The Art of Control: Balancing Autonomy, Authority, and Initiative in Human-AI Co-Creation

In the expanding domain of artificial intelligence, creativity is no longer a human-only endeavor. From music composition to visual art and storytelling, AI agents are taking on increasingly creative roles. But as these systems become more proactive, one question looms large: who’s really in control?

Enter MOSAAIC — a framework developed to guide the design of co-creative systems by managing autonomy, initiative, and authority in shared human-AI decision-making.

The Three Pillars: Autonomy, Initiative, and Authority

The authors define three interrelated yet distinct aspects of control:

• Autonomy: the degree to which an agent (human or AI) can operate independently without direct supervision
• Initiative: the agent’s ability to propose new ideas or actions without external prompts
• Authority: the power to finalize decisions or direct the creative path

These variables form a multidimensional space that helps analyze how control is distributed in any human-AI collaboration.

For example, ChatGPT often exhibits high autonomy and initiative (e.g., when completing prompts), but retains low authority — the human can override or ignore suggestions anytime. In contrast, systems like Shimon, a robotic jazz musician, may have shared authority with humans in live performance, dynamically responding with creative inputs that alter the musical direction.

Two Design Strategies: Adaptation vs Configuration

The framework describes two major strategies for managing the balance of control:

1. AI-Controlled Adaptation: the system dynamically adjusts how much initiative and autonomy it takes. For example, LuminAI — a dance improvisation partner — modulates its contributions based on the human dancer’s movements and responsiveness, shifting between leader and follower roles¹.

2. Human-Controlled Configuration: the user manually sets boundaries, priorities, or goals for the AI. For instance, in Reframer, a collaborative drawing tool, users choose when and how the AI suggests content, effectively constraining its initiative¹.

A strong co-creative system may use both: AI adapts to context within constraints defined by the user.

Case Comparison: How Six Systems Distribute Control

The MOSAAIC paper evaluates six co-creative systems by mapping their control profiles across the three dimensions. Here’s a comparison summary:

This table makes clear that systems can combine different configurations of control. For example, Reframer excels in letting users steer the creative flow, while SnakeStory leans more toward an autonomous narrative generator that adapts to feedback.

Implications for Cognaptus and XAgent

At Cognaptus, we’re already exploring multi-agent architectures (like XAgent) where agents must negotiate control over workflows — e.g., when to take initiative in drafting financial insights or visualizations, versus when to wait for human cues.

MOSAAIC’s framework offers a valuable design lens for structuring these interactions:

• Should a “Chart Agent” propose visualizations on its own, or only upon request?
• Can a “Summary Agent” override prior content to improve clarity, or must it defer?
• How does the system escalate when agents disagree (e.g., between accuracy and creativity)?

By explicitly modeling autonomy, initiative, and authority in our agents — and enabling configuration or adaptation — we move closer to building AI collaborators that are trusted, transparent, and effective.

Final Thoughts

In the era of co-creation, AI is no longer just a tool — it’s a partner. But successful partnerships depend on the balance of control. MOSAAIC provides both a conceptual map and practical strategies for achieving that balance, helping designers create systems that amplify rather than diminish human creativity.