Human Learning by Model Feedback: The Dynamics of Iterative Prompting with Midjourney
It addresses the problem of potential bias in user data for AI training, as prompts may align with model preferences rather than human intentions, which is incremental as it builds on existing cognitive and alignment research.
This paper analyzes how human users iteratively update prompts for the Midjourney text-to-image model based on feedback, revealing that prompts predictably converge toward specific traits, with initial evidence suggesting this is due to both users realizing missed details and adapting to the model's preferences.
Generating images with a Text-to-Image model often requires multiple trials, where human users iteratively update their prompt based on feedback, namely the output image. Taking inspiration from cognitive work on reference games and dialogue alignment, this paper analyzes the dynamics of the user prompts along such iterations. We compile a dataset of iterative interactions of human users with Midjourney. Our analysis then reveals that prompts predictably converge toward specific traits along these iterations. We further study whether this convergence is due to human users, realizing they missed important details, or due to adaptation to the model's ``preferences'', producing better images for a specific language style. We show initial evidence that both possibilities are at play. The possibility that users adapt to the model's preference raises concerns about reusing user data for further training. The prompts may be biased towards the preferences of a specific model, rather than align with human intentions and natural manner of expression.