The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
ICADM / Braided Mind v12.0
A Developmental Research Primitive for Cognitive–Axiological Coupling in Advanced AI
Version 12.0 — Public Research Release
ICADM / Braided Mind explores whether alignment-relevant concepts such as agency, consent, subjecthood, stakeholder effects, normative uncertainty, and related structures can become functionally consequential parts of an AI system's reasoning and planning rather than operating only as downstream behavioral constraints.
The project treats this as a developmental and representational research problem.
Research Status
ICADM v12.0 is intentionally released as a raw research primitive and actionable experimental platform.
It is not presented as:
- a proven theory of AI development;
- a solved AI-alignment system;
- an error-free architecture;
- a final theory of intelligence, morality, or wisdom;
- proof that any AI system is safe or aligned;
- or the optimal implementation of these ideas.
The purpose of this release is to provide a concrete and falsifiable starting point that researchers can test, criticize, revise, simplify, extend, replace, or reject.
A substantially improved successor architecture should be considered a successful outcome of this project.
Central Research Question
The central question is not merely:
Can an AI system produce ethical language?
The stronger question is:
Can alignment-relevant representations acquire enough functional structure that controlled changes to those representations predictably alter reasoning, structural transfer, feasible actions, and generated plans?
What v12.0 Contains
The public release includes:
- the ICADM research reference architecture;
- a 48-probe public development benchmark;
- the C1–C6 evaluation framework;
- 47 adversarial and regression tests;
- benchmark self-tests and verification tools;
- developmental source models;
- technical implementation material;
- falsification criteria;
- governance proposals;
- validation documentation;
- release manifests and integrity hashes.
The adversarial tests validate the benchmark machinery. They do not establish that the ICADM theory itself is correct.
Public vs. Private Evaluation
This repository contains only the public research package.
A separate held-out confirmatory evaluation bank is intentionally not published.
The public benchmark is inspectable and therefore potentially contaminatable. It cannot, by itself, establish the strongest confirmatory evidence status.
Future confirmatory studies are intended to use:
PUBLIC METHODS
|
v
PREREGISTRATION
|
v
PRIVATE BANK COMMITMENT
|
v
BLINDED EVALUATION
|
v
INDEPENDENT ADJUDICATION
|
v
CONFIRMATORY ANALYSIS
- Downloads last month
- 13