SOURCE READING
Original and translation
The excerpt and translation use the primary source below. Further references inform SHIO’s analysis.
Primary source
- Original title
- Agile Deliberation: How AI helps humans define the gray areas
- Original author / institution
- Otilia Stretcu and Leijie Wang / Google Research
- Primary source
- Google Research
- Source language
- English
- Original publication date
The five-word excerpt below names two stages in Google Research’s November 4, 2026 blog post. The post explains earlier research; the authors’ paper was first submitted on December 11, 2025. The branding scenario and work suggestions that follow are SHIO’s independent analysis.
Basis for use：Short attributed excerpt; full-blog translation rights not established
This edition uses the source language, so the original appears once.
concept scoping and concept iterationSHIO analysis
Make visual judgment something a team can discuss
The following is SHIO’s independent judgment and discussion of methods. It is separate from the source text and its translation.
SHIO reading note
Google Research’s November 4, 2026 post presents Agile Deliberation, an approach to clarifying subjective image categories through borderline examples. For a branding team, it raises a practical question: what should an image help this particular piece of communication express?
The central idea
SHIO would use AI to organize visual decisions while keeping the image, its purpose and the reason for choosing it connected. A shared adjective gives the discussion a starting point.
Put an abstract requirement beside actual images
The workflow described in the post first scopes what a concept includes and excludes, then presents ambiguous images for discussion. Human feedback informs revisions to the definition and classifier. Its subject is visual categories whose meaning can shift with context and understanding; a fixed label cannot capture every requirement.
Imagine a brand team seeking photographs of everyday life in a particular place and asking for a local feel. SHIO would make the intended meaning more concrete: the relationship between people and a setting, an activity still taking place, or how a space is used. Whether an old building belongs in the selection depends on the purpose of this communication. The subject and its historical appearance need an explainable connection to what the brand is saying. This is a hypothetical work discussion, not a completed client project.
Use difficult examples to explain a disagreement
For methodological context, version 2 of the authors’ paper describes nine participants without professional content-moderation experience completing 18 interactive sessions across two subjective concepts. The paper was first submitted in December 2025 and revised in April 2026. This scope does not establish how much selection time a branding project would save or whether different teams would reach the same judgments.
In SHIO’s proposed selection discussion, clearly suitable, unsuitable and uncertain images can sit together, each with an explanation. If two owners disagree, ask what each is attending to: a clear subject, a connection to the text, or missing context. AI can organize those explanations and retrieve similar candidates. The person who understands the communication task should explain the final choice. Source, shooting date and usage rights need separate attention; visual suitability does not establish those conditions.
Leave the reasons for the next person doing the work
SHIO suggests keeping a short note with a selected group of images: the text each supports, the reason to retain or replace it, and the context still needed. The note should let an editor, designer or photographer taking over understand the same task. A conclusion such as premium, authentic or modern leaves the next person guessing. Connecting a reason to a particular image also gives the team a way to notice when one word is carrying different meanings.
If the purpose changes from introducing a product to explaining how it is made, revisit both the images and the criteria. Keep the earlier selection and its context, then record what has changed. As AI retrieves more candidates, clear purposes, versions and ownership become useful. This is SHIO’s proposal for branding work, not a validated method for reducing cost or producing agreement. Its usefulness needs to be observed through discussion, selection and subsequent use on the actual project.
A question for practice
When AI helps find, classify and compare images, keep their purpose, selection reasons and applicable conditions alongside them. A brand’s visual quality needs concrete judgments behind it.
REFERENCES & EDITORIAL NOTE
Sources & further reading
Source excerpts, translations, and SHIO’s independent analysis are presented separately. The resources below offer additional context and further reading.
Google Research · Primary source
Agile Deliberation: How AI helps humans define the gray areas
Original source date：Accessed：A five-word attributed excerpt accompanies separate SHIO analysis. No open license for full reproduction or translation of the blog post was identified. No images, charts or full text are reproduced; no Google endorsement is implied.
Leijie Wang, Otilia Stretcu, Wei Qiao, Thomas Denby, Krishnamurthy Viswanathan, Enming Luo, Chun-Ta Lu, Tushar Dogra, Ranjay Krishna and Ariel Fuxman / arXiv
Agile Deliberation: Concept Deliberation for Subjective Visual Classification
Original source date：Accessed：The authors’ arXiv paper is marked CC BY 4.0. It was first submitted on December 11, 2025 and revised as version 2 on April 4, 2026; it is not a newly published November paper. Only brief methodological context is paraphrased. No figures, charts or full text are reproduced; SHIO’s application discussion is separate.