What question did this study set out to answer?

The aim is to explore how managing ambiguity in workflows affects collaboration between humans and large language models (LLMs).

May 16, 2026Open Access

Controlled Ambiguity Curve

Key Points

The aim is to explore how managing ambiguity in workflows affects collaboration between humans and large language models (LLMs).
Introduces the Controlled Ambiguity Curve as a model for interaction-level analysis.
Defines types of ambiguity: operational, interpretive, generative, and expressive.
Analyzes conversation trajectory instead of isolated prompts.
Quality of outputs is linked to modulating ambiguity over time across various workflow phases.
Initial grounding requires reduced ambiguity, while generative phases utilize expanded ambiguity.
Final production involves reducing ambiguity again, demonstrating the importance of context in human-LLM interactions.

Abstract

This working paper introduces the Controlled Ambiguity Curve as an interaction-level model for Human–LLM collaboration. The central claim is that, in complex workflows, quality does not depend on minimizing ambiguity, but on modulating it across time. Ambiguity should be reduced during initial grounding, selectively expanded during reflective and generative work, and reduced again during final production. The paper distinguishes operational, interpretive, generative, and expressive ambiguity, and argues that the human collaborator acts as an ambiguity modulator rather than merely a prompt writer. The CAC shifts the unit of analysis from the isolated prompt to the conversation trajectory, offering a vocabulary for understanding how advanced users conduct LLM-mediated work through phases of constraint, openness, and closure.

Controlled Ambiguity Curve

Key Points

Abstract

Cite This Study