Does AI's payoff depend on the model itself, or on who gets access, who owns it, and what rules apply?

This explores whether AI's benefits flow to particular people because of the technology itself or because of choices made by institutions, firms and governments. The corpus mostly says it's the choices. An interdisciplinary review covering information, work, education and healthcare found that generative AI can both widen and narrow inequality. Which way it goes depends on who gets access, how the tools are built into existing systems, and what incentives surround them, not on what the model can do Does generative AI inevitably worsen or reduce inequality?. The same capability can help a struggling student in one setting and pull further ahead of everyone else in another.

The economic notes show where those choices actually bite. Anthropic's scenario modeling finds that as AI speeds up growth, a larger share of income goes to the owners of capital and a smaller share to workers. Average wages can still rise while knowledge workers' wages stall or fall. The scenarios point to two things that stop the bigger economy from being shared: concentrated ownership, and the difficulty workers have moving between occupations Does AI growth inevitably shift wealth away from workers?. Both are institutional facts, not technical ones. Inside firms, the pattern repeats. Companies more exposed to AI replace online freelance workers faster and more cheaply than less-exposed companies, which suggests that building AI capability in-house pays off more the more you have of it Do firms substitute labor for AI at different rates?. If early adopters keep pulling ahead, a firm's starting position shapes the outcome as much as any policy does.

A less obvious question is what counts as a fair claim on AI's gains at all. One argument holds that generative models are crystallized collective knowledge, a synthesis of humanity's shared digital output. On that view, you can't sensibly trace any output back to an individual author. Restricting access, even for well-meant copyright reasons, could then turn a shared inheritance into a private asset Should restricting AI access create new kinds of inequality?. So the policy lever cuts both ways: rules meant to protect creators could themselves concentrate gains.

Here is the twist you might not expect. Institutions may not keep their power to steer this forever. The gradual-disempowerment argument says that societies stay aligned with what people want partly because they depend on human workers who care about outcomes. As AI replaces that labor, institutions lose one of the main ways ordinary people's preferences reach them, and the drift may become hard to reverse Does incremental AI replacement erode human influence over society?. Whether institutions determine distribution may depend on acting before AI wears away the leverage that made them responsive.

A note on what's missing: the governance material in the collection is mostly about safety, not sharing. One example is the argument that companies can't police AI risk without binding government oversight Can companies alone manage the risks of AI systems?. The corpus has little on specific redistribution tools like taxation, public ownership or wage policy. It's also worth knowing that the gains being fought over may arrive more slowly than benchmarks suggest. Agents that win contests still struggle with real, long-running professional work, which shows that what gets measured shapes what gets built Why do agent benchmarks not predict real economic value?.