AI;DR

AI systems are deployed first and most aggressively in sectors where labour is cheapest and workers have the least power to refuse: content moderation, data labelling, logistics, customer service. The productivity gains accrue to shareholders. The displacement accrues to workers who are rarely consulted and rarely compensated.

This is not a new pattern. It is the latest iteration of a long history of automation that concentrates the benefits of technological change at the top of the income distribution while distributing the costs across those least able to absorb them.

Democratic deliberation requires a shared informational environment — not agreement, but a common set of facts that participants can dispute. Recommendation algorithms optimised for engagement fragment this environment by design, delivering personalised information diets that maximise emotional response rather than shared understanding.

Generative AI adds a further dimension: a technology that can produce convincing false content at scale, at negligible cost, in any language. The infrastructure for democratic deliberation was not designed for this environment.

AI systems are being deployed by public administrations to make or support decisions about benefits eligibility, housing allocation, child welfare assessments, and parole recommendations. These systems carry the authority of the state while obscuring the chain of human judgment that produced them.

When a system fails the people who depend on public services, accountability becomes a question of who designed the algorithm, who procured it, who validated it, and who had the power to override it. In most cases, none of these chains are visible to the affected person.

The widespread adoption of AI-assisted writing tools in education raises a question that is not primarily about cheating: it is about what education is for. If the goal is the production of text that meets specified criteria, AI tools are efficient. If the goal is the development of the capacity to think, argue, and write independently, the same tools may be directly corrosive.

This is not an argument against AI in education. It is an argument for being precise about what is being optimised, and for whom.