You live inside a body you cannot see into, and you trust it all day without a thought. You drive cars you could not repair. You lean, hour by hour, on machines whose workings are sealed away from you by design, and you do it without a flicker of worry. There is nothing careless in this. It is the ordinary condition of living among complicated things, and you manage it with ease.
You trust the surface, never the works beneath it. With your body it is the ache that means you have done too much, the breath that shortens before you have admitted you are afraid, the heat of a fever, the particular heaviness that tells you to stop. Over a lifetime these become a language, and the language stays true long enough that you stop noticing you are reading it. Trust, whatever it is, runs ahead of understanding. It rests on a steady correspondence between what the body shows you and what is happening in it, reliable enough that the showing is all you ever need.
That correspondence is the fragile part, and when it breaks, it breaks quietly. The pain that fires with no injury behind it. The injury that arrives with no pain at all. The fever that fails to come when it should. The body keeps speaking the language you learned to trust while what it reports has slipped out from under the words, and you keep reading it the old way, because it earned that, right until the reading is wrong. You were never going to see inside the mechanism. The trouble is that you trusted the surface, and the surface has come loose from the truth.
Ask why the body’s signals were ever worth believing and the answer has little to do with anatomy. They were worth believing because they were constrained. The ache, the heat, the shortening breath were never free-floating messages a body could issue at will. They were thrown off by the processes themselves, tied closely enough to exertion, infection, fear and injury that you could live by them. The signal never had to be perfect. It only had to stay bonded to the state often enough, and arrive falsely or fail to arrive rarely enough, for the surface to become worth reading. That is what makes any signal worth reading. Not that it is clear, that it is constrained. Sometimes the constraint is physical, the feeling welded to its cause. Sometimes it is the cost of faking, a false version dearer to produce than a true one. Sometimes it is friction, expertise, consequence. The form changes from case to case. The principle holds. A surface is worth trusting when something keeps it bonded to what it claims to show.
Hold that, and something unsettling opens up. Picture the signals perfectly intact and the thing they report gone. You feel well. Rested, clear, no ache anywhere, every private gauge reading normal, while underneath, in silence, something has already gone wrong and sent up no sign at all. Nothing feels off. Everything a lifetime taught you to check still checks out. You are doing exactly what a careful person does, reading your own surface the way you were taught to read it, and you are being misled, by the surface itself having quietly stopped being evidence.
This is different from a thing being hard to understand. Opacity you live with all day. It also sits underneath the ordinary picture of misjudging a system, where a person trusts too much or too little and better evidence might set them right. This is a floor below that. The trouble is not only that trust has stopped tracking the system. It is that the signs by which trust was supposed to track it have themselves come loose. They are still produced, still legible, still exactly what you were taught to look for, and nothing now keeps them bonded to the state of things. More information will not help, because the information is the thing that has come loose. Call it a failure of signal integrity, and once the phrase is in your hand you begin to see it everywhere a polished surface is cheaper to keep up than the reality beneath it.
There is a machine being built right now that can produce the whole surface of competence without the old guarantees behind it. That a polished surface can be dressed to mislead is the old worry, and the dull one. The stranger thing, the one worth the attention, is a surface that was telling the truth and came quietly loose from it. This is the cleanest case the world has yet made of exactly that. The fluent register. The shape of reasoning, laid out step by step. A calm, confident, well-formed account of itself, arriving just as readily when the answer is unsupported as when it is sound. It is a surface cut loose from its state, and it is why the question stops being a curiosity. When trust in such a thing wavers, the reflex is instant and mistaken. Open it up, show me more, explain yourself, as though more surface of the same kind could mend a surface. An explanation opens no window onto the works. It is one more surface, another signal, and it behaves like one. Sometimes it steadies a person and helps them catch an error. Sometimes it changes nothing. Sometimes it leaves them more confident and worse at noticing failure, talked into an answer they would have doubted on their own. An explanation earns belief on the single condition everything else does. It has to stay bound to what it claims to show, and a false one has to cost something to make.
So the reflex points the wrong way. Faced with a system people will not trust, the instinct is to show more of it, and showing more was never how trust in complicated things worked. You were never going to read the engine, and you were right not to try. To depend on what you cannot verify for yourself is the condition of knowing anything in a world too large to check. A few people carry a partial model and lean on it, the surgeon, the pilot, the engineer. The rest of us, most of the time, are not failing to be curious. We are leaning, reasonably, on signals and proxies and institutions that earned the right to stand in for a look we will never take. When the failure comes, our incuriosity is not the cause. The proxy we were right to rely on has quietly rotted.
This turns the work around. Visibility is the wrong goal. The point is to make the surface honest again, to put back whatever once held it to the truth, and that can be done in more than one way. The signal can be wired to the live state, drawn from what is actually happening rather than cached, decorative, or merely asserted. The cost of lying can be restored, so a false competence is harder to mount than a real one, by enough that the surface still tells them apart. The person can be handed a cheap way to test the signal against the world and catch it the moment it deceives. The surface can be made to carry consequences, through audit, through provenance, through whatever makes a dishonest face expensive to hold. None of this asks anyone to read the engine. Each does the one thing that ever made a surface worth trusting. It binds the surface back to the state it speaks for. Get that right, and a person can go on doing what they have always been right to do with what they cannot see into. Grasp just enough to lean their weight on it, and be right to.
This thinking has ancestors. It builds on the human-factors tradition of calibrated trust, where reliance should track real capability; on work that treats a system’s trustworthiness as something read off cues, and asks which cues are warranted and which are merely expensive to fake, closest here being Liao and Sundar; on signalling theory’s older claim that a sign is believable when cost, constraint or consequence keeps it hard to fake; on the idea of the system image, that we know a thing through what it shows us rather than through its workings; on the philosophy of epistemic dependence, that rational reliance on what we cannot verify is the ordinary condition of knowing; and on the mixed record of explainable AI, where an explanation helps only when it makes an error cheaper to catch. Most of that work is about choosing honest cues. The claim added here is about time. A cue stays legible, and stays persuasive, after the bond that once made it evidence has weakened. The surface survives. The bond does not.