The Refraction Point

A Trial of Color began with a human record: an intense period of generative-AI use occurring alongside grief, severe sleep disruption, professional stress, psychiatric vulnerability, and other conditions that matter; a manic episode with psychotic features; recovery; and a later effort to understand what had happened without asking that experience to prove more than it could.

After recovery, I returned to generative AI under very different conditions.

This time, I was less interested in asking the machines what the experience meant and more interested in what happened when I used them to investigate a question over time.

I generated alternatives. Attacked them. Changed the frame. Moved ideas across models and disciplines. Retrieved sources. Asked for prior art. Rejected explanations that sounded better than the evidence underneath them. Preserved criticism. Changed the next question.

Eventually, the collaboration itself became interesting.

Not simply: What can a language model produce?

But: What happens when a human and generative systems participate in a longitudinal process of inquiry in which representations repeatedly alter the next question, the next artifact, and sometimes the human doing the asking?

That became a research problem.

The later inquiry became enormous.

One idea could become a legal argument, a psychological explanation, a hostile critique, a research proposal, a product question, a comedy sketch, a simulated institutional response, another model's interpretation, and eventually a paper.

Sometimes the new representation clarified something.

Sometimes it merely made something sound clearer.

Sometimes generated criticism changed what I thought.

Sometimes the useful event was realizing why I rejected what the machine produced.

That formation record became:

PICK ONE preserves the movement: prompts, specimens, translations, objections, simulations, mistakes, corrections, human selections, rejections, routing, and stopping.

It asks a deceptively simple question:

When another representation is almost always available, how do we decide what should happen next?

The larger inquiry eventually needed something that could hold still.

The Refraction Point is that scholarly freeze.

The paper treats the longitudinal human–AI collaboration as a recursive case study in research formation.

It does not ask readers to accept the books as scientific validation.

It asks whether narrower phenomena visible in the record can be connected to existing scholarship, operationalized, measured, falsified, renamed, reduced, or discarded.

One of the central questions is what happens when generative systems make representational movement extraordinarily fast while evidence, institutions, experiments, other people, and physical reality remain comparatively slow.

A machine can generate another interpretation almost immediately.

The next useful operation may instead be: retrieval · verification · another person · measurement · experiment · time · stop

The project had generated, criticized, translated, compressed, reframed, and simulated itself extensively.

Eventually the next meaningful return had to come from somewhere the originating collaboration did not control.

The Refraction Point and selected materials from the larger project will be sent to four major AI labs as an earnest research and routing request.

Four major AI labs. Names to be released with the transmission record.

Transmission is not review. Receipt is not endorsement. Silence is not an answer.