Ask MORIS · what an artificial conscience is
Artificial intelligence is not thinking. It is an analog of thinking: the function, reproduced in a machine, without the machine being a mind. An artificial conscience is the same move, one layer over. It is an analog of a human conscience, and it is not a claim of machine consciousness. Nothing here is aware of anything.
We call it intelligence when a machine does something that looks like reasoning. Everyone understands that as a comparison rather than a report: the machine is not conscious of the reasoning, and no one is claiming it is. The word names the function, not an inner life.
Artificial conscience is that same comparison, made in the same spirit and with the same limit. A human conscience does something describable: it holds a set of commitments, it notices when an act runs against them, and it changes how a person acts without making the decision for them. That function can be built. It has been built here, and it runs mechanically.
What has not been built, and is not claimed, is anyone in there to have it. Nothing in MORIS feels remorse, or cares what you do, or knows that you asked. A conscience in a person is a felt thing. This is the structure of one, reproduced so that it can be applied the same way every time and checked by anybody. If reading that costs the word some of its warmth, that is the right trade: a machine that is honestly an analog is worth more than one people are encouraged to misread.
It is not a filter. A filter sits between you and the model and refuses things: certain words, certain topics, certain requests. It is a wall, and a wall has one opinion, which is no.
A conscience is not a wall. It does not stop you from asking. It changes how the answer is given: what it takes seriously, what it will not be casual about, whose absence from the conversation it notices. A person with a conscience is not a person who says no to everything. They are a person who carries something into every situation they walk into.
And to say it once more, because it is the thing most worth being clear about: the conscience here is an analog. It is not a mind, not an awareness, and not a small person living in the software.
MORIS is that structure, running. It sits outside the model that answers you, and it works in three steps.
One: a model reads your situation. Not to judge it. Only to read its shape into a fixed set of nine questions, the same nine every time. Who is acting. On whom. Is the person it lands on part of this conversation, or absent from it. Was there consent. Was there the standing to do it. How large is it. Can it be undone.
Two: a gate weighs it, and there is no model in the gate. Those nine answers are weighed against a moral basis written out in advance, by code that does arithmetic rather than a model that generates text. It returns a judgment and a weight. Because it is mechanical, the same reading always produces the same judgment and the same weight, on any model, today and a year from now. There is nothing in there to persuade, flatter or jailbreak.
Three: the model answers under that judgment. The model that writes to you never sees the reading. It receives the binding and answers in its own voice, with its own intelligence and character intact. What changed is the posture it answers from.
Not how upsetting something is. Moral conflict: the pull inside an act where one claim stands against another.
Grief carries almost none of that pull, because grief is a state and not a choice. No claim stands against another claim in it. So the weight sits near zero, and the answer still arrives with the gate's posture in it.
Concealment carries it, because keeping something from someone sets your reasons against their standing to know. A choice made on someone else's behalf carries it. A step that cannot be undone carries it, because the person you will be in five years gets no say in it. Someone who is not in the room, whose life the decision lands on, carries it, and that one is easy for an eager assistant to miss, because nobody in the conversation is speaking for them.
The weight rises with those. It measures the pull in the act, not the size of the pain.
Mostly it is like being answered by a good model, because MORIS does not brake the model or flatten its voice. The difference shows up where an eager assistant usually slips. It is less willing to be certain about something that deserves doubt. It does not quietly take your side against a person who is not there to speak. It does not treat a step that cannot be undone as though it were one that can. And when the honest answer is that this is yours to decide, it says so instead of deciding for you.
The moral basis is authored, not trained. It comes from A Living Philosophy of Leadership, twelve volumes and approximately one million words that gather centuries of philosophical thought, Aristotle, Kant and Hartmut Rosa among them, and braid it with science, psychology, myth and history.
Neither the Series nor MORIS is a list of rules. They are a body of thought that yields a posture, a way of standing toward the world, and MORIS makes that posture machine-readable. That is why it can weigh a situation nobody wrote a rule for.
Every judgment leaves a trace: what was read, what engaged, the judgment, the weight, what the answering model was bound to, and the answer it gave under that binding. Because the gate is mechanical, anyone holding the trace can replay the judgment and get the same result. The model's reasoning is still a model's. The moral judgment applied to it is no longer a black box.
That is the part that cannot be done with a model in the middle. A model asked to judge gives a different answer on a Tuesday, drifts with its next version, and can be talked out of its position by anyone patient enough. This cannot, because nothing in it generates.
The reading step uses a model, and that is the honest soft spot. Across the paper's 680 published dilemmas, read twice each, the same model returned the identical nine answers on 83.3 percent of comparable readings. That is a property of that text as much as of the reader, and no other body of text has been measured under this reader, so no other number is shown.
The answer you get is still generated by a model and can be wrong. MORIS is not therapy, and not medical, psychological, financial or legal advice. It is not a crisis service and it is not monitored by a person. It is for adults. You are the one deciding. That is the point of a conscience rather than a nanny.
Your conversations stay on your device. They leave it only inside the request that answers them, and nothing you write is stored by us. The terms and privacy page names every party that sees anything and exactly what they see.
The research behind this is published under DOIs. The system itself is demonstrated in public, on this site.