Human-AI Configuration

Definition

The specific arrangement of roles, responsibilities, and interactions between humans and AI systems in a given workflow. Not just whether humans are “in the loop,” but how the entire system is structured.

Configuration Dimensions

Task Allocation: What does the human do vs. the AI? Authority: Who has final decision rights? Initiative: Who initiates actions? Communication: How do human and AI exchange information? Feedback: How does the system learn from interactions?

Configuration Types

AI-as-Tool: Human initiates, AI responds, human evaluates AI-as-Assistant: AI suggests, human decides AI-as-Peer: Collaborative problem-solving AI-as-Supervisor: AI monitors and corrects human work AI-as-Autonomous: AI acts independently within bounds

Why Configuration Matters

Same AI capability can be deployed in radically different configurations. A medical diagnostic AI could be:

  • A second opinion for doctors
  • A screening tool for nurses
  • An autonomous triage system

Each configuration has different risk profiles, trust requirements, and value propositions.

Design Implications

Configuration should be intentional, not emergent. Match configuration to task characteristics, error costs, and human expertise available.

Related: 01-atom—human-in-the-loop, 07-molecule—hybrid-human-ai-workflows, 01-molecule—appropriate-reliance-framework