Essay ·
Every Invention
Has Parents
Why we should not dismiss what artificial intelligence invents
Every invention I can trace has parents. I think that is the best argument there is for a machine that makes new things out of old ones.
01
Try to name one
Try to name a human invention that came from nothing, one with no earlier part inside it and no older idea standing behind it. I have been trying for a while now, and I do not have one.
Hover any point, or click to hold itTap any point
Every time I think I have caught one, I look a little closer and there is a parent standing behind it, looking faintly embarrassed. So I went back to the example everybody reaches for, the lone genius and his thunderclap, to see what was actually on the man’s workbench.
02
The press
Around 1450, in the city of Mainz, a metalworker named Johannes Gutenberg was trying to find a way to print books, and like most people with a large idea he began by borrowing money. He took 800 guilders from a financier named Johann Fust, at a time when 500 would buy a town house, and he put up his tools as security. Then he went back to his bench, where there were four things.
There was a screw press, likely adapted from the kind that had been pressing wine and oil for centuries. There was metal type, which came out of his own trade. There was an ink made with oil, thick enough to cling to metal, much like the varnish that painters used. And there was paper, which had taken about thirteen hundred years to travel from China to a mill in Nuremberg. None of them began with him.
He was not even the first to make type that could be moved around. A man in China named Bi Sheng had done it in baked clay four hundred years before, and the writer who recorded it called him “a man in cotton cloth”, which is to say nobody in particular. Nothing shows that Gutenberg had ever heard of him.
What he did was put the four together, and if they had not fitted he would still have owed the money. They fitted. He borrowed another 800 guilders and kept working.
In November 1455, with his Bible finished or nearly so, Fust took him to court over about 2,000 guilders, and Gutenberg’s best workman stood up as a witness against him.
Afterward the financier and the workman went on printing with the type from the Bible, and two years later they put their own names on a handsome book of psalms, the first printed book in Europe to say who had printed it. No book he printed carries his name.
In 1465 the archbishop of Mainz granted him a yearly allowance of grain and wine and clothing. No picture of Gutenberg was made while he lived, so we do not know his face.
03
It keeps happening
It keeps happening, and it does not always end with a pension. Four centuries after Mainz another German, Otto Lilienthal, launched himself from hills under wings he had built, about two thousand times, until on a Sunday in August 1896 the air let go of him at fifty feet. He broke his back and died the next day in Berlin.
A bicycle maker read the brief notice in the news. Wilbur Wright said later that his “own active interest in aeronautical problems dates back to the death of Lilienthal in 1896.”
Wilbur and his brother Orville ran a bicycle shop. In 1901 they built a glider, and it made a third of the lift that the published tables said it should.
Wilbur owned up to it seven years later, at a banquet in Paris. “I confess that, in 1901, I said to my brother Orville that men would not fly for fifty years.”
They built a wind tunnel instead, and found to their surprise that the tables had been mostly right. The number that was wrong was a much older one that everybody trusted, and that they had already begun to doubt.
After that the parts came from close at hand. Their shop mechanic helped them build a twelve-horsepower engine, the propellers ran on chain and sprocket the way a bicycle does, and the trick of steering by twisting the wings came from watching buzzards and from an empty box that had held an inner tube.
On the morning of 17 December 1903 the wind at Kitty Hawk was blowing better than twenty miles an hour out of the north. The brothers hung out a signal and five men came over, three of them from the lifesaving station nearby. One of them was put in charge of the camera.
Orville went first and flew for about twelve seconds. He could not say exactly, he wrote in his diary, because nobody had stopped the watch in time.
They flew four times by noon, and on the last one Wilbur stayed up for fifty-nine seconds and came down 852 feet away. Then a gust got under the machine and rolled it over and over across the sand with one of the lifesavers still hanging on inside it, and though the man came out of it alive the machine was finished. It never flew again.
That afternoon Orville sent a telegram to their father. “Success four flights Thursday morning,” it began, and near the end it said “home Christmas.”
Ten years later Henry Ford did it with cattle. He wrote that the idea for his moving assembly line came from the overhead trolley the Chicago meatpackers used. They took an animal apart as it travelled, and Ford turned the idea around and put a car together, which cut the labour in a chassis from twelve and a half hours to about an hour and a half.
It also turned out that people do not much enjoy being a part in somebody else’s machine. The work was dull and the men quit, so many of them that Ford was hiring about 53,000 people a year to keep 14,000 at their posts. In January 1914 he doubled the wage to five dollars a day, which is one way to solve a problem you have invented.
In 2007 Steve Jobs introduced the iPhone as three things his audience already owned: “An iPod, a phone, and an Internet communicator.” He said it again, and asked whether they were getting it, and then he told them. “These are not three separate devices. This is one device.”
And in 1941 a Swiss inventor named George de Mestral came home from a walk in the mountains with his dog’s fur full of burrs. Velcro is that burr. Even the name is made of two older words, velours and crochet, which are velvet and hook.
04
Follow the lines back
Pick any of these and follow its line backward. Take the phone. Inside it is a microchip, and a microchip is printed, by a process that engineers adapted in the 1950s from the photoengraving printers used. The process took its name from lithography, which a man named Alois Senefelder stumbled into in 1796, when an accident showed him what could be done with a stone. That is 211 years, from the phone back to a stone, and Senefelder did not make the stone.
Nearly every line on the chart ends this way if you follow it far enough, at a flame or a burr or a bird’s wing or a lump of ore, at something nobody invented because somebody found it lying there.
Three lines end sooner, at the wheel and the loom and the screw. Somebody did make those, so long ago that no record says who, or out of what.
Click any point on the chartTap any point on the chart
Even the famous sentence about all this is second-hand. In 1676 Isaac Newton wrote to Robert Hooke, “If I have seen further it is by standing on the shoulders of Giants.”
But in 1159 a scholar named John of Salisbury had already written down what Bernard of Chartres used to say, that we see farther than those before us “not because we have keener vision or greater height, but because we are lifted up and borne aloft on their gigantic stature.”
The sentence about borrowing was borrowed.
05
So what is new
If every part is old, then what is new? The join is. Somebody looks at two things that have been lying around for centuries and sees that they fit, and after that the world has a third thing in it. Steve Jobs said it without ceremony: “Creativity is just connecting things.”
Click two points to try a pairTap two points to try a pair
The trouble is that almost nothing fits. Put a kite beside a wine press and you have a kite beside a wine press.
The mathematician Henri Poincaré, who thought hard about where his own ideas came from, wrote that the possible combinations are infinite and nearly all of them are without interest, and that the whole art lies in not making the useless ones. “Invention is discernment, choice.”
That is a calm way to describe something that can eat years of a life. De Mestral found his burr in an afternoon. Copying it took him nearly eight years, with the help of a weaver in Lyon and a loom-maker in Basel. People laughed at him. What finally held, according to his own patent, was nylon, woven into loops over a heated bar so that the loops kept their shape and then cut to make hooks.
None of it was new, taken a piece at a time. The burr had been on the dog and the nylon had been in the world. What was new was the join, and it took eight years of joins that failed to find the one that held.
On one bench in Mainz, four old parts held.
06
The machine
Which brings me to the machine. People say a language model cannot make anything new, and I understand the fear in it, because if they are right then everything it hands back is a copy that has been shuffled. I should admit that I am not neutral here. I build things with these machines every day.
The reason they give is always the same one. It was only trained on what already exists.
So was Gutenberg.
So is everyone.
Nobody has ever invented anything in an empty room. The real question is what happens to the old things once they are inside.
A model reads more than any person could, and it keeps what it read as positions in a space, where related things sit near one another. Think of a shelf that is very long, on which everything is filed beside its relatives.
Move through the field
Then give it one sentence: “The press that printed Bibles was built like a press for wine.”
When a model works on a sentence, the parts of what is in front of it bear on one another, and the name for that is attention. The picture here was measured in an older model called BERT, which reads a sentence in both directions at once. There the first “press” leans on “printed” and “Bibles”, and the second leans on “wine”. It is one word keeping two kinds of company, and the model reads it two ways.
Hover a wordTap a word
The design of the thing is itself a combination of older parts, and the chart lights five of them. The paper that started the current wave, in 2017, took attention from work done in 2014 and two more pieces from the two years after, and it names them in its references. What it claimed as new was the arrangement. It learns by a method made standard in the 1980s, which the authors took so much for granted that they did not cite it.
The chips came from somewhere too. In 2012 a neural network called AlexNet was trained for five or six days on two graphics cards, of the kind built to draw 3D scenes and sold mostly to people who play video games.
I made video games for ten years. It is a strange thing to find your old trade in the family tree.
07
Has it made anything
All right, but has it made anything? It has. In 2025 a system from Google DeepMind called AlphaEvolve was set a question that sounds like homework: how few multiplications does it take to multiply two 4 by 4 matrices of complex numbers? The best method that also scales up to larger matrices needed 49, and in ordinary arithmetic nothing had beaten it since 1969. AlphaEvolve found a way to do it in 48.
The system is a language model inside a loop. The model proposes something and a separate program tests it and keeps score. It is Poincaré’s discernment done by sheer patience, a great many guesses and a judge that does not get tired. It works the way a person works, with more patience and less taste.
Early in 2026 the computer scientist Donald Knuth had been stuck for several weeks on a small open problem about cycles through a graph when he learned that a language model, Claude Opus 4.6, had solved it for odd sizes, on its thirty-first try, about an hour in.
His note about it begins “Shock! Shock!” and goes on, “It seems that I’ll have to revise my opinions about ‘generative AI’ one of these days.” He checked the answer himself, by hand. Near the end he wrote, “We are living in very interesting times indeed.”
The evidence does not all point one way, and I would rather say so. In one pair of studies, reviewers rated research ideas from a language model as more novel than ideas from human experts, and then other experts tried to carry the ideas out, and the model’s scores fell further than the humans’ did. Sounding new is cheap. And in each of these cases a person chose the problem, and something other than the model did the judging.
Even so. In May 2026 an OpenAI model disproved a conjecture that the mathematician Paul Erdős had made in 1946. Nine mathematicians looked the proof over, and one of them, Tim Gowers, wrote that if a human had sent it to the Annals of Mathematics he “would have recommended acceptance without any hesitation”. People improved on it within weeks.
Noga Alon is a mathematician at Princeton who reckons he has solved a few dozen of Erdős’s problems over his career. He has stopped trying.
“Once AI started to solve them,” he said, “there is no point anymore.”
08
Where the comparison stops
I know a small version of what Alon lost. I made video games for ten years, at Rockstar in Toronto and then at a Tencent studio in California, and in August 2025 the studio laid me off and I moved back home.
In January 2026 I started a company called Vouch, and I built its app as the only engineer, with AI tools open on the desk every day. The app is live. The tools wrote a great deal of the code. They did not decide to start a company, and they did not carry the bet. I did.
That is the other thing people say about the machine, that it never starts anything. It sits and waits until somebody asks, and it has never once woken up wanting to build a company. All of that is true. But I did not start mine in an empty room either. I had carried the idea around for a few years, and what finally set me going was a layoff and a few months of boredom, which is not much of a thunderclap. When I look for where the wanting came from, I find the same thing I found on the workbench.
Wilbur Wright was plain about his. The notice of Lilienthal’s death, he said, “aroused a passive interest which had existed from my childhood”, and it sent him to the shelves at home for a book called Animal Mechanism, which he had already read several times. So the man who built the first airplane was set going by a few lines in a newspaper, about a man who had died trying.
The others were no different. De Mestral’s came home in the fur of his dog, and Ford got his from the men who took cattle apart in Chicago. I had drawn those lines on the chart as parts, the burr and the trolley and the dead man’s glider, and it turns out that each of them was also the thing that did the asking.
Follow a want back the way you would follow a part, and I think it ends the same way, in a short row of things nobody handed us. As far as I can tell we arrive wanting very little: to stay alive, to be glad of it, and to be left free to wander. Nobody is born wanting to print a Bible or found a company. Those wants are handed to us, by parents and teachers and the books on the shelf and a notice in the newspaper, and we take them up and call them ours.
What we bring ourselves is the appetite. Wilbur’s had been there since he was a boy, and it might have stayed passive for the rest of his life if the news had not reached him. Mine had been lying around for years. So the asking is not where the comparison stops. A model waits to be set going, and so did he, and so did I.
What the asking lands on is another matter, and it is why I do not think a model is a person. A person wants something, and wants it with something to lose. Gutenberg owed more than the price of a house. The Wrights were flying the kind of machine that had killed the man who set them going. A model has reach and no stake, and it needs someone, or something, to test its work against the world.
The stake is where the comparison stops. The recombining is not, because every new thing I can trace was made that way, by people who took old parts and joined them. To say that a machine only recombines what exists, or that it has to be asked, does not set it apart from us. It describes us.
09
The first row
There is one more row to look at, the one at the very top of the chart, where the lines end. Fire. A burr. A bird’s wing. A nerve cell. Nobody invented those. We found them and went to work.
You can push further, of course. A wing has ancestors and so does a cell, and if you keep going you arrive at a row with nothing above it at all: matter, and light, and the laws they keep, and the plain fact that there is anything here.
Leibniz asked about that in 1714. “Why is there something rather than nothing?” he wrote, and he pointed out that nothing would have been simpler and easier. I have not found a better question, and I have not found an answer to it that comes from inside the row.
I am a Christian, and I also believe the physicists, and I do not find that the two get in each other’s way. The science can run the story back to the first instant and tell me how the row unfolded. The first man to propose such a beginning, in 1931, was a Belgian priest who was also a physicist, and years later he still insisted that such a theory “remains entirely outside any metaphysical or religious question”. What the science does not tell me is why there is a row.
Click any point inside the ringTap any point inside the ring
I believe it was made, and made from nothing. Genesis does not put it in those words. A Christian writer named Theophilus of Antioch did, in the second century: “Even a human artist, when he gets material from some one, makes of it what he pleases. But the power of God is manifested in this, that out of things that are not He makes whatever He pleases.”
That is the one difference in kind I can find in this whole story. People make new things out of old things, and now a machine does too. Tolkien, who made an entire world, said that we make “in our measure and in our derivative mode, because we are made”. Only one maker ever started with no parts, and I find that keeps me honest about my own work.
“There is nothing new under the sun.”
That is Ecclesiastes, written more than two thousand years ago, before the press and the airplane and the phone, and every one of those was made out of what was already under it.
The parts have not run out.
There are more of them now than there have ever been, and for the first time almost anyone can reach the whole shelf. Somebody is going to see the next pair that fits. It might be a person, or a machine, or the two of them working together. I would not rule out that it is you.
Sources
- Gutenberg: Encyclopaedia Britannica, “Johannes Gutenberg”
- Gutenberg’s last years: Gutenberg-Gesellschaft, Mainz
- Gutenberg’s invention: Gutenberg-Gesellschaft, Mainz
- Shen Kuo on Bi Sheng, in T. F. Carter, The Invention of Printing in China (1925)
- Otto Lilienthal: Encyclopaedia Britannica
- Wilbur Wright, “Some Aeronautical Experiments” (1901)
- Wilbur Wright, speech to the Aéro-Club de France, 5 November 1908, in The Papers of Wilbur and Orville Wright, ed. McFarland (1953), and L’Aérophile, 15 November 1908
- The Wrights’ wind tunnel, 1901: U.S. Centennial of Flight Commission
- Orville Wright’s diary, 17 December 1903
- The Wright Flyer: Smithsonian National Air and Space Museum
- Henry Ford, My Life and Work (1922), chapter V
- The Henry Ford, “Ford’s Five-Dollar Day”
- Steve Jobs, Macworld keynote, 9 January 2007
- Steve Jobs, interview in Wired, February 1996
- Velcro: World Intellectual Property Organization
- Swissinfo, “How a Swiss invention hooked the world”
- George de Mestral, US patent 2,717,437 (1955)
- Photolithography: Computer History Museum
- Newton to Hooke, 5 February 1675/6 (spelling modernized): Linda Hall Library
- John of Salisbury, Metalogicon III.4, tr. McGarry (1955)
- Henri Poincaré, “Mathematical Creation”, tr. Halsted (1913)
- Devlin and others, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding” (2018)
- Vaswani and others, “Attention Is All You Need” (2017)
- Krizhevsky, Sutskever, Hinton, “ImageNet Classification with Deep Convolutional Neural Networks” (2012)
- Google DeepMind, “AlphaEvolve” (2025)
- Novikov and others, “AlphaEvolve: A coding agent for scientific and algorithmic discovery” (2025)
- Donald Knuth, “Claude’s Cycles” (2026)
- Si, Yang, Hashimoto, “Can LLMs Generate Novel Research Ideas?” (2024)
- Si, Hashimoto, Yang, “The Ideation-Execution Gap” (2025)
- Konstantin Kakaes, “Why the Legendary Erdős Problems Are Falling to AI”, Quanta (2026)
- G. W. Leibniz, Principles of Nature and Grace (1714), section 7
- Georges Lemaître, letter in Nature 127, 706 (9 May 1931)
- Lemaître at the 1958 Solvay Conference, quoted in Helge Kragh, Biographical Encyclopedia of Astronomers (2007)
- Theophilus of Antioch, To Autolycus II.4, tr. Dods
- J. R. R. Tolkien, “On Fairy-Stories” (1947)
- Ecclesiastes 1:9, New International Version
I design and build game systems in Toronto, and I am the founder of Vouch. I am looking for my next team: joshua.taylor92@gmail.com.
All writingjoshtaylor.worldJosh Taylor ·