CONCEPT
The Autonomy Slider
Andrej Karpathy’s instrument for thinking about human-AI collaboration: a continuum from full human control to full machine autonomy, calibrated not by ideology but by the cost of verifying that the machine did the right thing in each specific domain—a practical answer to the binary that dominates public debate.
The public debate about AI autonomy tends toward binaries: either humans control the machine or the machine controls itself, either we trust AI or we do not, either agents are safe or they are dangerous.
Andrej Karpathy’s autonomy slider replaces the binary with a continuum, and replaces ideology with a concrete calibration principle. Rather than a fixed level of human oversight for all tasks, he proposes a slider the human moves depending on the task, the stakes, and the trustworthiness of the system in the specific domain at hand. At one end, the AI suggests and the human approves each step. In the middle, the AI executes larger chunks while the human supervises and verifies. At the far end, the AI runs autonomously and the human checks only the final result. The key insight is that the correct position on the slider is not a matter