Human Agency Scale

A spectrum describing the degree of human control and decision-making authority in human-AI systems, from full human control to full AI autonomy.

The Scale

1 - Full Human Control: AI provides information, human decides and acts 2 - Human Decides, AI Executes: Human chooses action, AI implements 3 - AI Recommends, Human Approves: AI suggests, human reviews and authorizes 4 - AI Acts, Human Monitors: AI executes within bounds, human oversees 5 - Full AI Autonomy: AI decides and acts without human involvement

Design Considerations

Moving down the scale (toward autonomy):

• Requires: Higher AI reliability, better error detection
• Enables: Scale, speed, consistency
• Risks: Error propagation, accountability gaps

Moving up the scale (toward control):

• Requires: Human availability, expertise
• Enables: Nuance, values alignment, accountability
• Risks: Bottlenecks, fatigue, inconsistency

Dynamic Positioning

Agency level should vary by:

• Stakes of the decision
• Confidence of AI system
• Available human expertise
• Time constraints

Related: 01-atom—human-in-the-loop, 01-molecule—human-ai-configuration-concept