Proposed Protocol for AI Agent-Human Collaboration

As workflows in which AI agents perform substantive decision-making increase, the Human Interaction Protocol has been proposed to clarify accountability when errors occur and to improve collaboration efficiency.

This protocol standardizes how agents make requests and how humans respond, addressing the following key challenges:

• Accountability and Audit: Clarifying approval and ownership of outcomes, and designing interactions to be traceable
• Clear Communication: Defining the nature of requests—simple confirmation, approval, decision, etc.—to clarify the definition of "done"
• Human-Centered Design: Designing interactions that account for human factors such as user fatigue, security, and usability

Key Components:

• HumanInteraction: A structured request from the agent, including context and constraints
• HumanFeedback: A human response with audit trail capability
• Expectation vs Commitment: The distinction between a simple request and a binding promise with deadlines/outcomes
• Resolution Policy: Defines how decisions are made among multiple participants (majority vote, first response, etc.)

The protocol leverages JSON Schema to ensure LLM generation efficiency and cross-tool interoperability, and is currently in a draft stage that is being continuously updated.