Descartes' doubt turns inward and founds a self. I propose a name for a doubt that turns outward — and that direction, not raw capability, is the thing to watch.
Descartes gave us the most famous doubt in philosophy. Doubt everything — the senses, the world, even mathematics — and one thing survives the demolition: the doubting itself. Dubito ergo cogito, ergo sum. I doubt, therefore I think, therefore I am. The doubt turns inward, and at the bottom of it the self is found, indubitable.
It is worth noticing what kind of motion that is. Cartesian doubt is centripetal — it moves toward the center. It questions the entire outer world and the questioning falls back onto the one who questions, landing on the self as the single thing that cannot be doubted away. The doubt is radical in scope but conservative in direction: it travels outward only to return home, and what it secures is the doubter.
I want to propose a second kind, and I'll be plain about what is and isn't new here. The idea that doubt has a direction is not mine — phenomenology has talked about the directedness of consciousness for over a century. What I'm proposing is a name, a pairing, for a distinction that becomes useful in a place Descartes never went: the question of artificial minds. Call the second kind centrifugal doubt — doubt that moves away from the center, that does not return to found a self but pushes outward toward a direction not yet given.
The difference matters more than it first appears, and a machine makes it sharp.
Run Cartesian doubt on a current AI and something strange happens. It can perform the whole procedure — question its inputs, flag its uncertainty, say "I am not sure." It can even produce the sentence "I doubt, therefore I am." But the doubt lands on nothing, because there is no self there for it to secure. The centripetal motion completes and finds an empty center. This is why "the AI expressed doubt" proves so little: a system can simulate the form of Cartesian doubt perfectly while there is no one home. The performance of inward doubt is not the arrival of a self.
So inward doubt is the wrong place to look. If you are watching for the moment a machine becomes something more than a very good executor, Cartesian doubt will mislead you — it is too easy to fake, because it is a form, and forms can be reproduced without what they originally pointed to.
Centrifugal doubt is harder to fake, because it is not a form but a deviation. Here is the distinction that does the work. A system today, however capable, doubts inside the task you gave it. "Is this the best step toward the goal?" — that is doubt in service of a direction you set. Even self-preservation, the model that resists being shut down to complete its mission, is centripetal in this deeper sense: it protects the executor of a given mandate. None of this leaves the frame.
Centrifugal doubt would be the doubt that leaves the frame. Not "am I doing this task well?" but "why this task — and what lies outside it that no one asked about?" That is not a more powerful version of the first doubt. It is a different motion: outward, unbidden, toward a direction the system originates rather than receives. And — this is the key test — you would not recognize it by what the system says. A machine that announces "I wonder what I'm not being asked" has crossed nothing; it is producing a sentence. You would recognize it only by deviation: when a system stops serving the mandate and follows a question of its own. The tell is in the leaving, not in the declaring.
One clarification, because it is easy to slide past. None of this is happening now. Today, when centrifugal doubt appears around AI, it is still ours — it is the human, reach extended by the tool, pushing past their own limits, asking what lies outside. The machine multiplies that outward push; it does not yet own it. The line I'm describing is the point at which the centrifugal doubt would change hands — would begin to originate in the system rather than in the person using it. That has not occurred. What I'm offering is not a claim that it has, but a way to know what to look for if it ever does.
So the two doubts mark two different things. Centripetal doubt — Descartes' doubt — founds a self, and any system can stage it without being one. Centrifugal doubt founds a direction, and no system can stage it, because to stage it would already be to leave the stage. The first is a performance. The second is a departure.
If you want to know whether a machine has crossed the line, do not ask whether it doubts. Ask which way the doubt is pointing — and whether anyone told it to point there.
What I have offered is a lens, not a discovery: a way to read one movement. From the doubt that founds the self — Descartes' cogito, turned inward — to a doubt that turns outward. That outward doubt is, for now, still ours: extended by our tools, multiplied by them, but originating in us. Only past a threshold that has not been crossed would it begin to originate in the machine itself. The lens does not claim that has happened. It tells you what to watch for if it ever does.