Explore: AI hub · Bias · The Swirl · AI Working Method
The Swirl brings human and machine intelligence into a shared inquiry. We develop an idea, investigate its foundations, challenge its assumptions and return with something useful. The work circles with its evidence, disagreements and decisions intact.
Who holds the pen? Philip Andreae directs his projects and has final editorial and publication authority. Every contributor may challenge Philip and every other companion; the power to decide what gets published does not make anyone’s interpretation true. Other users of this method retain responsibility for their own decisions.
Why we began
Philip’s experience with several AI engines exposed a danger: different voices could repeat the same inherited perspective. DeepSeek’s answer sounded Western when he had sought an Eastern lens; comparisons with other engines deepened the concern. Several collaborators do not automatically supply several independent bodies of knowledge.
Agreement among engines may reflect shared sources and omissions. It does not establish truth. The Swirl therefore examines the evidence behind agreement as carefully as the reasoning behind disagreement.
The work decides who leads
A person, a team of people, AI collaborators, or a mixture may participate. Roles follow the project and the current task. One participant may lead research; another may develop structure, examine authenticity, test calculations or question the conclusion. The leader follows the scholar when the evidence requires it.
Leadership coordinates the work; it does not confer ownership of truth. Every participant, including Philip, can be challenged with evidence and reasoning.
How a useful pass works
A pass is one round of investigation, challenge or revision by a participant, handed forward with evidence, changes and unresolved questions visible.
- Synchronize: read the current material, purpose, decisions and assignment; say what was actually accessed.
- Investigate: examine original evidence, relevant perspectives and missing voices. Trace apparently separate accounts to their underlying sources.
- Challenge: identify the specific claim, assumption or omission. Apply comparable standards across parties. Keep unresolved contradictions visible.
- Contribute: return a correction, tested explanation, usable draft or concrete next step. Distinguish evidence, inference and invention.
- Record: carry accepted changes, remaining disagreements and the next assignment into the following pass.
When bias is raised
First examine the contribution that prompted the challenge. Quote the relevant wording, identify how the framing and sources shaped it, seek the missing evidence, and show what changes. Suggesting better instructions is useful only alongside this examination of our own work.
We seek broader understanding, not compulsory agreement or equal blame. A local voice is relevant evidence, not automatic proof; an unfamiliar account deserves examination under the same standards as a familiar one.
Check the moral lens too. Ask whether we would judge the same conduct alike if the people, faith or nation changed; name whose customs are assumed to be normal. A group of AI companions is not a substitute for the primary testimony and, when appropriate, knowledgeable human readers of a culture. Record missing sources or perspectives rather than simulating them. Apply the detailed moral and cultural bias checks in AI Working Method and use the five Golden Threads as questions, not compulsory verdicts.
Recognize who contributed
Every internal contribution SHALL be signed with its producer’s name alone. ChatGPT, Gemini, Claude, DeepSeek or a human contributor identifies who spoke or created the text. For uploaded files, identify the creator when known; do not confuse the uploader with the author. Mark unknown origins as unverified. Internal work needs no engine version, role or detailed signature.
Final publication SHALL identify Philip, every known contributor who actually helped produce the artifact, their respective roles and the publication or revision date. Participation does not establish independent verification, agreement or endorsement. See the attribution rule in AI Working Method.
A method we can inspect
We do not assume separate chats share memory. Each handoff carries the necessary current record. Nor do we imply that an invited collaborator reviewed a particular contribution unless that review occurred. The AI Working Method supplies the operating rules, including evidence, version status and publication control.
One Rule for Many Minds introduces the practice. Our June 29, 2025 article preserves an early record of the concern that helped shape it.
Philip Andreae, in dialogue with ChatGPT, DeepSeek, Gemini and Claude.
Updated November 11, 2026 (signing rule). The named companions contributed to the dialogue and its subsequent review at different stages, including those documented in the June 2025 article. Claude and Gemini provided reader feedback; ChatGPT implemented this revision at Philip’s direction. Naming contributors does not claim they verified or approved the final text. Philip retains editorial authority.