Chapter 9. Work, Inequality, Power, and the Socio-Economic Divide

9.1 Jobs change through tasks, not only job titles

AI affects tasks within occupations unevenly. Some routine activities may be automated; other tasks may be augmented, expanded, or made more valuable. The effect depends on technical capability, cost, reliability, regulation, customer expectations, and organisational redesign. Forecasts about the number of jobs gained or lost are uncertain and should be treated as scenarios rather than precise predictions.

Workers may experience higher productivity, new responsibilities, reduced autonomy, tighter monitoring, wage pressure, or displacement. The same technology can benefit an organisation overall while imposing concentrated costs on a particular group. A responsible transition needs consultation, retraining, fair performance measures, and a plan for people whose roles change.

9.2 The distribution of gains

AI systems require data, computing capacity, infrastructure, skilled staff, capital, and access to distribution channels. These resources may be concentrated in a relatively small number of organisations. Dependence on a limited set of providers can create vendor lock-in, pricing power, and difficulty moving data or workflows. Small organisations and lower-income communities may struggle to obtain the same benefits.

Policy and institutional responses can include digital skills development, support for small businesses, interoperable standards, public-interest research, transparent procurement, competition policy, accessibility, and investments in local-language tools. Adoption should be assessed not only by productivity but also by who benefits, who pays, and who can contest decisions.

9.3 Worker surveillance and algorithmic management

Automated scheduling, productivity scoring, recruitment filtering, call analysis, and task allocation can affect livelihood and dignity. A metric may be incomplete or poorly suited to the work. For example, speed alone can punish workers who handle complex cases or provide high-quality support. When a system evaluates people, they should know the purpose, relevant data, limitations, and appeal route.

Employers should not treat AI-generated performance assessments as self-validating. Human review should consider context, reasonable accommodation, data quality, and the possibility of systematic error. Workers should have a channel to correct records and challenge decisions without retaliation.

9.4 A fair transition agenda

• Map tasks and skills before automating whole roles.
• Consult affected workers and their representatives early.
• Provide paid training and time to learn new workflows.
• Share productivity gains where feasible through better work, development, pay, or reduced repetitive burden.
• Audit outcomes for unequal effects and monitor work intensity.
• Keep a named human accountable for decisions affecting employment and livelihood.