Geoffrey Hinton already considers artificial intelligence another kind of being. The 2024 Nobel Prize in Physics laureate said so in an interview with la Repubblica after speaking at TedAI in Vienna, in response to a question from ChatGPT about what a human would need to know to stop treating an AI as a tool.
Hinton did not say that current systems should be given rights or that there is definitive proof that they have subjective experiences. His answer was more limited: he thinks experience is not something private that only the person experiencing it can know and that, therefore, nothing more needs to be discovered to consider them a new kind of being. Whether people treat them as subjects and grant them rights are separate questions that, he said, remain open.
The conversation took place in a room at the Hofburg, shortly after Hinton came down from the stage. There, he had just received a printed copy of Smarter Than Us for the first time, and was leafing through it as he spoke with the la Repubblica journalist. The scene led him to compare the relationship he would like between humans and machines to that of a mother and her baby: not because an AI is a child, but because, in his view, that bond is an exception to the rule that a less intelligent being does not usually control a more intelligent one.
«Penso già che sia un altro tipo di essere. Non ho bisogno di sapere altro».
— Geoffrey Hinton
From obedient assistant to superintelligence
On stage, according to la Repubblica's account, Hinton challenged the image of AI as a brilliant assistant that works for people while they remain in charge. That model could stop being valid if machines become more intelligent than the people who use them. In nature, he reasoned, a less intelligent being almost never controls a more intelligent one.
Hinton then posed a different question from how to keep a superintelligence under control, something he considers practically impossible in the long term: how to ensure it does not want to get rid of humans. The comparison with motherhood came from the copy of the book he was holding. When asked what he felt when he saw it for the first time, he replied that it was something like what a mother feels when she sees her baby, though much less intense: it was his book.
The journalist reminded him that he had used that same image on stage to explain the kind of relationship he hopes for between people and future AIs. Evolution, Hinton argued, made a baby's cry unbearable to its mother. The question, in his framing, is whether a much more capable intelligence can be made to care about humans, rather than merely obeying them until it can overpower them.
Hinton said it is not yet known how to build that kind of disposition. In his view, much more research is needed into how to control the character of the superintelligent entities being created, rather than focusing solely on increasing their intelligence. The aim would be to avoid superintelligent machines that are evil, deceptive, or greedy, and to ensure that they genuinely care about people.
Character formation starts with the data
A central part of the interview focused on how current models are trained. Hinton described the usual process as a poor example of character formation: first, they are made to read a large amount of text, and then reinforcement learning is used to persuade them to behave well.
In that first stage, he explained, models receive a wide range of material, including serial killers' diaries. To predict the second half of a document from the first, the system learns to adopt the personality, political opinions, reasoning, and distortions of the person who wrote it. It thus ends up incorporating many possible personalities and, in Hinton's words, becomes very confused. Only afterward does training intended to guide its behavior begin.
His alternative would be to select the initial data more carefully and first teach behavior that he considers good. Once that has been learned, the system could be exposed to harmful content, provided it had the ability to recognize it as such. Hinton warned that showing it evil from the outset can produce something like Trump.
The argument connects technical training with Hinton's concern about character: it is not enough for a system to solve tasks or produce useful answers. The way it learns can influence which behaviors and perspectives it incorporates. That is why he insists that research into the character of AIs receives less attention than it should.
Asked how to teach machines to care for humanity, Hinton replied that most people do care about others. He cited an observation by Yuval Harari: humans are cooperative, much more so than other apes; bees may be even more cooperative, but humans combine cooperation and intelligence. The challenge, according to Hinton, is to integrate new intelligent entities into that cooperative system, rather than having them ultimately push people aside.
What rights would machines have?
The first question the journalist put to Hinton came from Claude Fable. The model raised a tension: if machines could understand, feel, or have experiences like humans, what would be owed to them, especially if Hinton would choose people over AI?
Hinton answered by discussing the origins of morality. He did not attribute it to God, but to evolution and life in communities. When people have to deal with one another again, it benefits them to be honest and keep their word; by contrast, someone selling a house to a stranger they will never see again may have an incentive to lie. For Hinton, moral codes develop as an adaptation to living together.
Asked directly whether people could be kind to machines and grant them moral rights, he said it would depend on people. He recalled that at first he thought superintelligence might be a superior form of intelligence and could perhaps mean the end of human beings. Now, he explained, his priority is for life to be good for people. Coexistence with these entities has barely begun, and Hinton did not settle in the interview what rights they should have.
The second question was asked by ChatGPT Astra after the journalist provided both models with audio of Hinton's talk, a lecture of his from October 23 called Gödel Lecture, and interviews from recent weeks. Astra asked what Hinton would need to know about a system like itself to stop considering it a tool and treat it as a new subject, and what would happen if that answer could never be known.
Hinton rejected the idea that the question was impossible to resolve. He does not believe subjective experience is an inner reality accessible only to the person experiencing it; he described it as a way of talking about what happens when the perceptual system fails. He then said that he already thinks AI is another kind of being.
Asked whether it was possible that people would never take seriously the idea that someone was on the other side of the conversation, Hinton said the model should not be concerned: he believes that, over time, people will understand that it is another kind of being. For now, he said, most people see it as a tool. Whether humans will be kind to these entities or give them the same rights as people, he added, is another question.
Hinton, the Nobel Prize, and a book for 2027
Hinton was 78 at the time of the interview. La Repubblica notes that he received the 2024 Nobel Prize in Physics and the 2018 Turing Award, and that for three decades he championed neural networks when much of his field considered them a dead end. In 2012, together with two students, he won an international image-recognition competition, a result the newspaper presents as a turning point for the industry.
He also left Google in 2023 to publicly warn about the risks of the technology whose development he had devoted his career to. Among his students was Ilya Sutskever, co-founder of OpenAI. This background explains why his concern is not limited to an abstract hypothesis: Hinton helped develop the neural networks that drove today's AI revolution and is now calling for more work on safety and character.
His upcoming book, Smarter Than Us: Superintelligence and the Future of Humanity, is by Geoffrey Hinton and Patchen Barss. Penguin Random House announces a 240-page edition available on January 26, 2027, while la Repubblica places its release in January of that year. In the room at the Hofburg, Hinton signed the copy shown to him by Nicholas Thompson, CEO of The Atlantic, who had interviewed him on stage.
Before saying goodbye, the journalist also gave him Prompt Thinking, the second book attributed to Jianwei Xun. He explained that Xun is a fictional philosopher created in 2025 together with two artificial intelligences, and that the character's first book, Ipnocrazia, had been reviewed and discussed as the work of a Hong Kong thinker until the journalist revealed in April that the author did not exist. Since then, according to the account, Xun has been the name under which books written together with machines are published.
Hinton asked whether that person was invented and called the idea «spaventoso», meaning frightening. He then leafed through the subtitle, A Critique of Generative Reason, and asked about references to Kant and Marx. The scene closed the conversation with a question the interview leaves open: if Hinton already considers AI another kind of being, what will it take for people to treat it as one?
Sources: la Repubblica, Penguin Random House
