Why does AI money keep flowing toward replacing workers instead of making them more capable?

This explores why AI money has tended to flow toward replacing workers rather than empowering them, and whether that tilt is inevitable. The clearest answer in the collection is about incentives, not technology. Acemoglu, Autor and Johnson argue that a single firm usually earns more by automating existing expertise than by inventing new tasks for people to do. That leaves a gap no individual company has a reason to fill. Worker-friendly AI may be better for society, but no one firm captures enough of the payoff to justify building it Why do firms build automating AI instead of pro-worker AI?. Brynjolfsson adds a feedback loop on top. Automation moves bargaining power to whoever owns the technology, and workers with less leverage can't push investment back toward augmentation. He describes this as a 'trap' that can become self-sustaining Does automating labor concentrate power more than augmenting it?.

The less obvious point is that the 'automation vs. augmentation' choice may be partly a framing artifact. Shneiderman argues that automation and human control are two separate design dials, not two ends of one slider. You can turn both up at once, as with patient-controlled pain pumps in hospitals, where the machine does the dosing and the patient keeps control Can automation and human control increase together?. Raisch and Krakowski go further. On any single task the two look like opposites, but over time they feed each other: automating one task opens room for augmenting another, and the reverse Are automation and augmentation really a simple trade-off?. If investors see the decision as an either/or, automation's easier-to-measure returns will usually win.

The pattern also isn't fixed. Anthropic's Economic Index found that as AI use deepens, especially in wealthier economies, people shift from handing off whole tasks toward collaborating with the model and learning from it Does AI adoption follow wealth and mature over time?. That fits with Agrawal, Gans and Goldfarb's argument that the bigger gains come from redesigning how an organization coordinates its decisions, not from swapping a person for a tool one task at a time Does AI adoption require rethinking how organizations coordinate?. Automation may simply be the first and easiest step, not the final destination.

There are two complications worth knowing. First, augmentation isn't automatically the 'safe' option. A study mapping more than 8,000 workplace risk scenarios found that leaning too heavily on AI helpers can slowly wear down the skills people need to supervise them Does AI augmentation protect workers from skill erosion?. Second, the stakes go beyond jobs. One line of argument holds that institutions stay aligned with human interests partly because they depend on human workers who care about outcomes. Replace that labor bit by bit and the institutions can drift in ways that are hard to undo Does incremental AI replacement erode human influence over society?. Even within AI research itself, some argue that human-AI collaboration moves faster and more safely than full automation Can human-AI research teams improve faster than autonomous AI systems?.

One gap: the collection explains the incentives well but has little long-run historical data on how investment actually split between automation and augmentation over past decades. Treat 'historically' here as an argued pattern, not a measured trend.