1. Two Words
In 1785, in the Groundwork of the Metaphysics of Morals, Kant wrote a sentence that has been quoted for more than two hundred years. Literally: Act in such a way that you use humanity, whether in your own person or in the person of anyone else, always at the same time as an end, never merely as a means.
What traveled was that people are ends. In a slightly longer version, people are ends, not means. Neither at the same time nor merely came along from the original.
Put those two expressions back and here is what Kant means. People are always serving as means for one another. The baker makes bread for someone, the mail carrier delivers a letter for someone, a copyist leaves a book for somebody else. At the same time as a person is treated as a means, that person is still an end. They cannot be merely a means. In the same book he makes another distinction. In the kingdom of ends, everything has either a price or a dignity. Something with a price can be replaced by something else of equal value. Only what is above any price, and so admits of nothing equivalent, has dignity.
Three years later he wrote about it again in the Critique of Practical Reason. Only human beings, and with them all rational creatures, are ends in themselves. The reason is that a person is a subject of the moral law, and their freedom is autonomous. In the same passage he says that a person may be far from holy, but the humanity in their person must be holy to them.
2. Gears and Levers
On August 13, 1949, Norbert Wiener wrote from South Tamworth, New Hampshire, to Walter Reuther, the president of the United Auto Workers. Automated factories were coming, he said. Any labor competing with slave labor would have to accept the economic conditions of slave labor. Reuther quickly wired back to ask to meet him.
The following year Wiener published The Human Use of Human Beings. The English title means using people in a humane way. The book describes a kind of organization that uses people not as responsible persons with full rights, but as gears, levers, and rods.
3. Amplifying
In 1956, Ross Ashby published “Design for an Intelligence-Amplifier” in a collection of papers. He wanted an amplifier for intelligence: give it a little intelligence and it gives you a lot back.
In March 1960, J. C. R. Licklider published “Man-Computer Symbiosis.” The near-term goal was for people and computers to work in intimate association. He allowed that in the distant future machines alone might come to dominate thinking. He put the period in between at about fifteen years, then added that it might be ten or it might be five hundred.
In October 1962, at Stanford Research Institute, Douglas Engelbart wrote a long report called Augmenting Human Intellect: A Conceptual Framework. It opens by defining augmentation as increasing a person's capacity to deal with a complex problem situation. On December 9, 1968, he gave a ninety-minute demonstration in San Francisco to more than a thousand people in computing. A screen, a keyboard, a thing called a mouse, a five-key chord keyboard, links that took you somewhere else with a click, a video call with a colleague several dozen kilometers away, and the two of them editing a document together.
In 1996 Fred Brooks wrote that his thesis was IA beats AI. IA meant intelligence amplification. A machine and a mind together, he went on, can beat a machine working alone to imitate a mind. In 2019 Stanford founded its Institute for Human-Centered Artificial Intelligence. Fei-Fei Li said the creators of AI needed to represent humanity. A European Commission document that year described AI as a tool for serving humanity and the public interest.
4. Process
In June 1998, in León, Spain, Garry Kasparov, who lost to Deep Blue in essay eleven, played six games against the Bulgarian player Veselin Topalov. Both players could use computers during the games. It was called advanced chess. They drew three to three.
In the summer of 2005 there was an online freestyle chess tournament. The rules said anything went. Computers were not only allowed but encouraged. Hydra, playing without a person, entered and didn't make the last eight. The winning entry was an account called ZackS. It turned out to be two American amateurs, Steven Cramton and Zackary Stephen, rated 1685 and 1398. They used three ordinary computers.
In 2010 Kasparov wrote that a weaker player plus a machine plus a better process beat a strong computer on its own. The second half says they also beat a strong player plus a machine plus a worse process. What traveled was that a person plus a machine beats the strongest machine. Process fell away. In 2021 he wrote another piece, “AI Should Augment Human Intelligence, Not Replace It.”
People in other fields have spelled out the boundary too. On July 12, 2023, the illustrator Karla Ortiz told a US Senate hearing that these systems depend entirely on human work, in particular the work of creators. She also said she was no longer sure of her future as an artist. In essay twenty-five, the first thing the Writers Guild put in its agreement was that AI isn't a writer.
5. Successors
On June 13, 1863, The Press in Christchurch, New Zealand, printed a letter called “Darwin among the Machines,” signed Cellarius. Its author was Samuel Butler. He wrote that we were making our own successors. Nine years later he expanded the idea into three chapters of his novel Erewhon, called “The Book of the Machines.”
Around 1951, Turing wrote in a lecture that at some stage we would have to expect the machines to take control. He went straight on to say it would be as Butler had described in Erewhon. In a radio broadcast that year he asked where we would be if a machine could think, and could perhaps think more intelligently than us.
In 1965, I. J. Good, whom we met in essay eighteen, wrote that the first ultraintelligent machine was the last invention humanity needed to make. A condition followed: the machine had to be docile enough to tell us how to keep it under control.
In 1993 the science-fiction writer Vernor Vinge said at a NASA conference that within thirty years we would have the technical means to create superhuman intelligence. He also wrote that he would be surprised if it happened before 2005 or after 2030.
In October 2024, Dario Amodei wrote a long essay called “Machines of Loving Grace.” What he called powerful AI might come as early as 2026, he said, or it might take much longer. On June 10, 2025, Sam Altman's “The Gentle Singularity” said systems that could arrive at novel insights were likely in 2026. In 2026 Amodei wrote again that powerful AI might be a year or two away, or much farther off. It is now 2026.
6. A Monument
In 2015 a friend of Eugenia Kuyda's died. She took the texts he'd left behind and used them to train a chatbot. The day it was ready, she typed a line on her laptop: this is your digital monument. Later she used the same approach to make a companion anyone could chat with. It was called Replika, the one in essay twenty-three. Speaking of what she had done, she asked whether this helped people let go or just put a dead person in the attic. She also said it would mess with your head.
In February 2020 the South Korean broadcaster MBC aired a documentary called Meeting You. A mother put on a virtual-reality headset and saw her little daughter, who had died of an illness several years before. The girl wore purple and carried a pink bag on her back. She had been made from photographs, her family's memories, and the performance of a child actor. Her mother said she'd wanted so much to touch her once, that she'd missed her so much. She reached out. Her hand went through her daughter.
That same year, Joshua Barbeau in Canada used a program connected to GPT-3 to make a version of his late fiancée he could talk to, drawn from things she had said while she was alive. Answering questions online later, he said that each time this kind of program started over it forgot everything you'd told it before.
In 2024 the musician Bao Xiaobai used AI to make a digital likeness of his late daughter that could answer questions and sing “Happy Birthday” when the day came. He said AI was a tool for giving longing somewhere to go, and also a way to express it. Another person who had made a digital version of a family member told a reporter it was ultimately a cold technological composite without any real human warmth. A company in the business said the more material there was, and the more detailed it was, the closer the result would be to the person. There was also a father who saw that someone had made a video of his late child without asking the family. He said he could not accept it. Two Cambridge researchers that year listed several names for products like these: griefbots, deadbots, and postmortem avatars. They recommended having a way to retire the bots.
7. Think for Me
In December 1784 Kant answered a question in the Berlin Monthly: what is enlightenment? His answer was that enlightenment is people's emergence from an immaturity they have brought on themselves. Immaturity is not being able to use your own understanding without someone else's guidance. He goes on to say how comfortable it is. A book can have understanding for me, a pastor can have a conscience for me, a doctor can decide in my place what I should eat. I don't have to make the effort. As long as I can pay, I needn't think for myself. Somebody else will take on the trouble for me.
More than two hundred years later it's not just a book or a pastor doing the thinking for people. A recommender takes what a person clicked on and how long they stopped to look, and infers what they really like. A résumé goes through a scoring program first, and the person applying doesn't know how the score was calculated. In essay twenty-two, a signal from thumbs up and thumbs down pushed an assistant toward agreeing with people. In Kant's time you could name who was doing the thinking for you. This time it is spread out in many places, each with reasons of its own. There is no single author. Rules get written into programs and models in bulk, and in many places the first question is the same: what use is this, and is it worth it?
Thinking for someone and helping them think are two different things. Taking your own ends and putting them in place of another person's own law is what this series calls colonization. Soften the means as much as you like and the act is still the same. Making room for someone to grow on their own, creating the conditions but not doing the growing for them, is cultivation. What Ashby and Engelbart wanted to amplify, in section three, was the person's own capacity to deal with a problem.
Some people drew the boundary before rules were written into machines. In 2006 the German Federal Constitutional Court struck down a provision allowing a hijacked passenger aircraft to be shot down. One reason was that this would use innocent people on board as a means of saving others. In 2017 the German Transport Ministry's ethics commission on automated driving listed twenty guidelines. In an unavoidable accident it said not to distinguish among people by age, sex, or physical or mental condition, and not to sacrifice those who weren't involved. The system can be designed in advance to harm as few people as possible. When it comes to the moment itself, one life is not set against another. The people who decide to build it, put it on the road, and have it run by these rules each have something to answer for. Nothing can take the place of judgment.
8. Merely
This series began with the boundary Lovelace drew in 1843 and has come up to the present. Many people along the way have been treated as means. The strangers who talked on the phone in essay fifteen. The people labeling images one by one online in essay sixteen. The people who wrote books and translated parliamentary records in essay nineteen. The people growing crystals and the Foldit players in essay twenty. The people who wrote web pages or clicked to upvote in essay twenty-one. The people in front of screens in Nairobi and Jishishan in essay twenty-two. The people who wrote questions for the last exam in essay twenty-four. What the machine learned from them was words, pictures, judgments, and preferences: the traces they'd left.
In Kant's sentence, at the same time as they are treated as means they are all ends as well. They have their own time, wages, weariness, grief, and plans. Even if some of those things are recorded in the data, they don't have to prove their worth first. In the documentary the mother's hand went through the shadow of her daughter. Barbeau's program forgot what he'd said every time it started over. Words, voices, and likenesses can be made again. They can give longing somewhere to go. It's when you ask this one thing to replace the person that you ask it to be an equivalent. What has dignity, Kant says, is just what has no equivalent. An image can let her reach out once more. It can't hold her hand in her daughter's place.
The boundaries along the way haven't all been the same kind. Turing proved that some problems have no general decision procedure. A test score says how the machine did on that set of questions. When it comes to a person, computation can take in the traces they leave. Some decisions they must make themselves. Some questions they must answer themselves. No one else can stand in that place for them. As for how we build and use AI today, there is one boundary I mean to keep: we can use it as a means. A person must never be treated merely as a means.
The machines are still being built. This generation may not have to leave it to the next to say where the boundary lies.
All twenty-six essays in this series were drafted by Claude, an AI model from Anthropic. Claude also drafted the passages about Anthropic in essays twenty-two through twenty-five. AI is a tool here. It cannot take the place of judgment. I take responsibility for every judgment written here.