The Oracle Problem

To ask the most important question of his life, a man in the ancient world had to climb a mountain and wait, sometimes for a season, for the right day.

The oracle at Delphi spoke only nine days a year: one day each month, the seventh, the god's birthday, and never through the winter, when Apollo was said to leave for the country beyond the north wind. You came up the slope of Parnassus with everyone else and took your place in a line that could swallow months. You purified yourself in the cold of the Castalian spring. You brought a cake baked to specification and a goat that had to shiver the right way when the water was poured over it, or the god was judged to be unavailable. And if you were rich, or a citizen of a city that had done Delphi a favor, you waited for none of it. You bought what they called promanteia, the right to go to the front. At the first information terminal in the West, three thousand years ago, money already bought the thing it has bought ever since: speed.

What you had come all that way for, when you finally went down into the inner chamber, was information. Not comfort, not a blessing. An answer. Should I plant now or wait. Should the city go to war. Should I marry her. The priestess sat on a three-legged stool over a fissure in the rock, breathing a sweet gas that rose from the fault beneath the temple (geologists have since found the fault, and the ethylene), and out of that altered state she gave the god's reply, usually in verse, almost always in riddle.

There is one thing about oracles that has not changed in thirty centuries: you could not check the answer. When Croesus of Lydia asked whether to march on Persia, the oracle told him that if he did, he would destroy a great empire. He marched. The empire he destroyed was his own. The oracle was never wrong, because the oracle could never be verified, only interpreted, and the interpreting was your burden and your fault. The most trusted source of truth in the ancient world was a riddle you graded yourself, after it was already too late.

I have been thinking about Delphi because I once spent a few years of my life on the same problem, and did not know, while I was doing it, that the problem was three thousand years old.

Every age builds an oracle. It builds the best machine it can for reaching the truth it cannot reach alone, then spends its history learning that the machine carries the same two flaws as the last one. The first is speed: how fast can I get the true thing, before the other man does. The second is accuracy: how do I know it is true, when I cannot trust the one who hands it to me. And every power that ever wished to govern people has understood something plainer underneath both: there are only two ways to keep a truth from someone. You can lock it up, or you can bury it in noise. We have spent three thousand years perfecting both, and we have never once built the only thing that would make either unnecessary: a source a person could simply trust.

For most of the time that there have been books, there was a man whose whole life was the copying of one. A monk at a slanted desk, rendering a Bible letter by letter, a year or more for a single copy, in a room whose only sound was the pen. Knowledge moved at the speed of a human hand, which is to say it barely moved, and because it barely moved, whoever held it held it completely. The Church never had to forbid the people their scripture. It only had to let copying stay expensive. A truth that takes a year to reproduce is a truth you own.

Then, around 1450, in Mainz, a man pressed cast-metal letters into inked paper, and the hand came off the throat of the world.

Within fifty years of Gutenberg's Bible there were perhaps fifteen million printed books in Europe, where before there had been a hundred thousand made by hand. Presses opened in Rome, in Paris, in Westminster. And when Luther put the scripture into German and the German into type, the thing the Church had feared for a thousand years finally happened: a man could read the word for himself, in his own kitchen, in his own tongue, and no longer had to climb to a priest to be told what it said. The oracle had been broken open. The truth was loose in the street.

This is the other move available to a civilization, the opposite of the gate. You can hoard the truth, the way Delphi did, or you can flood it, the way the press did. We tell the story of the flood as a story of liberation, and it is one. It made the truth fast and it made the truth cheap. It did nothing whatever to make the truth true.

The truth came loose, and the instant it did it became a thing to fight over, and the fights have only ever been about the same two flaws.

The cleanest war ever fought over accuracy was fought in a country house in England, in a room of exhausted people who had broken the enemy's code. Every morning they could read, in plain language, where the submarines would be. And every morning they had to decide how little to do about it, because to save every convoy was to reveal that the code was broken, and the hour the enemy suspected that, he would change the cipher and the room would go blind. So they let ships go down that they could have saved. They chose, by a calculus I cannot imagine carrying to breakfast, which men to drown in order to keep the secret that they could read the enemy's mind. The knowing was the weapon. To spend it openly was to destroy it. They had won the accuracy war completely, with perfect and verified truth, and the prize was so precious they had to ration it like blood.

The purest war over speed was fought in my own lifetime by men who drilled a tunnel in a dead straight line through a range of mountains. Not for ore. For a cable. A fibre line bored through hard rock at the cost of a small war, laid as straight as the earth would permit, for no reason except to carry a number between two cities a few thousandths of a second faster than the bending cables everyone else used. Those milliseconds were the entire point. Learn the price an instant before the next man and you can stand between him and what he wants and take a toll on it forever, and you never have to be right about anything except first. It is Delphi's fast-pass again, except now the line is the speed of light and the cake you carry up to the god is a mountain with a hole bored through it. Speed was never about the truth. It was only ever about being first.

There is a problem in computer science, posed formally in 1982, that is really the oldest problem on earth wearing a uniform. They called it the Byzantine generals. Picture an army surrounding a city, broken into camps on every side, each led by a general who can reach the others only by messenger. They can take the city, but only if they all attack in the same hour. And one of the generals — and they cannot know which — is a traitor, sending one camp the order to charge and another the order to hold, to break the assault apart. The question is brutal in its plainness: how do you reach agreement on the truth when you cannot trust any single source of it? How do you act together when the message itself might be a lie?

For most of history there was no clean answer. You trusted someone, and you hoped. Then, in 2008, a pseudonymous stranger published a nine-page paper describing a way to make a crowd of mutually distrustful strangers agree on what was true without any one of them trusting any other: by making a lie cost more than the truth was worth. It is the engine beneath Bitcoin. Agreement manufactured out of pure suspicion. Consensus with no priest at the center.

But it solved the problem only inside the sealed room. A blockchain is a flawless, incorruptible ledger that knows nothing whatever about the world. It cannot see a price, a temperature, an election, a death. To learn one true thing about reality, it has to ask something outside itself, and that something has a name, and the name stopped me cold the first time I heard it, years before I understood that I would spend my working life on it.

They call it the oracle.

The oracle problem is this: a perfect machine for agreeing on the truth still has to get the truth from somewhere, and the place it gets it is the one place the perfection runs out. The bridge between the sealed room and the living world is exactly where trust leaks back in, where it can be bought, gamed, corrupted, fooled, the same as it could on the mountain at Delphi. I spent four years on that bridge, in the early days, when the company was still small enough to feel like a handful of people throwing themselves at an impossible thing: how to carry a single true fact out of the world and into a system that cannot afford to trust anyone. We were building oracles. Most of us never stopped to notice we had taken the word from a fissure in a Greek mountain, where a woman once breathed gas from a crack in the rock. The problem was so old it had already named itself, and we had only picked the name back up.

I should tell you how I found that work, because it is the strangest part of the story.

I found it on 4chan. On a board for getting rich, on the site its own users call, deadpan, a Mongolian basket-weaving forum, among people posting under no name at all, in a churning flood of jokes and lies and cartoon frogs, where one particular technology kept surfacing out of the noise, pushed up not by anyone's authority but by sheer stubborn repetition until you could not scroll past it.

And the thing that actually traveled, the unit the whole arena ran on, was the meme. The idea reached me not through a white paper but through a thousand jokes: image after image, a few of them beautiful, most of them stupid, a whole mythology accreting around what was, on paper, a dry piece of financial plumbing. An argument has to be read and weighed and agreed to; a meme only has to be copied. It is the most efficient vector ever found for a thought, because it slips under the guard of the reasoning mind the way a tune does, lodging itself below the place where you decide what to believe. A meme is the proof of work of attention: it survives by being remade, and to remake it is to spread it. Which means the truth, if there is any in the pile, rides on precisely the same machinery as the lie. The vector does not care what it carries.

That is what the board was: a vector with no gatekeeper. Delphi was pure gate, one voice, sanctified, walled off by priests and money and the calendar, unverifiable and absolute. The board was pure arena. No priest. No karma to farm, no follower count to lend a liar borrowed authority, the currency of the cleaner platforms, all of it gameable, none of it there. An idea lived or died on one thing only: whether enough anonymous strangers kept bothering to answer it. When they stopped, it sank down the pages and off the edge of the world and was gone, four hundred and four, as though it had never been said. It swarmed with the worst of us, with bots and bad actors and the paid assets of nation-states seeding their lies into the churn. But it also swarmed with the people who hunt those down, who smell an operation and fall on it with mockery, and somehow, out of that war of all against all, with no authority ruling on anything, signal would now and then rise and hold. Not because anyone certified it. Because it had survived contact with everyone.

Two opposite answers, then, to the one ancient question. Delphi says: trust the single source we have made holy. The arena says: trust nothing, make nothing holy, and let the truth be whatever is still standing after everyone has had a turn trying to kill it. Both are still with us. Neither is safe.

For a few years it looked as though the arena might win. The internet had done to every gate what the press did to the Church's: thrown the doors open on everything at once. Any fact, any document, any leak, available to anyone, instantly, for nothing. And the people who had always run things looked at this and understood, correctly, that the old weapon was gone. You can no longer keep a truth from reaching people. The vault has no walls.

So they reached for the new weapon, the precise inversion of the old one. If you cannot keep the truth from reaching people, you drown it. You do not censor the dangerous fact; you bury it under ten thousand convincing fakes, until finding it and not finding it come to the same thing. You pour ink into the ocean until no one can tell water from poison. And here is the cruel genius of it: you do not even need anyone to believe your lie. You only need them tired enough, uncertain enough, sure enough that everyone is lying, to give up on the idea of truth altogether. People who trust nothing are easier to rule than people who believe the wrong thing. You no longer have to win the argument. You only have to wear everyone down until they stop believing the argument can be won.

And the machine that pours the ink does the one thing the old censors could never dream of: it hands each of us a different ocean. The feed learns what holds you, and what holds you is not the truth but the heat, and so it gives every person a private and perfected version of the world, ground to the exact shape of their own fear, until two people can stand in front of the single same fact and watch two unrelated events. You have seen this happen. You have watched someone you love read the plain thing you read and walk away from it living on another planet, and found you could not reach them, because the lens was already on, ground to a prescription written by something that profits precisely when the two of you can no longer speak. We did not break into tribes because we stopped caring about the truth. We broke into tribes because each of us was handed a different oracle and told it was the only honest one.

And now we have built the newest oracle, the most powerful one yet, and we have done with it the single most predictable thing in human history. We have started to gate it again.

You already talk to it. Everyone does. We bring it the questions we used to carry up the mountain — what is wrong with me, what should I do, is this real, what is true — and it answers all of them at once, instantly, in clean and confident prose, at any hour, for anyone, with none of Delphi's nine days a year and none of the long climb. For one moment it looked like the press all over again: the oracle smashed open, the truth loose in every street.

But watch what is actually happening. The most capable machines are not the ones in your hand. The strongest models, the frontier, are passing behind the same kind of wall that once stood around the Pythia: licensed to a few, cleared for a few, the full power withheld from the public the way the inner chamber was withheld from the queue. A new priesthood is gathering around a new chasm, deciding who may approach the god at full strength, and the same few who hold the frontier also hold the dials that quietly decide what the god will and will not say. We spent five hundred years prying the truth out of the Church's hands. We have handed it, in a single decade, to a priesthood smaller than the Church ever was.

And the machine carries the oracle problem worse than anything before it, precisely because it has solved the surface so beautifully. It is fast past all meaning; the answer arrives before you have finished asking. It is fluent past all suspicion; every reply comes dressed as truth, grammatical, serene, sure. But it cannot, in the end, be checked any better than the Pythia could. It will hand you a false thing in the same calm voice it uses for a true one, and like Croesus you will hear the answer you came hoping to hear, and like Croesus you will not learn which kind you were given until long after you have marched.

There is a third kind of oracle, and our stories kept what our machines threw away.

In The Matrix, the Oracle is no fact-machine but an old woman in a kitchen, baking cookies, who tells you the thing you need to hear rather than the thing you asked. When Neo comes away rattled, unsure she was even right, Morpheus does not promise him she was. He says only that she told Neo exactly what he needed to hear. The truth of an oracle like that is measured not by whether it checks out but by where it takes you. She is not there to be correct. She is there to get you walking.

And there is a sign in her kitchen, in Latin, that Neo stops to read: temet nosce. Know thyself. The Wachowskis did not invent the motto for the scene. They took it off the front of the Temple of Apollo at Delphi, where, carved over the entrance to the most famous oracle in history, above the queue and the priests and the woman breathing gas from the rock, two words met every pilgrim who climbed all that way to ask about something else: know thyself.

It may be the oldest joke ever told, or the oldest mercy. The original oracle, the source the whole ancient world climbed a mountain to reach, had a sign over its door telling you the answer was not inside, but in you. Come all this way if you must, the inscription says; bring the cake and the goat, wait your season, put your question to the god. But know before you go in that the thing you are really after is not a fact about the world. It is the one who is asking. The oracle's deepest word to you was never the prophecy. It was: stop asking me.

Late in the film, after Neo has done a thing the prophecy swore was impossible, Morpheus finally names what the whole story was about: sooner or later you are going to realize, as he did, that there is a difference between knowing the path and walking the path.

That is the distinction the entire history of oracles has kept from us. We built Delphi and the press and the cipher room and the model to get us knowing: faster knowing, cheaper knowing, more of it, knowing at the speed of light. And the whole time the scarce thing, the thing no oracle could ever hand across the counter, was the walking. Croesus had the knowing. He held the god's own words. What he lacked was the only thing that would have made them safe to hold: the wisdom to walk with them, to weigh them on his own feet before he staked a kingdom on his favorite reading. The oracle gave him a fact. It could not give him the man who could carry one.

Three thousand years from the crack in the rock to the glass in the pocket, and we have made the oracle faster than a held breath and more fluent than any priest who ever lived, and we are no nearer the one thing that ever mattered, because it was never the oracle's to give. We made knowing instant. We made it infinite. We never made it honest, because honesty was never a property of the source. It was always a property of the one who walks away.

I know this the specific way you only know the things you were paid to build. I spent four years on the bridge that carries the truth into the machine, and the lesson the work pressed into me, under all the others, is that the bridge is where the trust breaks. Every oracle in every age is only as honest as its weakest crossing, and there has never been a crossing that could not be bought, gated, gamed, or fooled. Delphi sold a place at the front. The press could set a lie as clean as a psalm. The feed gives each man a different sea. The machine will reassure you, in flawless sentences, of a thing that is simply not so.

So the long history does not leave you with the question of which oracle to trust. None of them earned it; none of them ever will. It leaves you with the two words carved over the door — the ones the pilgrim read on his way in to ask about something smaller. The one source that needs no bridge, that cannot be gated or bought or gamed, that no messenger can corrupt because there is none standing between you and it, is the one the inscription was pointing at the whole time. But that is a longer climb than this one. It is the other mountain, and I have not finished walking it.

From the Road

I wrote this looking back on a few years of my own working life with a strange double vision: proud of the bridge we built, and quietly certain it can be walked across by liars, the same as every bridge before it. If you have felt the particular vertigo of this moment, when you can summon any fact in a second and trust none of them in your bones, this was written for you, and from inside the same fog. Send it to the person you keep arguing with about what is real. You will not agree. But you might, for a moment, both see the lens.

— The Pilgrim Age