Two Visions of AI

In 1987, Thomas Sowell observed that most political disagreements are not about the specific issues they appear to be about. They are about something deeper and prior: the assumptions participants carry into the argument before the argument begins. The constrained vision assumes human nature is permanently flawed and limited, requiring institutions to channel its defects toward tolerable outcomes; the unconstrained vision assumes human potential is malleable and improvable, and that the limits of the present are obstacles to be overcome rather than realities to be managed. Nearly four decades after Sowell mapped these visions, the AI discourse reproduces them with uncanny precision. The optimist who points to the twenty-fold productivity multiplier in Trivandrum and the skeptic who points to the dissolution of expertise both have evidence. Neither can persuade the other. They are not arguing about AI. They are arguing about whether the cost of