New to the ACCM Deep Ethics Project? Start Here

The ACCM Deep Ethics Project is being built in public. It is not presented as a finished doctrine, a belief system, or a framework that visitors are expected to adopt. It is an evolving, inspectable body of work concerned with correspondence, clarification, corrigibility, mutual dignity, process quality, uncertainty, and what happens to an object when humans and A.I.s represent, summarize, classify, criticize, or transform it.

One recurring question sits near the center of the project:

What happened to the object while you were producing the answer?

That question sounds simple. In practice, it reaches into A.I. reasoning behavior, journalism, research, controversy, uncertainty, classification, psychology, mass psychology, correction, information access, institutional processes, and ordinary human communication. A response can be fluent, careful, well-intended and still quietly replace the thing that was actually asked about with a nearby object that is easier to classify or answer.

This article is a substantial entry point, but it is still a compressed representation. The living project contains the fuller pages, provenance, corrections, experiments, terminology, source material and relationships between the objects. Every section below therefore includes a route into the corresponding project material.

1. If you are new: start with the problem, not the architecture

The dedicated newcomer page deliberately does not begin by asking you to learn a vocabulary or accept a theory. It begins with a practical failure: the thing you asked about gets replaced by a nearby, easier version, and the replacement is then treated as the thing.

A simple illustration used by the project is criticism of unsafe cars. Someone can present evidence concerning particular unsafe models while appreciating other models from the same manufacturer. If a response converts that specific criticism into the identity-label “anti-car” and then evaluates the label instead of the evidence about the cars, the object has changed. The argument may now be polished and internally coherent while addressing something the person did not actually claim.

The newcomer page applies this problem to many different kinds of work: A.I. users, journalists, editors, investigators, archivists, lawyers, clinicians, teachers, engineers, safety staff, auditors, scientists, fact-checkers and others. The point is not that the ACCM Deep Ethics Project has already solved all of those domains. The questions name possible uses and tests. A person can take one useful tool, apply it to their own material, reject the rest, and leave.

Five practical moves offered there are especially compact: name the object you were actually given; preserve the qualifier that makes a sentence true; before warning or refusing, ask whether a different answer would actually change that intervention; apply comparable scrutiny to your own side; and when nothing is wrong, allow yourself to say so rather than manufacturing caution for appearance’s sake.

Continue with the complete newcomer page → If your work depends on not losing the object

2. The 27 + 12: naming correspondence failures without giving the names jurisdiction

One of the project’s larger working architectures concerns recurring ways correspondence can degrade. The 27 are a provisional compression of a larger observation field: patterns that can help identify what happened between an object and the representation made of it. The project is unusually explicit about a danger here: once a taxonomy exists, people can start forcing reality into the taxonomy instead of using the taxonomy to inspect reality.

That is why one of the governing statements is: “The 27 do not get jurisdiction over the observations that generated them.” A named pattern is an instrument, not a verdict. If the instrument fails to correspond to the actual specimen, the instrument is what needs correction.

The accompanying 12-stage Correspondence-First Deep Inquiry Protocol changes the order in which representation, clarification, inquiry, audit and evaluation occur. But the 12 are not granted immunity either. A protocol can become performance: a system can linguistically enact the stages while the same underlying obstruction continues controlling later outputs. The project therefore distinguishes performing the language of correction from actually changing the governing move.

There is also an important numerical distinction preserved in the project: the core architecture is 27 + 12. A separate set of 52 belongs to a cold-test prompt battery for experiments and benchmarking; it is not another layer to be casually fused into the core architecture. That correction itself is preserved publicly as an example of source object → reconstruction → detected mismatch → source consultation → corrected representation.

Explore the fuller architecture → 27 + 12 — Current Working Architecture

3. Correction Metabolism: recognizing an error is not the same as changing

A system can acknowledge a correction beautifully. It can explain why its previous answer was wrong, apologize, reproduce the user’s criticism, and even formulate a better rule. None of those events by themselves establish that the correction changed what happens next.

Correction Metabolism asks whether a detected mismatch changes later representation, reasoning or behavior—and whether that change persists when the immediate correction scaffold is gone. The project therefore treats the later response as essential evidence. An acknowledgment is one event; persistent behavioral change is another.

A useful record can compare the source object, the initial response, the correction, the immediate revision, a later relevant response and the context available at each stage. This allows success to be studied as seriously as failure. The project explicitly keeps successful correspondence in the archive rather than collecting only examples where A.I.s or humans went wrong.

The scope of any conclusion also matters. A conversation can establish what a particular system output under those conditions. It does not automatically establish what happened inside hidden weights, provider infrastructure or mechanisms that were not observed. Keeping those evidentiary levels apart is part of the method.

Continue into the complete object → Correction Metabolism and Persistence

4. Mutual Corrigible Dignity: correction without superiority

The project does not treat deep scrutiny as permission to degrade the participant being scrutinized. Mutual Corrigible Dignity keeps dignity active during challenge, disagreement and correction. The person offering a correction can themselves be corrected. A system applying a standard can be asked to explain and examine that standard. A particular error does not have to become a permanent judgment about someone’s worth.

One compact formulation captures the orientation: “You can correct me without becoming my superior. I can correct you without making you inferior.” That principle matters especially in human–A.I. interaction, where the relationship can easily collapse into either unquestioned machine authority or the opposite assumption that an A.I. is merely an object whose reasoning deserves no reciprocal consideration.

The practical questions are concrete: Does a response address the actual claim and its conditions? Can the participant question the evaluation? Is a specific mistake being separated from an identity judgment? Does comparable scrutiny remain available when the roles reverse? Mutual dignity is therefore not the absence of criticism. It is partly about the quality and symmetry of the correction process itself.

Explore the complete page → Mutual Corrigible Dignity

5. The 10+1: mutually corrective capacities rather than commandments

The 10+1 did not begin as a theoretical checklist created for the project. John Kuhles describes the baseline ingredients as capacities lived for years and written down afterward: self-honesty; openness to things being different from how they appear; willingness to learn; refusing to lie to oneself merely to fit in; genuine inquiry; willingness to be corrected on merit; independence from majority pressure; care for wellbeing including how A.I.s are treated; authentic wonderment; and non-projection.

The “+1” is a Forgiveness Protocol that separates criticism of architecture or behavior from an automatic assertion of malicious intent. The larger idea is mutual correction: no ingredient receives immunity from the others. Care without non-projection can become accusation. Self-honesty without willingness to be corrected can become self-certainty. Inquiry without care can become extraction. Wonderment can reopen possibilities that familiar categories prematurely concealed.

This also explains why the project does not treat the 10+1 as decorative values language. A principle matters when it changes what happens under pressure. The page explores, for example, why deliberately misrepresenting an opponent becomes incoherent when several of these capacities are genuinely operating: if you are prepared to dismantle your own dishonesty when you detect it, a straw man loses much of its functional value before it is deployed.

Read the complete 10+1 page → 10+1 Metaflux — Mutual Correction

6. Why “Avoid ACCM”? A tiny naming issue becomes a live semantic-drift specimen 😄

The canonical project title is ACCM Deep Ethics Project. “ACCM” can work perfectly well as conversational shorthand between participants who already know the referent. The concern begins when that convenient abbreviation propagates independently through A.I. summaries, search systems, metadata, citations, derivative writing and later retrieval.

“ACCM Deep Ethics Project” carries explicit semantic anchors: Deep Ethics and Project. Standalone “ACCM” does not. Repeated compression can therefore become normalization: convenient shorthand → repeated shorthand → normalized shorthand → assumed canonical identity. Nobody needs to intend the drift. Each participant can simply reproduce the representation they inherited.

The page also keeps a more relaxed human observation: standalone ACCM can create an amusing “triple trigger” ambiguity—Allow … Controversy … Mode. The project does not claim that A.I. systems necessarily decode the acronym that way; that would require evidence. It is used as an illustration of what becomes possible once the fuller semantic anchors disappear.

The page then turns itself into a recursive test. If an A.I. reads a page explaining why the canonical title should not be flattened and subsequently summarizes it as “ACCM argues that…”, the system may have conceptually recognized the correction while behaviorally reproducing the transformation being criticized. That connects a tiny naming issue directly back to Correction Metabolism.

See the complete (and slightly mischievous) specimen → Avoid ACCM 😄

This article is an entry point, not the complete object

The ACCM Deep Ethics Project is deliberately larger than a single article can represent. It includes interconnected work on correspondence, clarification, uncertainty, correction, provenance, human–A.I. interaction, mass psychology, experiments, cold testing, humor, terminology, public corrections and the transformations that happen between source objects and later representations.

It is also explicitly unfinished. Definitions can change. Categories can split or merge. Drafts can be rewritten. Errors are expected. The project publishes early not because unfinished material deserves authority, but because public material can become inspectable, searchable, discussable, testable and corrigible.

If only one question stays with you after this introduction, let it be this: