MOSAAIC: A New Framework for Balancing Human-AI Control in Creative Collaboration

The rise of generative AI tools like ChatGPT, Midjourney, and Runway has transformed how humans and machines collaborate creatively. But as AI systems become more autonomous, a critical question emerges: How do we balance control between human creators and their AI counterparts? A new research paper titled "MOSAAIC: Managing Optimization towards Shared Autonomy, Authority, and Initiative in Co-creation" introduces a framework to address this challenge.

The Problem: Who’s in Charge?

Human-AI co-creativity—where both parties contribute proactively—can produce outcomes that surpass what either could achieve alone. However, as AI systems grow more capable, they risk overshadowing human agency. The paper, authored by Alayt Issak (Northeastern University), Jeba Rezwana (Towson University), and Casper Harteveld (Northeastern University), argues that control—defined as the power to initiate, direct, and decide in co-creation—must be carefully managed to maintain a productive partnership.

The MOSAAIC Framework

Through a systematic review of 172 research papers, the authors developed MOSAAIC, a framework that breaks down control into three key dimensions:

1. Autonomy – The ability to independently choose creative actions.
2. Initiative – The capacity to proactively contribute rather than just react.
3. Authority – The power to make decisions and direct the creative process.

These dimensions form a 3D control space, where human and AI contributions can be mapped along continuums ranging from full human control to full AI control, with shared control in the middle.

Strategies for Balance

MOSAAIC proposes two primary strategies to optimize control:

• AI-Controlled Adaptation – The AI adjusts its autonomy, initiative, and authority based on context (e.g., workload, task complexity).
• Human-Controlled Configuration – Users customize AI behavior to fit their preferences and expertise.

Case Studies: How Control Plays Out in Real Systems

The researchers analyzed six co-creative AI systems, including ChatGPT, LuminAI (dance improvisation), and Shimon (a robotic musician). Key findings:

• Most systems (5/6) exhibit shared autonomy, meaning both humans and AI make creative choices.
• Initiative is split—some systems (like ChatGPT) wait for human prompts, while others (like LuminAI) proactively contribute.
• Authority leans human—in 4/6 systems, users retain final decision-making power.

Why This Matters for Business

As businesses increasingly integrate AI into creative workflows—whether in marketing, design, or product development—understanding control dynamics is crucial. Unbalanced AI dominance can lead to:

• Loss of human creative ownership
• User frustration from overbearing AI
• Missed opportunities for synergistic collaboration

MOSAAIC provides a structured way to design AI systems that enhance rather than overshadow human creativity. For companies deploying AI in creative roles, this framework offers:

• Guidelines for adjustable AI behavior (e.g., letting users tweak autonomy levels).
• Benchmarks for evaluating human-AI collaboration.
• A tool for cross-domain comparisons (e.g., how control differs in writing vs. music AI).

Future Directions

The paper acknowledges limitations—such as the need for context-specific optimization rules—and suggests future work on factors like trust and ownership in co-creation. As AI continues evolving, frameworks like MOSAAIC will be essential for ensuring human-AI partnerships remain balanced, ethical, and truly collaborative.

For businesses leveraging AI in creative processes, MOSAAIC isn’t just theoretical—it’s a practical roadmap for designing AI that empowers rather than replaces human ingenuity.