusion from the 18-month ordeal is stark: ‘it isn’t AI we should fear, but the handful of billionaires who control it and foist it on us at every turn.’ She asserts that the tech giants’ treatment of workers ‘shows the opposite’ of having humanity’s best interests at heart. Knox echoes author Karen Hao, stating that ‘the big AI companies are becoming the new empires,’ deliberately portraying AI as an ‘unstoppable force’ to be ’embraced’ to avoid being ‘left behind.’ She views AI as a ‘wage depression tool,’ exacerbating the wealth gap, while tech leaders ‘blithely predict swathes of job losses’ and an ‘idyllic picture of a future world without work.’
Also Read:
- Experienced Professionals Lead AI Adoption, Challenging Generational Assumptions in the Workforce
- Navigating the AI Landscape: Why Businesses Must Diversify Beyond Single Models and Avoid Over-Reliance
Knox concludes with a call to action, noting that people are ‘fighting back’ through documenting worker stories, resisting data centers, and advocating for humane AI policy. She stresses that meaningful change against the ‘enormous wealth, political power and selfishness in Silicon Valley’ will only occur if people believe it is possible.
