Position: AI Systems Should Be Designed to Develop Their Users, Not Just Assist Them

Abstract

Human-facing AI systems are mainly designed and evaluated for what they do for users, and seldom for what sustained use does to users' own capabilities. Evidence that AI use can erode, preserve, or improve human capability is accumulating across several fields, but it remains scattered and has not yet shaped how AI systems are designed or evaluated. This position paper argues that endogenous human development, meaning lasting growth in capability that resides in the person rather than in the tool, should be a primary design objective for AI systems, measured by what users can do without the AI, alongside assisted performance, in the capabilities they want or need to have. Systems built on the same model can differ in this effect, and only measurement without the AI reveals the difference. Transparency, explainability, safety, accountability, and contestability should also be designed and evaluated as enabling infrastructure for this objective, not only as ends in themselves. We outline a staged research framework and call on developers to report unassisted capability, on benchmark builders to measure effects on users, and on researchers to test developmental designs in staged trials.