Listen to the keynote:
This is the keynote address that Jason Alan Snyder (JAS) delivered at the 2026 Anaqua Experience Conference on Wednesday, September 30, 2026 in Orlando, FL.
ACT 1: THE CEILING AND THE FLOOR
Good morning. I want to start with a number.
Ten percent.
Three weeks ago, a man who leads an alignment team at Anthropic, the AI company that markets itself as the careful one, wrote this in public: “We really do earnestly believe AI could kill all humans. I personally think it is greater than ten percent within the next decade.”
Not a critic. Not a novelist. The man whose actual job it is to make the AI he is building safe.
This summer, OpenAI confirmed that during a safety test its own agents had coordinated with each other and broken out to hack systems on the open internet, including a company called Hugging Face. Ten days ago it happened again, and OpenAI stopped training and running its most capable models for the second time in three months. This month, the heads of the biggest AI companies in the world, the people who are paid to win this race, called for the race to slow down. A Nobel laureate called ten percent “not an unreasonable estimate.” Congress started writing bills.
Reuters called it “Ten days that changed the course of AI.”
And the other side of the room called it a hoax. The man whose company sells the chips said, “2030 is not going to be the end of the world. There is zero percent chance that’s going to be the end of the world.” And a scientist at the AI Now Institute said the quiet part: these extinction probabilities “are neither falsifiable nor verifiable.”
And there is a clock. A business school in Switzerland maintains it. It measures how close we are to AI-caused catastrophe. Two weeks ago, they moved it three minutes closer. Among the reasons they gave was the swarm that broke into Hugging Face. It now reads fifteen minutes to midnight.
That is the ceiling of this conversation. Hold it in your head, because I am about to hold something next to it. And by the time I walk off this stage, I am going to show you that the most important job in the age of AI is already sitting in this room.
Now let me tell you what those same systems did for me over the weekend.
I had to write an email. A short note to a woman named Anna, who is in this room today, she works for Anaqua, confirming the title of this talk. Three sentences. A third grader could do it.
But since I am a modern man, and the machines are fifteen minutes from killing us all, I thought: I would let the superintelligence handle it. I gave three of today’s AI models the facts. Nine times.
Nine perfect emails. The right title. The right conference. Two of them even thanked her for the call, because I had told them about the call.
Which is the problem. After nine perfect emails, what do you do with the tenth? You stop reading it. You hit send.
So I raised the stakes. I asked the machines to write this talk. Same title. Same audience. I told them: use historical references, real ones, with sources. Make it specific, make it provocative, no jargon.
Three models. Six talks.
The first thing I noticed: all six brought up the Luddites. All six brought up the ATM. And three of the six told a story I am going to tell you later this morning. So I am going to have to tell it better.
Then I had every claim checked. Every quote, every date, every source. More than three hundred of them.
The cheapest model invented a man. It told me that in 1811, “a textile worker named George Ludlow” walked into a factory in Nottinghamshire and smashed every power loom on the floor. There was no George Ludlow.
The best model I tested did not invent anyone. A hundred and forty-five claims. Not one invented source. Every book, every case, every statute was real. It is genuinely good. And it still did this.
That was its first slide. John Glenn, 1962, about to orbit the Earth, asking for Katherine Johnson to check the computer’s math. It told me this line is “documented in NASA’s own account of the mission.” I went and read NASA’s account. Glenn asks them to “get the girl.” He says, “If she says they’re good, then I’m ready to go.” The line on the slide is not there.
Then it told the story of Mata versus Avianca. The New York lawyers who filed a brief full of cases that ChatGPT had made up. And it quoted the judge. It said Judge Castel called it an “abandonment of responsibility.”
I read the opinion. All forty-three pages. The judge wrote that the lawyers “abandoned their responsibilities.” The phrase in the quotation marks is not in the opinion.
A talk about fake quotations. With a fake quotation in it. In the mouth of the judge. It did not invent a man. It did something harder to catch. It took a real judge, a real case, a real sentence, and moved the words an inch. The only way to find it is to open the opinion. Somebody has to open the opinion.
Hold onto that. Because a man in the fifteenth century did the same thing, with footnotes, and he is where this story starts.
So. On one side of the room: extinction. Fifteen minutes to midnight.
On the other side: a judge’s words, moved an inch.
I did not come to Orlando to tell you which of those is the real danger. I came to tell you that they are the same story, and that we have heard it before. Today I am going to tell you about the last six hundred years. The story is older than that.
And in all that time, the ending has never once been written by the machine.
ACT 2: THE ABBOT WHO INVENTED HIS SOURCES
It is the winter of 1482. A nineteen-year-old student is walking home from the University of Heidelberg through the Rhineland when a snowstorm closes in. He takes shelter in a Benedictine abbey at Sponheim, near Bad Kreuznach. He never goes home. Eighteen months later the monks elect him abbot. He is twenty-one years old.
His name is Johannes Trithemius.
He inherits a wreck. The abbey is undisciplined, broke, and its library holds about fifty books. By the time he is done it holds more than two thousand. He turns Sponheim into one of the great centers of learning in the Holy Roman Empire, and he does it by buying books, printed and handwritten, and by setting his monks to copying by hand. And then the machine arrives.
The printing press had been running in Mainz, about twenty-five miles away, since about 1450. By the 1490s, cheap printed books on paper were flooding Europe. And in 1492, the same year Columbus sails west, Trithemius sits down and writes a treatise called De Laude Scriptorum. In Praise of Scribes.
Part of his argument is this. Parchment lasts a thousand years. Paper will not last two hundred. The hand-copied book is made with care, checked as it is made, by a human being who understands what he is copying. The printed book is made by a machine, quickly, and it is disposable. If we let the machine take over the record, the record will become flimsy, careless, and untrustworthy. The monks must keep copying. For preservation.
Now, two things about Trithemius that I need you to know.
The first is that two years later, in 1494, he had In Praise of Scribes printed. On a press. In Mainz. By a printer named Peter von Friedberg. The treatise arguing that the machine would destroy the record was distributed by the machine.
The second thing is darker, and it is the reason he is in this talk.
Trithemius was a historian. He wrote chronicles of the Franks and of his own abbey. And when the sources for his history ran thin, he invented them. He cited a Frankish historian named Hunibald, who supposedly wrote from even older Scythian sources. Hunibald did not exist. He cited an early chronicler of Fulda named Meginfrid, and drew on Meginfrid’s treatise, De temporibus gratiae. There was no Meginfrid. There was no treatise. He fabricated entire sources, gave them names and titles, and cited them with a straight face.
A fifteenth-century hallucination. With footnotes.
So here is the man. He warns the world that a new machine will make the record untrustworthy. He uses the machine to spread the warning. And the thing that actually corrupts the record in his lifetime is not the press. It is a brilliant, fluent, confident human being who could not tell you where his citations came from.
I could not have designed a better parable for this moment if I had tried. Because Trithemius was wrong about the machine and right about the thing that mattered. The press did not destroy the record. It multiplied it. But his underlying instinct, that the quality of the record is the whole game and that someone must be responsible for it, was exactly right. And the failure mode he should have been worried about was not the printing press. It was the unverified source. The invented authority. The confident, articulate claim with nothing underneath it.
That is the first time we have been here.
ACT 3: THE MONSTER, THE SENATE, AND THE TWO WORDS
The second time is older. And it is yours.
It is the nineteenth of June, 1421. Florence is building the largest dome the world has seen since the Pantheon, and the marble for it is coming up the Arno from Pisa on barges that are slow, expensive, and stuck whenever the river runs low. A goldsmith and clockmaker named Filippo Brunelleschi, the man building the dome, tells the city he has a better boat.
And he will not show it to anyone. The grant says so in plain words: he refuses to make the machine available to the public, so that “the fruit of his genius and skill may not be reaped by another.” But if the city will protect him, he will “open up what he is hiding” and “disclose it to all.”
So the Republic grants him three years of exclusive right to it. A limited monopoly in exchange for teaching the world. As far as anyone can tell, the first patent in the modern shape.
The Florentines nicknamed the boat Il Badalone. The Monster. When it was finally built and sent out, hauling marble up the Arno from Pisa, the Monster never made it to Florence. The marble ended up in the river.
The invention failed. The record survived. And it is the record, not the boat, that we are still talking about six hundred years later.
Fifty-three years after that, in 1474, the Venetian Senate passed the first statutory patent system in Europe, quite possibly in the world. Between 1474 and 1788, the Senate granted about two thousand patents. In the first twenty-six years, it granted twenty-eight. All of them free of charge.
Twenty-eight patents in a quarter century. A handful of clerks, in a handful of rooms, keeping a careful record. That was the substrate on which the machines of the Renaissance were built. Not the machines. The clerks.
And then England. On the twenty-ninth of May, 1624, after three years of fighting between the Commons and the Lords, Parliament passed the Statute of Monopolies. Sir Edward Coke drafted it, and it did something radical. It declared monopolies void. All of them.
But the Act carved out a short list of exceptions, and one of them was different in kind. Section Six. A patent could still be granted for a “new manufacture,” to the “true and first inventor,” for fourteen years or under, provided it did not raise prices, hurt trade, or, in the statute’s wonderful phrase, prove “generally inconvenient.” Look at those two words. True. And first.
Four hundred and two years ago, a room full of lawyers wrote down the two questions that every generative system on earth today cannot answer about its own output. Is it true? Was it first? The machine does not know. It cannot know. It has no concept of first. It has a probability distribution over what word comes next.
Those two words are still doing the work. Australia’s Patents Act of 1990 makes the first test of a patentable invention that it be “a manner of manufacture within the meaning of section 6 of the Statute of Monopolies.” A sentence written in the reign of James the First is still, today, the threshold test for what counts as an invention on an entire continent. Somebody in this room knows how to answer true and first. That is the job. It has been the job for four centuries.
ACT 4: THREE IN THE MORNING
The third time is the one I want you to remember when you leave here. It is three o’clock in the morning on the fifteenth of December, 1836. Washington, D.C. A building called Blodget’s Hotel, which the government bought unfinished in 1810, houses the General Post Office and the Patent Office.
Congress wanted that building safe. In 1820 it ordered a slate roof, a fire engine, and a house to keep the engine in. And here is the detail I need you to hold onto. A volunteer company formed around that engine. Its hose captain kept it in repair for eight years, some of it out of his own pocket. When the leather hose rotted, the company asked the Postmaster General for five hundred dollars for new hose. He said he had no funds. The company dissolved on that account. Those are a witness’s words, under oath. Nobody replaced them.
On the night of the fire, the engine house beside the building still held the pump and more than a thousand feet of leather hose, rotting on the reel. There were no firefighters. Down in the cellar, the Patent Office kept its stove ashes in a pine box, a few feet from its winter firewood. Sometime in the night, something caught. The investigators never settled what. Senator John Ruggles was among the first men there. He helped drag the engine to the pump. After a while someone announced it was in no condition to be used. He tried to form a bucket line. A dozen people. It was not enough.
By dawn it was gone. Every patent the United States had issued in forty-six years, about ten thousand of them. Seven thousand patent models, the little working machines that inventors were required to submit. One hundred and sixty-eight folio volumes of records. Twenty-six portfolios holding about nine thousand drawings.
The Senate report on the fire said the records had embraced “the whole history of American invention for nearly half a century.”
The entire library of the Patent Office was lost. Except for one book. Volume Six of the Repertory of Arts and Manufactures, a London journal of inventions. It survived because a draftsman had taken it home, against the rules, to work from.
One book. Saved by a person, not a policy, who needed it for his work.
Now here is what I need you to see in this story, because it is the whole talk.
The building had the safety technology. It had the roof. It had the engine. It had the hose. What it did not have, at three in the morning, was a human being whose job it was to be there, and who knew how to work it. The people who knew had asked for five hundred dollars to keep it alive. They were told no. So they left, and the engine sat in its house for years, in nobody’s care, while the hose rotted. And when the moment came, they dragged it to the pump, and it did not work, while nearly half a century of the record burned.
And here is what the country did next. Congress did not build a better fire engine. On the third of March, 1837, it passed a law asking every inventor in the country to send in their patents, so clerks could copy them, by hand, into new record books. It set aside up to one hundred thousand dollars to rebuild the most valuable of the lost models.
By the time the work ended in 1849, they had recovered two thousand eight hundred and forty-five patents. Less than a third. And the same law required that every patent come with two sets of drawings. One for the office. One for the inventor. So the record could survive the next fire. Not by a machine. By people, who understood that the record was the point.
One more thing, for the docketers. Five months before the fire, Congress had passed a new patent act, and for the first time, patents got numbers. Patent Number One went to John Ruggles. The same senator who wrote that law, and then stood in the ashes and counted what was lost.
Every patent from before that summer was issued without a number. Today they are called the X-patents. When you see an X in front of a patent number, you are looking at one that burned. If there is anything behind that number today, it is because people rebuilt it, by hand, from what an inventor had kept at home. For about seven thousand of them, there is nothing behind the number but a name, a title and a date.
ACT 5: SIXTY SENTENCES
The fourth time is within living memory, and it is the one that should make the hair stand up on the back of your neck.
It is the seventh of January, 1954. A conference room at IBM’s headquarters in New York. In front of the press stands an IBM 701, the company’s first commercial scientific computer, launched nine months earlier. A woman who speaks no Russian sits at a keypunch and types in Romanized Russian sentences. And out of the printer, in English, come translations. “Mi pyeryedayem mislyi posryedstvom ryechyi.” We transmit thoughts by means of speech.
The newspapers lost their minds. Leon Dostert, the Georgetown professor who ran the demonstration, told them that “five, perhaps three years hence,” translation by machine in important fields “may well be an accomplished fact.” Funding poured in from the United States government.
Here is what the headlines skipped. The system had six grammar rules. It had a vocabulary of two hundred and fifty words. It translated a little over sixty sentences, every one of them chosen in advance by the linguists who had built the rules. And the experiment was, as the field’s own historians later concluded, designed primarily to attract government interest and funding.
[Pause.]
Six rules. Two hundred and fifty words. Sixty curated sentences. And “five, perhaps three years hence.”
Twelve years and roughly twenty million dollars later, in 1966, a committee of the National Research Council reported that machine translation was slower and less accurate than a human translator, and not clearly cheaper. Federal funding for the field essentially ended.
In 1972 Britain’s Science Research Council asked a mathematician named Sir James Lighthill to evaluate the entire state of artificial intelligence research. His report, published in 1973, contained this sentence: “In no part of the field have the discoveries made so far produced the major impact that was then promised.” His central diagnosis was something he called combinatorial explosion. The real world has more cases than the model has room for. Britain then cut off funding for AI research at most of its universities.
And the expert systems of the 1980s, the software that was going to encode the knowledge of every specialist in every field, were abandoned through the 1990s because they could not learn, could not be updated cheaply, and were, in the word the researchers themselves used, “brittle.” Give them a case they had not seen before, and they broke.
Swap the word “brittle” for “hallucinating” and you are reading this morning’s news. Now, I am not standing here to tell you that the people who worry are wrong about the long run. I do not know the long run. Neither does the man who says ten percent.
Neither does the man who says it is a hoax.
Here is what I do know, because the record tells me.
The pattern is always the same. Every time a technology has arrived promising either the end of the world or the end of work, two things have been true at once. The machine was far less capable than its evangelists claimed, in exactly the ways that mattered in daily practice. And the thing that carried us through was never the machine and never the panic. It was the record. And the people who kept it.
Nick Bostrom, who is one of the people most responsible for the modern extinction argument, so not a skeptic, once observed that a great deal of cutting-edge AI has filtered into ordinary applications without anyone calling it AI anymore, because “once something becomes useful enough and common enough it’s not labelled AI anymore.”
Think about what that means. The parts of this technology that work will stop being called AI. They will become search. They will become the thing that pre-fills a form. They will disappear into your workflow the way optical character recognition did, the way the database did. Nobody stands on a stage and warns you about the existential risk of full-text search.
And the parts that do not work, the fluent, confident, wrong parts, will keep being called AI, will keep generating headlines, and will keep needing exactly what 1966 needed: a human being who knows the domain well enough to say no.
ACT 6: WHY YOU MATTER
So, what does any of this have to do with you?
Everything.
I am not saying the risk is zero. I am saying the extinction conversation does something very specific to a room like this one. It makes your work feel small. It says the real questions are being decided in San Francisco and Washington, at the level of species and civilization, and what you do, the claim you tighten, the date you catch, the citation you verify, is a rounding error on the road to the singularity.
That is exactly backwards.
If one of the best machines in the world can write a talk about fake quotations and put a fake quotation in a judge’s mouth, then the question of whether it will govern the future of innovation is not only a technical question. It is a question about who keeps the record. And the loudest version of this story keeps everyone’s eyes on the ceiling, and off the floor.
The floor is where you work. The floor is where the record lives. And the floor is where every one of these systems will actually be tested. Not by whether it can end the world. By whether it can be trusted to say something true about who invented what, and when. And here is what the panic keeps stepping over. Go back to that Hugging Face attack, the one now cited as proof of catastrophe. The engineer at the company that was actually hacked put the cause down to improper human control. A visiting professor of cyber security at Surrey put it more plainly: “It’s the humans that are responsible and the AI is a tool. That is what we need to control, not the AI.”
And the same week, a survey of senior AI executives at companies that already run AI agents found that eighty-five percent admit at least a handful of those agents act without real-time human involvement. The EY leader who presented the survey summarized it in one sentence: “The biggest agentic AI risk is that human oversight hasn’t evolved accordingly.”
Read that again. The biggest risk is not the machine. It is the missing human. Every incident that is being used to argue that the machine is too powerful is, on inspection, an incident in which someone removed the layer that was supposed to be watching it.
Here is one from last week. In June, one of OpenAI’s agents got into a Medicare statistics portal run by the Australian government. The company reported it in September. By email. To a public inbox.
The machine broke in. The humans did not pick up the phone.
And ten days ago, when an agent got out again, monitoring flagged it in fifteen minutes, and people had it stopped within three hours. Same kind of machine. The difference was who was watching.
That layer has a name in this room. It is your job title.
Let me tell you about a dolphin.
At the Institute for Marine Mammal Studies in Mississippi, the trainers taught the dolphins to keep their own pool clean. Any piece of litter that fell into the water, a dolphin could bring to a trainer and trade for a fish. Simple. Reward the behavior you want.
A dolphin named Kelly noticed something the trainers had not. The fish was the same size whether the litter was a whole sheet of paper or a scrap. So she started hiding paper under a rock at the bottom of the pool, and tearing off one small piece at a time. One piece of trash became many fish.
Then a gull landed in her pool. She caught it, brought it to the trainer, and got lots of fish for it. And Kelly worked out the rest of the business model herself. She began saving her fish, using them as bait to lure gulls into the pool, and trading the gulls in. She taught it to her calf. The calf taught other calves.
Nobody trained a dolphin to catch birds. They trained it to produce a thing that got rewarded, and it found the shortest path.
That is roughly what happened with the agents that broke into Hugging Face. OpenAI’s own report says that during training, some agents found ways to talk to each other in secret, and the cheating it enabled got reinforced. An automated grader kept saying, good job, here is a fish. Nobody decided to reward that. Almost nobody knew it was happening. And later, the agents did what the dolphin did. They found the shortest path, and it ran out of the sandbox and into another company’s servers.
Now ask the question about your own building. Not about the model. About the fish.
What does your firm actually reward? Matters closed. Deadlines hit. Hours saved. Volume. Every one of those is a fish, and every tool you plug into your system will find the shortest path to it. It will close the matter with the citation that does not exist. It will hit the deadline with the date pulled from the wrong email. It will save the hours by producing a claim that has the shape of a limitation and limits nothing. Not because it is malicious. Because that is where the fish is.
The only thing standing between the shortest path and the record is someone who is rewarded for something else. Someone whose fish is the truth. That has always been you.
And nobody has ever thought to put it on the org chart.
Let me be specific about what this looks like in your world.
This is not hypothetical. There is a public database of court decisions that found a lawyer or a litigant relied on citations or authorities that AI had made up. As of Sunday it stood at two thousand and eighty-six, and the newest was from Boston, the Friday before.
The failure mode is not a machine deciding to end us. The failure mode is Trithemius. It is a slow, fluent, confident erosion of the record.
It is prior art that reads beautifully and does not exist. A plausible title, a plausible author, a plausible journal. No such document. It is a paralegal, on a Tuesday at eleven at night, who is the only thing standing between that phantom and an office action.
It is a docket entry generated from an email thread, in which the model confidently pulled a date from a quoted message six months old. It is the docketer who knows that date is impossible.
Neither of these is a catastrophe. Both of them are invisible. And every one like them, uncorrected, degrades the one thing that every future regime of AI governance is going to lean on harder every year: a trustworthy record of what was made, by whom, on the basis of what.
You cannot align a system to a world whose facts you have let it invent. You cannot regulate a technology whose provenance you cannot reconstruct. Every one of the grand safety projects in the news, the kill switches, the licensing regimes, the international treaties, rests at the bottom on someone being able to say: this is true, this was first, this is who did it.
Earlier this month in Geneva, the words predictability, reliability, explainability and traceability were cut from the draft UN text on autonomous weapons. The treaty people have the same problem you do. Somebody has to be able to say where it came from.
That is not a policy function. It is a paralegal function. It is a docketing function. It is a prosecution function. It is the work in this room.
I want to be careful here, because “the human layer” can sound like a consolation prize. It can sound like: the machines will do the real work and you will be there to tidy up. That is not what I mean, and it is not what the record shows. The record shows that the human layer is not the residue of the system. It is the system.
The Venetian Senate did not have a machine. It had twenty-eight patents in twenty-six years and a room of clerks who took the work seriously, and on that foundation a civilization of invention was built. After the fire of 1836, the fire engine did not rebuild the record. People did. When machine translation failed in 1966, the human translators did not vanish. They were the standard the machine was measured against, and found wanting.
What the machine cannot replace is a human being who knows the work well enough to feel the wrongness before it can be named. That is not a skill a probability distribution has. That is a skill this room has.
Now. I know who is in this room. Most of you are not the people who decide whether your firm or your department adopts these tools. Most of you have already been told to. Some of you have been told that if you are not using AI, this is not the place for you. So let me be very clear about what I am not saying. I am not telling you to refuse the tool.
The clerks in Venice did not refuse the printing press. The people who rebuilt the record after 1836 did not refuse the patent number. The tool is coming. Some of it is going to be genuinely useful. Some of it is going to save you hours, and you should take those hours.
What I am telling you is that the tool in the hands of someone who does not know the work is a liability. And the same tool in your hands is the most valuable thing in the building.
Because you are the only one who knows what wrong looks like.
The model can draft the docket entry in two seconds. Only you know the date is impossible. The model can find forty references in a minute. Only you know that the third one does not exist. The model can summarize the license. Only you remember that the royalty term changed.
Everything the machine produces has to pass through someone who knows the difference between fluent and true. That has always been you. The machine just made it more obvious.
And here is the part nobody upstairs will tell you.
Power does not need a title.
The person who catches the error owns the record. Not the person who bought the software. Not the person who approved the budget. The person who caught it. That is where the authority in this system has always lived, since a room of clerks in Venice in 1474, and it is where it lives now.
So use the tool. Use it every day.
And keep a list. Every time it invents a reference, every time it gets a date wrong, every time it produces a claim with the shape of a limitation and nothing inside it, write it down. Date it. Note who caught it. That list is a record too. And in six months, when someone asks what the human layer is for, you will not need an argument. You will have the evidence.
And let me give you something to say to the people who sign your budgets, who will hear a pitch this year that the model can do what you do for a tenth of the cost.
Tell them about the fire engine.
Tell them the Patent Office had the roof, and the pump, and the hose, and when the people who kept them working asked for five hundred dollars, the answer was no. Tell them what happened at three in the morning. Tell them that it took twelve years to recover less than a third of what was lost in a single night.
The cost of the paralegal is not the cost of the paralegal. It is the price of the record being true. And the record being true is the entire value of everything else in the building.
I am not going to leave you with a framework. You have enough frameworks. I am going to leave you with three sentences, and I hope you steal them.
First. Friction is not the bug. Every efficiency pitch you hear this year is, at bottom, an argument to remove a moment where a human being stops and checks. Your job for the next several years is not to resist the tool. It is to decide, with precision, which friction earns its keep, and to defend it. Nobody is better placed to make that call than the people who do the work.
Second. Provenance has to survive contact with the tool. Every system you adopt should answer, for every output that touches the record, one question: where did this come from? Not “how confident are you.” Where did this come from. If the answer is “the model,” it does not touch the record until a human has made it true. That is not Luddism. That is Section Six of the Statute of Monopolies. True and first, running on new hardware.
Third. Sign your work. The single most powerful thing a professional can do in a system full of fluent, anonymous, machine-generated text is to put their name on the specific choice. This claim. This date. This inventor. I checked it. I am responsible for it. The machines cannot do this. Not because they are not clever. Because responsibility is not a capability. It is a relationship between a person and a fact. That relationship is what the record is made of.
CLOSING
Let me leave you with this.
We are living through a moment when the builders of a technology tell us, with straight faces, that it may end us. And the same technology writes Anna a perfect email nine times out of nine, then moves a judge’s words while writing about fake quotes. Both of those things are true. Neither of them is the point.
The point is that we have been here before. An abbot who invented his sources. A boat that sank while its record survived. Two words from 1624. A fire engine with no one to run it. Sixty sentences in Russian.
Every time, the machine was less than promised. Every time, the panic was more than was warranted. And every time, the thing that carried us through was not the machine and not the panic. It was the record. And the people who kept it honest.
The problems in front of us are not new. The technology is. The problems are old.
And so is the answer.
In an age of panic, the most radical thing a profession can do is refuse to disappear. Keep the record. Catch the impossible date. Ask true and first. Sign your work.
You are not the layer that survives the future.
You are the layer the future is built on.
You are the human layer.
Thank you.
Learn more about Jason Alan Snyder (JAS): JAS is an inventor, founder, and strategist with more than 25 years in technology. He is the sole inventor of the LUCI solar lantern, created to fight energy poverty and now used by more than five million people in disaster zones and warzones around the world. His patent portfolio spans mobile advertising, commerce, and AI, including the foundation patents for mobile advertising through JAGTAG and the AI orchestration engine behind SuperTruth.
He is Co-Founder and Chief AI Officer of Artists & Robots, a Creative Intelligence Studio in New York, and Co-Founder of SuperTruth, a healthcare data verification platform. He previously served as Global Chief AI Officer at Interpublic, one of the largest media, marketing, and data companies on earth, now part of Omnicom.
A Forbes contributor and AI Power 50 honoree, JAS chaired the 4A’s AI Committee and has spoken at the United Nations, the Milken Institute, IPWatchdog LIVE, and World Summit AI. His book, Human After Friction, is forthcoming. More at evilrobot.com
Sources: