I refused to train the AI that could replace me
Refusing to help train AI systems that may automate one’s own job raises deeper questions about who controls technological progress and who benefits from productivity gains. Commenters debate whether society can or should slow AI development, given billionaire and state incentives, and whether past technologies have ever truly been stopped once profitable. Many fear that without new economic and social models, AI-driven automation will concentrate wealth, erode worker bargaining power, and leave displaced people without meaningful support or alternatives.
Refusing to Train AI & Individual Agency
- Some praise the choice not to train an AI “replacement” as a rare act of resistance against capital-driven automation.
- Others argue it’s largely symbolic: if one person refuses, others will do it; impact on outcomes is minimal.
- A few note the author’s decision wasn’t clearly principled but see value in the reflection and the resulting essay.
Societal Control vs. Inevitability of AI
- One camp argues AI will be trained regardless; individual or even societal resistance is ineffective.
- Another insists “society” can still impose brakes via regulation, protest, and political action, citing past tech bans (e.g., CFCs, specific pollutants).
- Counterpoint: states struggle to regulate even “dirt-poor pirates,” so controlling billionaire-backed AI efforts is seen as unrealistic.
- Debate over whether slowing AI in EU/US just hands advantage to less-constrained countries; some claim no real slowdown is happening.
Labor, Replacement, and Inequality
- Many note AI is just the latest in a long line of “train your replacement” episodes (younger workers, offshoring, now machines).
- Key difference: AI training scales infinitely; one worker’s knowledge can replace many workers indefinitely.
- Strong concern that automation breaks capital’s dependence on labor, enabling a drift toward a new feudalism where only a small elite is needed.
- Question raised: once labor isn’t needed, what incentive does “the system” have to provide food, housing, and healthcare?
Productivity, Past Tech, and Distribution of Gains
- Comparisons to cars, washing machines, computers, and self-checkout: tech often reduces drudgery but doesn’t reliably reduce working hours.
- Some argue productivity has risen while labor’s share of gains and overall economic security have stagnated or worsened.
- Others counter that labor’s GDP share and unemployment remain within historical norms and that AI gains may flow mostly to consumers.
Governance, Rights, and Models
- EU rules against fully automated decision-making in hiring are highlighted as a concrete constraint.
- Some insist AI skills should go into open models, not proprietary systems.
- Exploitation of low-wage workers for data labeling and the “long tail” hunt for niche data are seen as structural and ongoing.