Some AI tools make workers more valuable; others replace them outright — what actually decides which happens?

This explores whether some kinds of AI reliably help workers while others push them out, and what decides which side a tool lands on. The corpus offers one useful way to sort them, but its stronger finding is that the same tool can help or hurt depending on how it is deployed, who uses it, and what the worker is trying to do. The sorting comes from Acemoglu, Autor and Johnson, who divide AI into types by their effect on labor. Their sharpest point is economic rather than technical: firms make more money automating expertise than building tools that create new tasks for people. So firms underinvest in pro-worker AI, even though it would leave society better off overall Why do firms build automating AI instead of pro-worker AI?. In this view the split between helping and replacing workers is a matter of incentives, not of which technology gets invented.

The clearest evidence that AI can help workers comes from a 758-consultant experiment with GPT-4. The bottom-half performers gained the most, which narrowed the skill gap. The catch is that AI use was mandatory. When people choose for themselves whether to use AI, the effect may reverse, because stronger performers tend to adopt more Does AI help lower-skilled workers catch up to top performers?. There is a second limit. Productivity gains show up when workers apply skills they already have, and they disappear when workers use AI to learn something new When does AI actually boost worker productivity?. Wu et al. found that people did better work with AI but showed no improvement when they later did similar tasks alone Does AI assistance help workers learn lasting skills?. So 'helping' may mean helping you perform, not helping you grow.

That changes how safe 'augmentation' looks. It is often treated as the worker-friendly alternative to automation, but a map of 8,356 workplace AI risk scenarios found that leaning on AI agents can slowly wear down workers' skills and their ability to catch the AI's mistakes Does AI augmentation protect workers from skill erosion?. Scaled up to whole societies, the 'gradual disempowerment' argument holds that institutions stay aligned with human interests partly because they depend on human workers who care about outcomes. Remove that dependence piece by piece and the systems can drift away from what people want Does incremental AI replacement erode human influence over society?.

On displacement, the data points to targeted effects rather than mass job loss. ADP payroll records show no economy-wide job losses from AI, but young workers in AI-exposed jobs are hired at a 19% lower rate, while experienced workers show no such gap Is generative AI displacing workers at economy-wide scale?. The damage shows up as entry-level jobs that never open, not as layoffs. How the exposure is spread matters too. When AI touches only a few of a job's tasks, workers can shift to the rest and net employment barely moves. When exposure is spread across many tasks, labor demand drops Does concentrated AI exposure enable workers to adapt and reallocate?. Firms also differ: the more AI-exposed ones replace online freelance workers with AI faster and more cheaply, which suggests that in-house AI capability compounds Do firms substitute labor for AI at different rates?. Real delegation of tasks to AI clusters in information-heavy jobs and follows what the technology can do, which is a different pattern from the old 'routine jobs go first' prediction Where have workers actually delegated tasks to AI?.

The distribution question has a twist. Acemoglu's model predicts AI's wage effects will be spread more evenly across demographic groups than past automation was, yet income from capital still pulls ahead of income from labor Will AI increase inequality less than past automation?. And 'evenly' hides a gender split. In male-dominated jobs, exposure falls mostly on high-paid, high-skilled workers. In female-dominated jobs, it spreads across all skill levels, so lower-paid women with fewer resources to adapt are disproportionately exposed Does AI exposure hit low-wage workers harder in some fields?. The takeaway: whether AI is 'pro-worker' depends less on the tool than on whether its use is mandatory or optional, whether it touches a few tasks or many, whether you are already established or just starting out, and whether the goal is getting today's task done or building skill for tomorrow.