AI - Part 2: The Bad — Workers in the Age of AI
This is the second part of a 4 part post, the first post is here.
Part one described what AI could do for working people under the right conditions, this section describes what it is actually doing under current conditions. The evidence is mounting that AI, as presently deployed, is intensifying inequality, eroding worker autonomy, and creating new forms of exploitation that are less visible and less accountable than the old ones.
Job Displacement: The Scale and Shape of Losses
The labour market effects of generative AI are no longer speculative. A World Bank study analysing 285 million U.S. job postings found that occupations with above-median AI substitution scores experienced a 12% decline in postings relative to below-median occupations following ChatGPT's release. The effect accelerated over time, growing from 6% in the first year to 18% by the third year. Entry-level positions requiring neither advanced degrees nor extensive experience were particularly hard-hit, with losses of 18–20%. Administrative support roles fell by 40% and professional services by 30%1.
These are not abstract percentages. They represent people—disproportionately young workers, career-changers, and those without the credentials to compete for the shrinking pool of positions that AI cannot yet touch. A separate Harvard Business School working paper confirmed a heterogeneous pattern: AI-driven automation reduces labour demand and skill requirements in structured cognitive-task jobs, while increasing demand only in positions specifically built around human-AI collaboration. The jobs being lost and the jobs being created are not the same jobs, and they are not held by the same people2.
The mechanisms of displacement are often obscured by managerial language. A 2025 Gartner survey found that 80% of companies piloting AI reported workforce reductions, with the cuts occurring regardless of whether the technology was generating returns. AI is being used as a justification for downsizing even when the economic case for it is absent. This is not technological inevitability; it is a choice, driven by short-term cost reduction rather than long-term value creation.
The Ghost Workforce Behind AI
Every AI system that appears autonomous rests on a foundation of human labour that is systematically hidden from public view. Millions of workers worldwide annotate data, moderate content, and correct model outputs so that machine learning systems can function. Researchers call this "ghost work"—labour that is essential to AI's operation yet deliberately obscured by technology companies invested in the narrative that their systems are self-learning.
The conditions of this work are stark. Data workers typically earn per microtask—an annotation, a check, a translation—and must spend unpaid time searching for tasks across platforms. They often earn below minimum wage, even in countries with relatively low costs of living. Those employed through business process outsourcing firms in physical offices are paid by the hour but still frequently fall below the poverty line. Many ghost workers are highly educated people with distance from the formal labour market, their qualifications rendered irrelevant by the structures of platform-mediated work3.
The geographic dimension is significant. Workers in Colombia, Kenya, and the Philippines have reported extreme occupational, psychological, sexual, and economic harms. In Kenya, data workers have earned as little as $1.32 per hour labelling toxic content for large language models. The work is organised through global supply chains that allow technology companies to outsource risk while maintaining plausible deniability about conditions. Workers are often bound by non-disclosure agreements that prevent them from speaking publicly about their experiences.
Algorithmic Management: The Boss That Never Sleeps
For workers in the gig economy, the most immediate impact of AI is not job loss but algorithmic management: the use of automated systems to allocate tasks, measure performance, and impose discipline. A cross-sectional study of 1,204 gig workers across four countries found that algorithmic surveillance significantly undermined worker autonomy, fairness, and psychological wellbeing. Workers reported being unable to contact a human being during disputes or when their accounts were deactivated—a digital dismissal with no right of appeal.
The psychological spillover extends beyond the workplace. Recent research has identified intimate partner surveillance as a control-compensation mechanism triggered by algorithmic management: workers subjected to constant monitoring at work begin to replicate those behaviours in their personal relationships. This is a disturbing finding, suggesting that the psychological effects of algorithmic control cannot be contained within working hours.
Gig workers themselves have described the experience as one of relentless visibility, where every action is tracked, measured, and fed back into a system that can reduce pay or terminate access without explanation. The algorithmic manager is not merely more efficient than a human supervisor; it is less accountable, less transparent, and less capable of exercising the discretion that even the most rigid human manager retains.
The Degradation of Creative Work
Generative AI is reshaping creative labour markets in ways that devalue professional skill. A scoping review of creative professionals across visual art, writing, performing arts, and design found that AI tools risk "devaluing traditional notions of mastery, skill, and authenticity; thus, reducing the economic and cultural worth of professional creatives and their creations". Empirical evidence supports this concern: prices for AI-related gigs on a major freelancing platform dropped by approximately 33% following a ruling that AI-generated work lacked copyright protection, suggesting the economic floor is falling out.
A separate analysis in Minds and Machines argued that artists "will see their work monetarily devalued because the imitations generated by GenAI will be useful to the businesses that would be their clients". The technology does not need to match the quality of human creative work to undermine the market for it; it only needs to be good enough and cheap enough for commercial buyers who prioritise cost over craft.
The Common Structure: Externalising Costs, Concentrating Gains
Across job displacement, ghost work, algorithmic management, and creative devaluation, a structural pattern is visible. AI enables organisations to externalise the costs of production—psychological harm, economic precarity, environmental damage—while concentrating the productivity gains among shareholders and executives. The worker bears the risk; the firm captures the reward. This is not a malfunction of AI deployment under current conditions; it is the logical outcome of deploying any powerful technology within an economic framework that treats labour as a cost to be minimised rather than a source of value to be invested in.
Next part is here.
