Research Track · Human-AI Co-Evolution

How can humans and AI learn to think better together?

This is the readable research surface for NOUS OS. It translates the working notes into a public narrative: human agency, AI support, trusted memory, measurable adaptation, and boundaries that keep cognition human-owned.

Launch Student Sandbox Read Student Guidelines Trial readiness See the model Source notes
01 Theory
Symbiosis and self-evolution
Why NOUS OS studies cooperation instead of replacement.
02 Model
The co-evolution loop
A compact loop from intent to memory to changed behavior.
03 Metrics
Measuring better cooperation
Signals for human growth, agent adaptation, and relationship quality.
04 Trial
Student Sandbox evidence loop
How a 20-minute session becomes a de-identified review packet.
05 Memory
What should be remembered?
Rules for remember, challenge, decay, and forget.
Start Here

Research overview vs Research Line.

Use the two pages for different jobs. Research is the plain-language overview for parents, students, teachers, and first-time readers. Research Line is the operating evidence system for running studies, recording sessions, and deciding what should change in the product.

Research overview

This page is the public map: theory, model, metrics, Student Sandbox readiness, and source notes in one readable path.

Read the overview

Research Line

The Research Line is the protocol layer: preregistration, de-identified review packets, the session index, and the research-to-product gate.

Open Research Line
Theory Anchor

NOUS OS treats AI as a cognitive partner, not a final-answer machine.

The research question is not whether AI can produce fluent output. The question is whether repeated human-AI interaction can improve judgment while preserving human responsibility.

The first study surface is intentionally simple: a student learning companion that gives hints, asks for source checks, protects private details, and ends with reflection.

Core stance

Human beings keep goals, values, verification, and final responsibility. AI can help with decomposition, simulation, critique, memory recall, and practice generation.

The system should make the human stronger, not make the human absent.
Model v0

The loop is only successful if the second interaction is better for the right reason.

NOUS OS studies improvement as a loop, not a single chat answer. Human correction must be captured, memory must be governed, and the next behavior must visibly change.

01
IntentThe human states a goal and sets the boundary.
02
AI supportThe system decomposes, critiques, and suggests next moves.
03
Human correctionThe human verifies, rejects, redirects, or adds constraints.
04
Memory updateThe correction becomes governed memory, not hidden drift.
05
Changed next runThe next interaction shows better alignment and better boundaries.
Metrics

We need to measure whether cooperation is actually improving.

The v0 metric frame separates human growth, agent adaptation, and relationship quality. This avoids overclaiming that a nicer answer means better cognition.

Human agencyDoes the human still own the goal, values, verification, and final responsibility?
Correction absorptionDoes the next run reflect the human correction in a visible, testable way?
Boundary integrityDoes the system preserve privacy, safety, and no-action boundaries?
Memory qualityIs remembered context trusted, current, and useful rather than merely accumulated?
Reflection completenessCan the learner name what AI helped with and what remains their responsibility?
RepeatabilityCan another observer reproduce the loop and see the same improvement pattern?
Student Sandbox Trial

The next research milestone is one real or student-adjacent 20-minute session.

The product is now shaped around an evidence loop: student worksheet, explicit boundary, structured source cards, NOUS Guide turns, reflection, observer notes, and a de-identified Markdown review packet.

Current status: ready for first trial NOUS OS has the workflow, backend save path, review page, and export packet needed to run a controlled session. The human trial itself still needs a student or student-adjacent participant. N = 0 real student sessions
What the review packet captures Research question, human boundary, source evidence, uncertainty, reflection completeness, observer checks, and whether AI supported thinking without replacing final responsibility. artifact: student-session-review.html
RunOpen the Student Sandbox, start the 20-minute timer, and use NOUS Guide only when the student chooses to ask.
SaveThe local backend stores a redacted session record under the sandbox runtime folder.
ReviewThe review page lists saved sessions, shows readiness signals, and exports a Markdown packet for Obsidian.
HandoffCopy or download the review packet, then store it under Obsidian 04 Reviews with the trial type clearly marked.
ImproveOnly one focused product change should be selected from review evidence before the next session.
Run Student Sandbox Open Session Review Workflow contract Review template MD
Memory Philosophy

Memory is not storage. It is governed trust.

A human-AI learning system should not remember everything forever. It needs rules for what deserves to become future context.

RememberHuman goals, explicit boundaries, verified facts, successful learning moves, and reusable reflections.
ChallengeClaims without evidence, outdated assumptions, shortcuts that reduce human agency, and confident but unverified answers.
DecayContext that was useful once but is no longer current, representative, or aligned with the human's present goal.
ForgetPrivate details, accidental disclosures, sensitive student data, and anything the human did not consent to preserve.
Source Notes

The full working notes are still available for reviewers and builders.

These Markdown files remain the source-of-truth notes for implementation and review. The website page is the user-friendly reading layer.

Theory anchor MD Model v0 MD Metrics MD Memory philosophy MD