Some years ago a person I trusted looked me in the face and told me something that was not true. I will keep the details out of it, because the details were never the point. It was not a dramatic lie. It was a small one, the kind told to smooth over an awkward evening, the kind people tell themselves does no harm. And I believed it completely, the way you believe someone you have no reason to doubt — which is to say without doing anything at all, without a moment's checking, because checking is not what you do with people you trust. I simply took it, the way we take almost everything.
Weeks later I found out. And what I have never forgotten is what I felt, because it was not what I would have felt if I had merely been mistaken. That same month my weather app told me, with great confidence, that a day would be clear, and it rained on me for an hour and I was soaked, and I felt nothing — a flicker of annoyance, gone by the time I was dry. The app had been wrong. Confidently, badly wrong. But I did not feel hurt by it. I did not lie awake going over it. Nothing between me and the app had been broken, because there was nothing there to break. The app cannot betray me. It can only be wrong.
The person could, and did. And the word for what I felt is not mistaken, and not misinformed, and not that I had updated on bad data. The word is betrayed. Something had been done to me — not just a false belief placed in my head, but something with a moral shape, an injury of the kind other people can do to us and the weather cannot. I had been let down by someone I had made myself open to. And you cannot be let down by someone you were never open to in the first place.
Last time I left you holding one word and asked you to keep it, because I would need it tonight. The word was that one: betrayed. I said that when someone you trusted turns out to have lied to you, you do not feel the clean disappointment of a wrong forecast; you feel something that has an injury in it, and an injury means there was something exposed there to be injured. Hold on to that, I said, because whatever trust turns out to be, it is the kind of thing that can be betrayed — and a weather forecast is not. Tonight we find out what that tells us, and it tells us more than you might expect.
This is Philosophy for Us — philosophy for everyone, no degree required. This is the sixth of eight episodes on how you know anything you did not see for yourself, in a world where the recordings have begun to lie. Last time we reached a conclusion that does not reverse: there is no place to stand outside of testimony from which to check it, no firsthand ground beneath your trust that you laid down yourself. Trust is what you start from, and the checking is only a little verification laid over the top of it. And I handed you a harder question than the one you arrived with — not how do I verify, which has no answer, but whom am I already trusting, and am I doing it well or badly. Tonight we examine what that question is made of. Because even now, after everything last time took away, you are probably still picturing trust as a kind of number — a private estimate you keep of how likely a person is to come through. Tonight that picture breaks as well. We are going to ask whether trust really is just a number you assign to a person, and, if it is not, what the number leaves out. And that begins with the injury.
I want to catch a belief that many of us still hold, even after last time appeared to remove it.
For five episodes we took apart the idea that you are an auditor — someone who grants belief only after verifying, standing on a floor of facts established first-hand. We followed that idea all the way down, and I hope you came out of last time genuinely convinced: there is no such floor, trust is where we start, and we have been trusting all our lives. Good. But almost no one notices what comes with that change of mind, and it is the reason we are not finished. You can give up the idea that you verify before you believe and still hold, underneath it, a particular picture of what trusting itself is — and that picture is mistaken, in a way that is about to matter.
Here is the picture. Ask what you think you are doing when you trust someone — not the verifying, which we have agreed you do not do, but the trusting itself. For most of us, raised in this particular moment, the answer that surfaces is something like this: to trust a person is to keep a running estimate of how reliable they are. A number. A score. You meet someone, you watch how they behave, you note when they come through and when they do not, and you assign them, somewhere at the back of your mind, a rating for credibility — high for the friend who has never let you down, low for the one who is always late, somewhere in between for the colleague you are still reading. To trust someone, on this picture, is just to have given them a high enough number. Trust is your private estimate of the probability that this person, on this matter, will deliver: a bet on their reliability, calculated from their record, revised each time they prove you right or wrong.
And this picture survived last time completely intact. We removed the idea that the number rests on facts you checked yourself; granted, you concede, the number rests instead on other people's word, it is trust all the way down. But the number is still there. You have admitted that it floats, ungrounded, on further trust — and you are still picturing trust itself as the keeping of that number, as though trusting were a kind of forecasting, as though a person were a system to be predicted the way we predict the weather: gather the data, estimate the reliability, place the bet, update on the outcome. You have stopped believing you stand on solid ground. You have not stopped believing that trust is a calculation. Tonight I want to show that it is not, at bottom, a calculation — and that you already know it is not, and knew it long before you heard of this programme, from the last time someone you trusted lied to you.
There is a question that breaks the number. If trusting someone really were keeping a high estimate of their reliability, what happens when they let you down? On the number picture the answer is clean: you got the estimate wrong. You had them at ninety per cent and they turned out to be sixty. New data has arrived; you revise the number down; you recalibrate. That is all that should happen, if trust is a forecast. Being lied to by a friend ought to feel exactly like a weather app being wrong — a failed prediction, a number to adjust, mildly irritating and quickly over. You overestimated; now you know better; move on.
But that is not what it feels like, and you know it is not. When someone you trusted lies to you, you do not feel like a forecaster who set a parameter too high. You feel betrayed. And those are not the same feeling at different volumes; they differ in kind. The forecaster who got the weather wrong feels that he was inaccurate. The person who was betrayed feels that he was wronged — that something was done to him, that something he had given was thrown back, that there was a bond and it has been broken. There is a moral injury in it: the particular distress of having been made a fool of by someone you opened yourself to, which is quite different from the distress of merely having held a false belief. A false belief you correct. A betrayal you recover from, and people take years.
The number picture cannot account for this, because it does not contain the parts. If trust is just a high estimate of probability, then betrayal is just an estimate that was too high — and an estimate that was too high does not wound anyone; it breaks nothing, because a probability is not the kind of thing that can be broken. It can only be wrong. The weather app makes the point. It is exactly what the number picture says trust is — pure tracking of reliability, nothing left over — and it cannot betray you. Not because you do not depend on it; you do, you plan your day around it. It cannot betray you because there is nothing between you and it to break.
What the injury tells us is this: that a person can betray you — not merely disappoint you — and that a thermometer or an app or a tossed coin never can, shows that trusting a person is not the same activity as relying on a thermometer. It is not, at bottom, the keeping of a reliability estimate, because if it were, betrayal would be impossible. The injury is your own evidence, paid for in your own past, that when you trust a person you are doing something the number picture cannot see — something that leaves you open to being hurt in a way no machine, however much you rely on it, can hurt you. You have been treating trust as a forecast. But no one grieves a forecast, and you have grieved, or one day you will. So whatever trusting is, it is something else. Tonight we go and find out what.
To find out, we slow right down on the most ordinary event in the world, one that happens ten thousand times a day and that we have somehow never looked at closely: someone tells you something.
A friend says the meeting was moved to three. A stranger in the street says the station is two blocks that way. A doctor says this will help. A sister, on the phone, in tears, says she is in trouble. In each case a person opens their mouth and puts a piece of the world into your head on their say-so. Through this whole arc we have called this testimony and treated it as a source of evidence — a poorer source than seeing for yourself, one you have to assess for reliability, but evidence all the same, the same kind of thing as a footprint or a fingerprint or a reading on a dial. Someone tells you the meeting moved; that is a data-point; you weigh it; you update. That is the picture we have worked inside since the first episode, the picture that Hume and the careful checker share: a person's word is evidence, evidence that happens to come out of a mouth.
There is a philosopher, Richard Moran, who says: look again, because that is not what telling is. Moran sets out the argument in an essay from 2005 called "Getting Told and Being Believed," and later in a book, The Exchange of Words, in 2018. When someone tells you something, he argues, they are not doing the same kind of thing a footprint does. A footprint is simply there. It does not care whether you believe it; it is not addressed to you; it has no stake in how you take it. It is a part of the world that happens to carry information, and you read it as you read any clue, by inference, on your own, at your own risk. But when a person tells you something, something different is happening. They have turned toward you. They are not leaving a clue for you to find; they are addressing you, speaking to you, and in doing so they do something that has no equivalent among footprints and dials. They give you their word.
And giving you their word is not handing you a piece of evidence. It is making you an offer. Moran's term for it is assurance. When I tell you the meeting was moved to three, I am not setting a fact in front of you and inviting you to evaluate it like a footprint. I am inviting you to trust me. I am saying, in effect: you can take this from me, you need not go and check, I am vouching for it, I am putting myself behind it. I am offering you my assurance, and asking you to take the matter on my say-so — which means I am taking responsibility for it. If it turns out to be wrong, that is now on me, in a way that a misleading footprint is never on the footprint. I have staked myself. I have made myself answerable to you for the truth of it. That is what telling is: not the deposit of evidence but the giving of a guarantee, one person to another, with the teller's own standing as what is at risk.
Now see what believing someone is, on this account, because it is the mirror image, and just as different from what we assumed. If telling is an offer of assurance, then believing the teller is accepting the offer. It is not weighing a data-point and finding it probable enough. It is taking the person up on what they extended — receiving their word, letting them be the reason, entering what they opened. When you believe someone who has told you something, you are not concluding; you are trusting. You are doing the second half of a two-person act. They invited; you accepted. And the whole exchange lives in the space between the two of you — which is exactly the space the footprint picture has no room for, and the space where, weeks later, a betrayal can occur, because a betrayal requires two people and something that passed between them. You cannot betray someone you only left a footprint for. You can only betray someone you told.
Underneath Moran there is an older and barer way of putting it, from the philosopher Annette Baier, in an essay from 1986, "Trust and Antitrust," that more or less reopened the whole subject. Baier asks the simplest question — what is trust, the bare thing itself? — and her answer is not about belief or evidence at all. To trust someone, she says, is to accept being vulnerable to them. It is to put something you care about within their reach, knowingly, and to count on their goodwill not to harm it. That is the whole of it. When you trust a surgeon you are not chiefly estimating her competence; you are lying down and letting her open you, placing your living body within reach of her care and skill and good faith, where she could ruin you and you could not stop her. When you trust a friend with a secret you hand them something that could hurt you and count on them not to use it. To trust is to give someone the power to hurt you and to believe they will not. It is deliberate, knowing, open-eyed exposure to another person's freedom. And that exposure is the thing the injury of betrayal proves you had. The betrayal hurts because you really were exposed; you really had put yourself in their hands; the vulnerability was not a figure of speech but the actual structure of what you were doing when you trusted them, and they used it.
Put the two together, because this is the account built about as well as anyone has built it, and see how much more it explains than the number ever did. To trust a person is to make yourself vulnerable to their goodwill — to hand them, knowingly, some power over you. And the central way they use that power well, its everyday form, is by telling you things: giving you their word, their assurance, and asking you to take it from them rather than go and check, which you accept by believing them, which is to say by trusting them. Knowledge from other people, on this account, is not a degraded form of seeing for yourself. It is something passed between people who have made themselves answerable to each other, and receiving it is not a calculation but a relationship — brief and slight with the stranger giving directions, vast and lifelong with the people who raised you, but always the same shape: someone stakes themselves, and you let them, and for that moment each of you holds something of the other's. That is what you have actually been doing, ten thousand times a day, the whole time you thought you were keeping numbers. You were not forecasting people. You were entrusting yourself to them. And the reason a lie feels like an injury and a wrong forecast does not is that one of them broke a relationship and the other was never a relationship to break.
Now I want to take what we have built — trust as a relationship of staked vulnerability, knowledge as something passed between people who have made themselves answerable to each other — and turn it back to face the phone call we began this whole arc with. Because it does something to the deepfake that nothing else we have said comes close to doing.
Here is the question. When a machine clones your sister's voice and calls you at eleven at night, what exactly is it doing to you? On the old picture the answer was straightforward and bleak: it is feeding you false evidence. It is a very good forgery of a data-point, a counterfeit footprint, and the misfortune is that your instruments cannot tell it from a real one. That is true as far as it goes. But consider what the relationship account lets us say instead, because it goes deeper and it changes the shape of the whole threat. A deepfake of your sister's voice is not your sister telling you anything. Consider that slowly, because it sounds almost too simple. Telling, we just said, is a person turning toward you, staking themselves, offering their assurance, taking responsibility, making themselves answerable to you for the truth. None of that is happening in the fake. There is no one behind it who has vouched for anything. No one has staked themselves. No one has turned toward you and said, take this from me. There is no person involved at all — only a pattern of sound, assembled by a machine, that has the shape of someone telling you something and none of the substance, because the substance of telling is a person putting themselves at risk, and there is no person there to be at any risk.
So the forgery, named precisely for the first time, is this: it is a counterfeit of the relationship itself. It does not merely plant a false belief. It imitates assurance — it fakes the very act of someone staking themselves to you — to borrow the trust that act is owed. And that is why it feels like a violation and not just a trick. The cloned voice does not merely lie to you; it takes on your sister's standing, it borrows the answerability she has built with you over a whole life and uses it, so that the trust you would rightly extend to her is diverted to a machine that has earned none of it and staked nothing. It exploits the exact vulnerability the relationship runs on — your openness to the people you love — by imitating the one thing that vulnerability answers to. That is a more intimate crime than fraud. It does to trust what a forged signature does to a name.
But here — and this part is almost hopeful, though it should not be mistaken for comfort — the same point that names the crime also names its limit. A forgery can imitate assurance, but it cannot produce it. It can take on the appearance of someone staking themselves; it cannot actually stake anything, because there is nobody there to be staked. The relationship is exactly what the fake is reaching for and exactly what it can never supply. Which means the deepfake, for all the fear it causes, does not reach the relationship at its root. Your sister, the real one, can still turn toward you and stake herself; the bond between you — the thing the forgery was trying to steal — is not itself forgeable, because it exists between two people who are actually answerable to each other, and the machine is not a party to it and cannot become one. The forgery depends on trust the way a parasite depends on a host: it needs the real thing to exist first, it produces nothing of its own, and it can only divert what the relationship has already generated. The deepfake is frightening precisely because trust is real and valuable and worth stealing — not because trust is fake.
There is a philosopher, Paul Faulkner, who has pushed this idea of assurance hard over the last two decades, in Knowledge on Trust in 2011 and later work, setting out what it commits you to and where it strains. One thing the whole tradition keeps arriving at is this: you cannot get knowledge from a teller by suspecting them all the way down, because to receive what they are offering you have actually to take it — you have to step into the relationship, not stand outside it auditing. Which brings us, by a completely different road, to exactly where Coady left us last time. There is no standing outside. Last time it was: there is no testimony-free vantage from which to check testimony. Tonight it is: there is no relationship-free vantage from which to receive a person's word. To be told something and actually get it, you have to let them be the reason — you have to trust — and trusting, we now see, is making yourself vulnerable. The thing the panic wants — a way to take people's word while risking nothing, exposing nothing, staking nothing of your own — was never available. To know what other people know, you have to be open to being hurt by them. That was always the price. The deepfake did not invent it. The deepfake merely found a way to charge it to the wrong person.
And one last thing this opens, which I will only point at, because it belongs to a much later week and not to tonight. If telling is a person staking themselves, and a forgery is the shape of telling with no one behind it, then somewhere ahead, when machines stop merely cloning a voice and begin generating the words — talking to you, answering you, seeming for all the world to tell you things — we will have to ask whether there is anyone behind that, whether a machine that produces sentences is telling you anything in the sense we have built tonight, or is only doing to language what the deepfake does to a voice: taking on the shape of assurance with no one inside it to be answerable. That is its own subject, for later. Tonight, simply note the point: the difference between someone telling you and something producing the words is going to turn out to be one of the most important differences there is. For now, back to the injury of betrayal, and to what all of this costs you — because it is about to ask more of you than you expect.
So — what tonight has done, and then I leave you standing in it, without a hand down; except that tonight there is more to stand in, because the reassuring half of this comes with a half that offers no reassurance, and I will not give you the one without the other.
What tonight did was break the number — the picture of trust as a private estimate of someone's reliability — and set a rival beside it, a picture with a person in it. On this account, to trust is to make yourself vulnerable to another's goodwill; to be told something is to be offered someone's assurance, their staked word; to believe them is to accept it, to let them be the reason, to enter a relationship rather than weigh a clue. And on this account that is why a lie from a friend is a betrayal and a wrong forecast is not; why the deepfake is a violation and not just an error; why knowledge from other people was never the cheap, second-best, arm's-length material you settle for, but something passed between people who have made themselves answerable to each other.
Look at why the number felt so natural in the first place, because it says something about the assumptions we live among. You heard, three nights ago, where the word trust came from — credere, the giving of credit, a commitment of oneself into another's keeping — and how, under the long pressure to treat every belief as a bet proportioned to the evidence, it narrowed into a score: your estimate of how likely a source is to deliver, a credit rating in the literal modern sense. The relationship drained out and left a calculation. And that narrowing is exactly why the deepfake era feels like the end of trust — because if trust simply is a reliability you compute, then wrecking your ability to compute it really does abolish the thing. Recover the older meaning, the one the forgery cannot reach, and the panic changes shape. The machine can sabotage your computing. It cannot touch the staking, because it has nothing to stake.
So that is the reassuring part, and it is real, and you should take it. Now the part that gives no reassurance, because if I stopped there I would be misleading you by leaving out the half that does not console.
The relationship picture is clearest, and at its most persuasive, right where the injury is — the sister on the phone, the friend who lied, the doctor leaning over you, the people who raised you. There, someone really does turn toward you and stake themselves, and trusting really is a bond that can be betrayed. But go back to the audit that started all of this — the molten core of the Earth, what is actually in the pill, and the rest of it. Who told you those things? Not your sister. Not a friend across a table. You got them from textbooks written by people who do not know you exist, from institutions that have never turned toward you, from a vast impersonal machinery of reports and records and screens where no single person staked themselves to you and no relationship was ever in the room. And that is most of what you know; it is by far the larger part of it. So the hard question the reassuring answer cannot dodge is this: when nobody is telling you — when there is no person turned toward you, no one made answerable to you, no relationship to be betrayed — what is your trust resting on there? The relationship picture, which explains the phone call so well, falls silent just where the largest part of your knowledge comes from. It explains the intimate case and has little to say about the impersonal one, and the impersonal one is by far the larger.
I am not going to resolve that for you, because it is not resolved, and because the next two episodes live inside that exact gap. Hold both halves at once, because that is the honest place to stand: trust is a relationship of chosen vulnerability, and yet most of what you know reaches you through no relationship you could name. And this does not bring back the auditor, and it does not cancel last time's careful checker. The person who weighs the second source, who asks who benefits, is still doing something real, still doing genuine work inside the trust, and nothing tonight retracts it. We have not replaced checking with feeling. We have found that underneath both the checking and the feeling there is a relationship of staked vulnerability that neither of them is, and that you cannot get out of.
So carry one thing out. The next time someone tells you something that matters — not a machine, but a person, looking at you, putting themselves behind it — you are not, in that moment, computing a probability. You are being offered something, and to take it you have to make yourself, a little, open to being hurt by them. And the harder thing: trusting them well is not just a matter of getting a number right, if there is even a clean number to get, but of being wise about whose assurance you accept, whom you let make you vulnerable, knowing all the while that you cannot verify, that you never could, and that the only thing worse than being open to the wrong person is refusing to be open to anyone — which is its own kind of poverty, and we will come to its cost before the end. Trusting well is staking yourself wisely. And I still cannot hand you a method for it, because there is none. There is only the question, whom and how well, now with a harder edge: whom will you let stake themselves to you, and what are you risking either way. That is not a comfort, and I will not dress it up as one.
But this is where I actually leave you, at the edge of next time. We have found what the forgery counterfeits and cannot supply — the staked relationship, the person behind the word. That is what is robbed. But the pessimist who began this whole panic was never really talking about the intimate call. He was talking about something larger and colder — the whole public apparatus that used to keep tellers honest and let you catch a liar: the recording, the tape, the footage, the thing you could once hold up against a person's word and say, but the camera saw. That was never a relationship. It was an external, public support, standing outside any particular trust, that held the whole thing up. And the claim coming for you next time — the one I have kept at bay since the first episode — is that this support, not the relationship but the public apparatus around it, is the thing the deepfakes are actually taking down. Whether he is right, and whether that apparatus was ever bearing the weight he says it was, is the argument we have next. Tonight you learned what this picture says trust is, and what, if it is right, the fake can and cannot reach. Next time we find out whether the ground beneath the whole practice — the public, checkable, impersonal ground that let trusting well be possible at all — is going out from under us, or whether that too is a story we are talking ourselves into.
Thanks for listening. I'll see you next time.