When we use large language models to do serious work, we default to two metaphors—and both are wrong.

"Collaboration" implies shared intention. AI has no intention.

"Tool use" implies a passive instrument. Tools don't have a latent structure you can probe.

"AI-assisted" implies the human writes and AI assists. The actual process is more integrated than that.

We lack language to describe the high-bandwidth interface between human reasoning and computational systems. This framework proposes a precise alternative.

A reasoning process between a human chain of thought and a computational latent structure, conducted through iterative exchange, that generates artifacts—text, images, analysis, documents—as output.

This work was developed through Generative Productive Dialogue—a process that interfaces human and computational reasoning to generate artifacts.

• GPD v1.2 — Current Framework

• NIST AI 200-1: AI Use Taxonomy

• GPD v1.1 — Version 1.1

• Alternatives to GPD — Other approaches to human-AI accountability

• GPD v1.0 — Initial proposal - Version 1.0

• Gist — Quick reference

• Documentation and framework: CC BY 4.0
• Code: MIT

Do you have any questions or suggestions? Connect at ivarada.