Google Research proposes Agile Deliberation for subjective classification

On November 5, 2026, Otilia Stretcu, Research Scientist, and Leijie Wang, Student Researcher at Google Research, introduced Agile Deliberation, a method aimed at building AI image classifiers for subjective categories. The approach guides users through an interactive deliberation workflow that runs from scoping sub-categories to resolving edge cases, and relies on an LLM-driven optimization loop to automatically build an accurate, aligned classifier.

The authors note that machine learning models learn to recognize the world around us from examples labeled by humans. Underlying that process is an assumption that the person doing the labeling has a clear and stable definition of the category in mind before they begin. That assumption holds well for objective categories such as a stop sign or a golden retriever.

For subjective categories, however, labelers' definitions tend to be unstable, which creates training difficulties. Agile Deliberation targets categories such as "healthy food" and "inspirational art," where the boundary of the category is itself open to judgment. In the material available, the account currently stops at the proposal and its rationale, with no further details on evaluation or deployment.