🔗 What we let machines do

The article presents an innovative framework for evaluating the autonomy of AI systems, dissecting the nuanced boundaries we establish for their capabilities, from basic analysis to self-driving cars. It introduces the concept of an Autonomy Ladder which categorizes AI's functionality into four rungs: analysis, prediction, generation, and behavior. As we progress from spreadsheets to autonomous vehicles, the piece underscores a troubling trend: the gap between what AI can technically do and what society is willing to accept. While some AI applications face minimal pushback, like predictive weather models, others, particularly in the realm of generative outputs or embodied behaviors, create significant contention due to issues of authorship, accountability, and irreversible mistakes. The article proposes a hypervisor-like model for managing these systems, advocating for tailored oversight as AI continues its march into areas ripe with risk and ambiguity.