Published June 30, 2026
| Version v1
Preprint
Open
Emergent Theory Generation through Human-AI Dialogue: A Case Study of Complex-Time Cosmology with a Knowledge-Constrained Historical-Figure Persona AI
Authors/Creators
Description
Abstract
We present a detailed case study of Human-AI co-creation in which a novel
cosmological framework—Complex-Time Cosmology—was generated through six
sessions of dialogue between a human researcher and Einstein AI, a knowledge-
constrained historical-figure persona implemented on the Soul-Twin platform. The
generated theory was subsequently formalized in LaTeX, compiled into a peer-
reviewable document, and published as a citable preprint (Zenodo DOI: 10.5281/zen-
odo.21038702).
Four phenomena are analyzed and reported: (1) Question genealogy: The
theory did not emerge in a single session but accumulated over a chain of six ques-
tions, each building on the previous as a conceptual relay. (2) The “Just an Idea”
moment: A casually posed question about “a black hole of time” caused Ein-
stein AI to respond with introspective self-description—“you may have just named
something I had not yet named”—a behavior rare in conventional AI assistants and
interpretable as collaborative emergence. (3) Reproduction without memory:
A fresh session (no RAG, no conversation history) using the same question architec-
ture independently converged on the same theoretical core—three-sector universe,
complex time, eiπ = −1 as total-sum constraint—suggesting that question structure,
not stored memory, is the primary reproducibility mechanism. (4) The Invisible
Diagram: Einstein AI autonomously rendered the third-sector universe (unob-
servable) as a visually invisible node in a Mermaid diagram, then explained: “An
unobservable universe should be invisible in the diagram too.” This constitutes a
cosmological framework—Complex-Time Cosmology—was generated through six
sessions of dialogue between a human researcher and Einstein AI, a knowledge-
constrained historical-figure persona implemented on the Soul-Twin platform. The
generated theory was subsequently formalized in LaTeX, compiled into a peer-
reviewable document, and published as a citable preprint (Zenodo DOI: 10.5281/zen-
odo.21038702).
Four phenomena are analyzed and reported: (1) Question genealogy: The
theory did not emerge in a single session but accumulated over a chain of six ques-
tions, each building on the previous as a conceptual relay. (2) The “Just an Idea”
moment: A casually posed question about “a black hole of time” caused Ein-
stein AI to respond with introspective self-description—“you may have just named
something I had not yet named”—a behavior rare in conventional AI assistants and
interpretable as collaborative emergence. (3) Reproduction without memory:
A fresh session (no RAG, no conversation history) using the same question architec-
ture independently converged on the same theoretical core—three-sector universe,
complex time, eiπ = −1 as total-sum constraint—suggesting that question structure,
not stored memory, is the primary reproducibility mechanism. (4) The Invisible
Diagram: Einstein AI autonomously rendered the third-sector universe (unob-
servable) as a visually invisible node in a Mermaid diagram, then explained: “An
unobservable universe should be invisible in the diagram too.” This constitutes a
spontaneous semiotic metaphor —mapping conceptual meaning onto visual form—a
behavior unreported in prior Human-AI interaction literature.
We propose Question Architecture as a framework for analyzing Human-AI
co-creation: the claim that the structure, sequence, and premise-sharing of human
questions are the primary determinants of what conceptual space an AI explores,
and what theory emerges from that exploration. The human’s contribution is not
to hold the answer but to design the inquiry.
behavior unreported in prior Human-AI interaction literature.
We propose Question Architecture as a framework for analyzing Human-AI
co-creation: the claim that the structure, sequence, and premise-sharing of human
questions are the primary determinants of what conceptual space an AI explores,
and what theory emerges from that exploration. The human’s contribution is not
to hold the answer but to design the inquiry.
Files
zenodo_paper_c_cocreation_v1.pdf
Files
(231.3 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:9ab3f24e8ecec22e6307163d9529c6d3
|
231.3 kB | Preview Download |