{"id":11305,"date":"2024-06-06T12:00:32","date_gmt":"2024-06-06T12:00:32","guid":{"rendered":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/"},"modified":"2024-10-17T13:43:32","modified_gmt":"2024-10-17T13:43:32","slug":"profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure","status":"publish","type":"post","link":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/","title":{"rendered":"How a reporter prepped to understand A.I. and the man who helped invent it"},"content":{"rendered":" \n
\n
\n \n

Journalists who write profiles don\u2019t go in cold. They pre-report to prepare for crucial interviews. They read widely and research previous stories that have been written about their subject. They think through questions they need to ask.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

To profile Geoffrey Hinton, the British-Canadian computer scientist considered the godfather of artificial intelligence, New Yorker writer Joshua Rothman did his homework, and then some. He read histories of A.I., consulted an oral history of neural networks \u2014 a machine-learning technique that teaches computers to process information like the human brain  \u2014 and plowed through a textbook about deep learning. He even took an online course in linear algebra before he felt ready to query Hinton.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

\u201cI\u2019d written about A.I. and machine learning before, but the idea of spending an extended period alone with Geoff, and asking him to explain his work \u2014 it made me feel like I really needed to be prepared,\u201d Rothman told me. \u201cI wanted to make the most of the time I had.\u201d<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

All that prep helped Rothman understand and then convey the science behind artificial intelligence, the technological development that has seized the public\u2019s attention ever since OpenAI released its chatbot, ChatGPT, in November 2022. Rothman\u2019s diligence grounded a 10,000-word nuanced, comprehensive and revealing portrait of Hinton, published in The New Yorker in November 2023.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

\u201c<\/strong>Why the Godfather of A.I. Fears What He\u2019s Built<\/strong><\/a>\u201d<\/strong> is an important story, deeply reported, creatively structured and written with literary grace. Rothman frames it within a four-day visit he made to Hinton\u2019s home on a private island on Ontario\u2019s Georgian Bay. The narrative glides through connections that track the history of A.I. and Hinton\u2019s career with the personal tragedies that have shaped him.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

\u201cOften, I think, we write about people because of something they\u2019re involved in, or because of something they\u2019ve done,\u201d Rothman said. \u201cBut when it comes time to write the profile, it becomes important to focus in on the person as a person \u2014 just as an unadorned human being, with all the richness and intensity that entails.\u201d Even the word \u201cprofile\u201d is telling; Rothman and Eric Overbey wrote in an introduction<\/a> to 12 classic New Yorker profiles: \u201cIt suggests catching sight of someone from an unusual angle.\u201d<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

The profile of Hinton opens with an examination of the brain at work and comes to a close when the natural world eclipses the technical one. In between, it explores a new and, to some, discomfiting frontier of knowledge and ethics. Rothman uses sharply-etched scenes, characterization, dialogue, metaphors and digressions to help readers grasp historical context and challenging technical details.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

Rothman, the Ideas Editor of The New Yorker, began his journalism career blogging for the ideas section of The Boston Globe and then freelancing before joining The New Yorker in 2012, where, as ideas editor of newyorker.com, he guides and writes stories about science, philosophy and technology.<\/strong> He helped conceive the Hinton profile for an issue devoted to artificial intelligence, pegged to recent concerns about the evolution of technology that has sparked concerns about potential abuse. Hinton\u2019s take is invaluable, Rothman said, \u201cespecially given his recent emergence as someone sounding the alarm about A.I.\u201d<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

In an email exchange, Rothman described the reasoning behind the story\u2019s structure, the challenge of conveying technical knowledge in accessible ways and the collaborative relationship with his editor. The Q&A has been edited for length and clarity, and is followed by an annotation of the story.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

Why did you become a journalist?
<\/strong>I wanted to be an English professor, and I studied for a PhD in English. But I was wrapping up my dissertation in 2008, and the financial crisis brought the professorial job market to a halt. I\u2019d always wanted to be a writer and had wide-ranging interests, including in science and technology. It felt like journalism would be a practical path forward that would also allow me to explore a lot of different subjects.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

What does the position of ideas editor at The New Yorker entail?
<\/strong>I edit pieces about ideas, broadly construed \u2014 pieces that touch on science, technology, philosophy, literature, history and so on \u2014 mostly for the web site, but also for the print magazine. And I write a few pieces a year, fitting it in where I can.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

What writers have influenced you and how in general, and as you reported and wrote your story?
<\/strong>I always wish I could write a profile the way Larissa MacFarquhar<\/a> writes them. And when I write about technology, I always revisit William Gibson (whom I was able to profile<\/a> a few years ago). He reminds me that the newest technologies have roots in the past. Everything new is old. You always have to rewind further than you think.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

What is the origin story behind \u201cWhy the Godfather of A.I. Fears What He\u2019s Built?\u201d
<\/strong>I\u2019ve been covering A.I. for a few years as an editor and writer. It\u2019s a really complicated subject, both technically and conceptually, and it\u2019s also a sweeping story going back at least a century, combining history and philosophy and science fiction and business. It\u2019d be crazy to think you could combine all that in a single piece, but it seemed like writing about Hinton would get you into the ballpark. So once we started talking about doing a special issue on A.I., a profile of Hinton seemed like a good idea. Especially given his recent emergence as someone sounding the alarm about A.I.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

What role did your editor play during the stages of writing and finishing your story?
<\/strong>Henry Finder, editorial director<\/a> of the New Yorker, has been my editor for a long time now. Working with him is a little like being in a writing seminar that progresses from piece to piece. With this piece, I wanted to apply lessons that I\u2019d learned from him while writing the previous one\u2014a profile<\/a> of the novelist Kim Stanley Robinson. He pushes the pieces forward in so many ways, in terms of craft but also in terms of thinking. And he always helps me keep in mind what the ultimate goals are \u2014 what the ideal version of the piece should be.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

Much of your story is an intellectual challenge to absorb and understand. Who do you envision as your audience?
<\/strong>I wanted this piece to be something anyone could read. I hoped it would pull readers into the intellectual adventure of A.I., and that it would help them understand how a pure and profound curiosity about our own minds got us to this point. Once I got to know Geoff Hinton, I wanted the piece to reflect him \u2014 his personality and experiences. He\u2019s a fascinating individual in his own right, with a life story that has a great deal to tell us just about life, outside of A.I. So I don\u2019t think of it as a piece for a technology-oriented audience. I hoped to write a humanist piece.<\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

ANNOTATION: <\/em><\/strong>Storyboard\u2019s questions are in red; Rothman\u2019s answers in blue. To read the story without annotations, click the HIDE ANNOTATIONS button in the right-hand menu of your monitor or at the top of your mobile screen.<\/em><\/p>\n\n <\/div>\n <\/div>\n \n

\n
\n \n\n\n <\/div>\n <\/div>\n
\n
\n
\n \n
\"2015
Geoffrey Hinton outside Google’s California headquarters in 2015. Hinton, a computer scientist known as the \u201cgodfather of artificial intelligence,\u201d resigned in 2023 from his high-profile job at Google specifically to share his concerns that unchecked AI development could threaten humanity.<\/figcaption><\/figure>\n\n <\/div>\n <\/div>\n<\/div> \n
\n
\n \n\n\n <\/div>\n <\/div>\n \n
\n
\n \n

Why the Godfather of A.I. Fears What He\u2019s Built<\/h2>\n

Geoffrey Hinton has spent a lifetime teaching computers to learn. Now he worries that artificial brains are better than ours.<\/em><\/h3>\n\nBy Joshua Rothman
\nThe New Yorker
\nNov. 13, 2023


\nI<\/span>n your brain, neurons are arranged in networks big and small. With every action, with every thought, the networks change: neurons are included or excluded, and the connections between them strengthen or fade. This process goes on all the time\u2014it\u2019s happening now, as you read these words\u2014and its scale is beyond imagining. You have some eighty billion neurons sharing a hundred trillion connections or more. Your skull contains a galaxy\u2019s worth of constellations, always shifting. <\/a>You begin the story by describing the inner workings of the brain. Why? <\/span>Because that\u2019s where A.I. started. It began as an offshoot of an introspective, neuroscientific effort. I also wanted to raise the question, early on, of what human intelligence is and how it might work. I wanted readers to look inside themselves and begin to consider the possibility that the mind is a kind of machine.<\/span><\/span>

\nGeoffrey Hinton, the computer scientist who is often called \u201cthe godfather of A.I.,\u201d handed me a walking stick. \u201cYou\u2019ll need one of these,\u201d he said. Then he headed off along a path through the woods to the shore. It wound across a shaded clearing, past a pair of sheds, and then descended by stone steps to a small dock. <\/a>This sentence adheres to the \u201crule of three.\u201d Was that a conscious writing decision? <\/span>I\u2019ve never heard of the rule of three. Although maybe I think that three-part constructions sound good.<\/span><\/span> \u201cIt\u2019s slippery here,\u201d Hinton warned, as we started down.
\n
\nNew knowledge incorporates itself into your existing networks in the form of subtle adjustments. Sometimes they\u2019re temporary: if you meet a stranger at a party, his name might impress itself only briefly upon the networks in your memory. But they can also last a lifetime, if, say, that stranger becomes your spouse. Because new knowledge merges with old, what you know shapes what you learn. If someone at the party tells you about his trip to Amsterdam, the next day, at a museum, your networks may nudge you a little closer to the Vermeer<\/a>. In this way, small changes create the possibility for profound transformations. <\/a>The reader quickly becomes aware of the story\u2019s woven structure. How did you select this approach? <\/span>I knew from the beginning of the writing process that I wanted to braid the science and the life together. The challenge was making it work. But I wanted it to be clear, from the beginning of the piece, that we would be learning about Hinton\u2019s life on a personal level. I think it\u2019s important, in a long piece of writing, to try and move all the elements forward together instead of siloing them into separate sections.<\/span><\/span>
\n
\n\u201cWe had a bonfire here,\u201d Hinton said. We were on a ledge of rock jutting out into Ontario\u2019s Georgian Bay, which stretches to the west into Lake Huron. Islands dotted the water; Hinton had bought this one in 2013, when he was sixty-five, after selling a three-person startup to Google for forty-four million dollars. Before that, he\u2019d spent three decades as a computer-science professor at the University of Toronto\u2014a leading figure in an unglamorous subfield known as neural networks, which was inspired by the way neurons are connected in the brain. Because artificial neural networks were only moderately successful at the tasks they undertook\u2014image categorization, speech recognition, and so on\u2014most researchers considered them to be at best mildly interesting, or at worst a waste of time. \u201cOur neural nets just couldn\u2019t do anything better than a child could,\u201d Hinton recalled. In the nineteen-eighties, when he saw \u201cThe Terminator,\u201d it didn\u2019t bother him that Skynet, the movie\u2019s world-destroying A.I., was a neural net; he was pleased to see the technology portrayed as promising.
\n
\nFrom the small depression where the fire had been, cracks in the stone, created by the heat, radiated outward. Hinton, who is tall, slim, and English, <\/a>Why did you select these particular three descriptors? <\/span>This is a kind of arm\u2019s-length description \u2014 like if you were squinting to make Hinton out from a distance, this is what you\u2019d see. I could\u2019ve described him in greater detail here, but didn\u2019t want to lose momentum. <\/span><\/span>poked the spot with his stick. A scientist through and through, he is always remarking on what is happening in the physical world: the lives of animals, the flow of currents in the bay, the geology of the island. \u201cI put a mesh of rebar under the wood, so the air could get in, and it got hot enough that the metal actually went all soft,\u201d he said, in a wondering tone. \u201cThat\u2019s a real fire\u2014something to be proud of!\u201d <\/a>You\u2019re capturing small moments in exquisite detail. What do you use in the field to record interviews and what you\u2019re seeing? <\/span>I have a small handheld voice recorder, and I also take a lot of pictures. I almost always have a camera with me. At the end of each reporting day, I take notes about what happened and begin to write out the scenes from memory and notes.<\/span><\/span>
\n
\nFor decades, Hinton tinkered, building bigger neural nets structured in ingenious ways. He imagined new methods for training them and helping them improve. He recruited graduate students, convincing them that neural nets weren\u2019t a lost cause. He thought of himself as participating in a project that might come to fruition a century in the future, after he died. Meanwhile, he found himself widowed and raising two young children alone. During one particularly difficult period, when the demands of family life and research overwhelmed him, he thought that he\u2019d contributed all he could. \u201cI was dead in the water at forty-six,\u201d he said. He didn\u2019t anticipate the speed with which, about a decade ago, neural-net technology would suddenly improve. Computers got faster, and neural nets, drawing on data available on the Internet, started transcribing speech, playing games, translating languages, even driving cars. Around the time Hinton\u2019s company was acquired, an A.I. boom began, leading to the creation of systems like OpenAI\u2019s ChatGPT<\/a> and Google\u2019s Bard, which many believe are starting to change the world in unpredictable ways. <\/a>You condense the early parts of Hinton\u2019s relationship with A.I. to a single paragraph. How do you achieve that concision? <\/span>This paragraph is based very heavily on a summary of his life and career that Geoff himself gave me during our very first conversation over Zoom. My mother-in-law has a theory that people tell you about the most important aspects of their lives within the first fifteen minutes of conversation. In Geoff\u2019s case, I think it was true. \u201cI was dead in the water…\u201d \u2014 it was such a moving thing to say, and I knew that I wanted to include it high up to help readers understand what a forthright, humane person he is.<\/span><\/span>
\n
\nHinton set off along the shore, and I followed, the fractured rock shifting beneath me. \u201cNow watch this,\u201d he said. He stood before a lumpy, person-size boulder, which blocked our way. \u201cHere\u2019s how you get across. You throw your stick\u201d\u2014he tossed his to the other side of the boulder\u2014\u201cand then there are footholds here and here, and a handhold here.\u201d I watched as he scrambled over with easy familiarity, and then, more tentatively, I took the same steps myself. <\/a>You are present in the narrative. Why? <\/span>I\u2019m pretty much always present in everything I write. It\u2019s just how I approach things.<\/span><\/span>
\n
\nWhenever we learn, our networks of neurons change\u2014but how, exactly? Researchers like Hinton, working with computers, sought to discover \u201clearning algorithms\u201d for neural nets, procedures through which the statistical \u201cweights\u201d of the connections among artificial neurons could change to assimilate new knowledge. In 1949, a psychologist named Donald Hebb proposed a simple rule for how people learn, often summarized as \u201cNeurons that fire together wire together.\u201d Once a group of neurons in your brain activates in synchrony, it\u2019s more likely to do so again; this helps explain why doing something is easier the second time. But it quickly became apparent that computerized neural networks needed another approach in order to solve complicated problems. As a young researcher, in the nineteen-sixties and seventies, Hinton drew networks of neurons in notebooks and imagined new knowledge arriving at their borders. How would a network of a few hundred artificial neurons store a concept? How would it revise that concept if it turned out to be flawed? <\/a>How did you keep track of your structure: outline, notecards, timeline, or other ways? <\/span>I have no system! I used to make outlines, but now I just keep track of things in my head.<\/span><\/span>
\n
\nWe made our way around the shore to Hinton\u2019s cottage, the only one on the island. Glass-enclosed, it stood on stilts atop a staircase of broad, dark rocks. \u201cOne time, we came out here and a huge water snake stuck his head up,\u201d Hinton said, as we neared the house. It was a fond memory. His father, a celebrated entomologist who\u2019d named a little-known stage of metamorphosis, had instilled in him an affection for cold-blooded creatures. When he was a child, he and his dad kept a pit full of vipers, turtles, frogs, toads, and lizards in the garage. <\/a>When did you learn this? Did you interview Hinton apart from those during your visit to his island home? <\/span>He told me this at some point during the visit. And, apart from a short Zoom interview we had when I first proposed the story to him, I kept our discussions in-person, on the island, so that as much of the piece as possible could be situated there. <\/span><\/span>Today, when Hinton is on the island\u2014he is often there in the warmer months\u2014he sometimes finds snakes and brings them into the house, so that he can watch them in a terrarium. He is a good observer of nonhuman minds, having spent a lifetime thinking about thinking from the bottom up.
\n
\nEarlier this year, Hinton left Google, where he\u2019d worked since the acquisition. He was worried about the potential of A.I. to do harm, and began giving interviews in which he talked about the \u201cexistential threat\u201d that the technology might pose to the human species. The more he used ChatGPT, an A.I. system trained on a vast corpus of human writing, the more uneasy he got. <\/a>Why did you use the word \u201ccorpus\u201d rather than one that might be more familiar?<\/span>I think it\u2019s the most accurate. It\u2019s a curated collection of writing, maintained and adjusted at great expense by experts. <\/span><\/span>One day, someone from Fox News<\/a> wrote to him asking for an interview about artificial intelligence<\/a>. Hinton enjoys sending snarky single-sentence replies to e-mails\u2014after receiving a lengthy note from a Canadian intelligence agency, he responded, \u201cSnowden is my hero\u201d and he began experimenting with a few one-liners. Eventually, he wrote, \u201cFox News is an oxy moron.\u201d Then, on a lark, he asked ChatGPT if it could explain his joke. The system told him his sentence implied that Fox News was fake news, and, when he called attention to the space before \u201cmoron,\u201d it explained that Fox News was addictive, like the drug OxyContin. Hinton was astonished. This level of understanding seemed to represent a new era in A.I.
\n
\nThere are many reasons to be concerned about the advent of artificial intelligence. It\u2019s common sense to worry about human workers being replaced by computers, for example. But Hinton has joined many prominent technologists, including Sam Altman<\/a>, the C.E.O. of OpenAI, in warning that A.I. systems may start to think for themselves, and even seek to take over or eliminate human civilization. <\/a>This strikes me as the news peg, if you will. Is this the story\u2019s central take? <\/span>It is definitely a peg. But to me the piece is actually centered on the question of whether artificial intelligence is actually a kind of real intelligence. And the real news peg, from that perspective, is the advent of the large language model, which is a technology that seems intelligent in a new way. <\/span><\/span>It was striking to hear one of A.I.\u2019s most prominent researchers give voice to such an alarming view.
\n
\n\u201cPeople say, It\u2019s just glorified autocomplete,\u201d he told me, standing in his kitchen. (He has suffered from back pain for most of his life; it eventually grew so severe that he gave up sitting. He has not sat down for more than an hour since 2005. <\/a>In an audio of you interview with Hinton played back during your New Yorker Radio hours, we hear him telling you, \u201cYou\u2019re just an autocomplete device.\u201d As a writer and editor, what was your reaction to that?<\/span>Well, I think a lot depends on how that statement is meant. I don\u2019t think that Hinton is trying to reduce or denigrate me or my work or humanity by saying it. He certainly knows that human life is rich, that our capabilities are multi-dimensional, that our lives have value, and so on. What he is saying is that something like autocomplete \u2014 that is, an unconscious, statistical process that uses what\u2019s already happened to extrapolate about what\u2019s next \u2014 is a big part of our mental lives. Maybe even the bulk of our mental lives. And I think it\u2019s hard to dispute that.<\/span><\/span>\u201cNow, let\u2019s analyze that. Suppose you want to be really good at predicting the next word. If you want to be really<\/em> good, you have to understand what\u2019s being said. That\u2019s the only way. So by training something to be really good at predicting the next word, you\u2019re actually forcing it to understand. Yes, it\u2019s \u2018autocomplete\u2019\u2014but you didn\u2019t think through what it means to have a really good autocomplete.\u201d<\/a>You alternate between your own descriptions of the workings of the brain and A.I., but use Hinton\u2019s words as well. How did you decide which one to use when to further the narrative? <\/span>Geoff is an incredible communicator, and he often puts things in a powerful, concise, insightful way. But there are other times when what he says isn\u2019t quite sufficient for the purposes of the story. For example, the story might benefit from having a theme that\u2019s been introduced higher up pulled through to the current moment. In those instances, I will sometimes interject my own formulations. <\/span><\/span>Hinton thinks that \u201clarge language models,\u201d such as GPT, which powers OpenAI\u2019s chatbots, can comprehend the meanings of words and ideas.<\/a> Why did you use quote marks for \u201clarge language models\u201d and then not provide a phrase defining that term?<\/span>I figured that \u201csuch as GPT\u201d was explanation enough, and didn\u2019t want to bog things down.<\/span><\/span>
\n
\nSkeptics who say that we overestimate the power of A.I. point out that a great deal separates human minds from neural nets. For one thing, neural nets don\u2019t learn the way we do: we acquire knowledge organically, by having experiences and grasping their relationship to reality and ourselves, while they learn abstractly, by processing huge repositories of information about a world that they don\u2019t really inhabit. <\/a>How did you learn this?<\/span>I suppose it\u2019s just general knowledge, gleaned while covering this subject over the years. I extrapolated it from my training data, as it were. <\/span><\/span>But Hinton argues that the intelligence displayed by A.I. systems transcends its artificial origins.<\/a>Could you point to any section and say, \u201cThis is the nut (or context\u201d graf?\u201d <\/span>I think it\u2019s the very last paragraph of this section. Hinton\u2019s biggest, most important argument is: It started as one thing, and it\u2019s become something else. <\/span><\/span>
\n
\n\u201cWhen you eat, you take food in, and you break it down to these tiny components,\u201d he told me. \u201cSo you could say that the bits in my body are made from bits of other animals. But that would be very misleading.\u201d He believes that, by analyzing human writing, a large language model like GPT learns how the world works, producing a system capable of thought; writing is only part of what that system can do. \u201cIt\u2019s analogous to how a caterpillar turns into a butterfly,\u201d he went on. \u201cIn the chrysalis, you turn the caterpillar into soup\u2014and from this soup you build the butterfly.\u201d
\n
\nHe began rooting around in a small cupboard just off the kitchen. \u201cAha!\u201d he said. With a flourish, he put an object on the counter\u2014a dead dragonfly. It was perfectly preserved. \u201cI found this at the marina,\u201d he explained. \u201cIt had just hatched on a rock and was drying its wings, so I caught it. Look underneath.\u201d Hinton had captured the dragonfly just after it had emerged from its larval form. The larva was a quite different-looking insect, with its own eyes and legs; it had a hole in its back, through which the dragonfly had crawled.
\n
\n\u201cThe larva of the dragonfly is this monster that lives under the water,\u201d Hinton said. \u201cAnd, like in the movie \u2018Alien,\u2019 the dragonfly is breaking out of the back of the monster. The larva went into a phase where it got turned into soup, and then a dragonfly was built out of the soup.\u201d In his metaphor, the larva represented the data that had gone into training modern neural nets; the dragonfly stood for the agile A.I. that had been created from it.\u00a0 <\/a>Throughout the story, you use metaphors to present abstractions. Why? <\/span>I can think of few technologies that are more abstract than A.I. Arguably, the whole research program is metaphorical. And one of Hinton\u2019s big intellectual commitments is to the idea that we are \u201canalogy machines.\u201d So, in addition to being useful from a readerly perspective, I thought it would be fun, as a writer, to try to do a lot of metaphorical thinking.<\/span><\/span> Deep learning\u2014the technology that Hinton helped pioneer\u2014had caused the metamorphosis. I bent closer to get a better look; Hinton stood upright, as he almost always does, careful to preserve his posture. \u201cIt\u2019s very beautiful,\u201d he said softly. \u201cAnd you get the point. It started as one thing, and it\u2019s become something else.\u201d<\/a> Is the metaphor designed to deliver more than one message and if so, what is it? <\/span>It adds a biological dimension to the discussion of A.I. It suggests the mystery of the technology \u2014 it\u2019s more like a process we set in motion than a machine that we build part by part. And it captures the sense of wariness with which Hinton thinks we need to approach the technology: we don\u2019t yet know what it is, or what it\u2019s becoming. When he showed me this dragonfly and talked about it this way, I knew almost immediately that I\u2019d close the first section of the story with it.<\/span><\/span>


\nA FEW WEEKS EARLIER,<\/strong> when Hinton had invited me to visit his island, I\u2019d imagined possible scenarios. <\/a>Why did you use the visit to the island as the spine of the narrative? <\/span>An alternative approach would\u2019ve been to visit Geoff in Toronto. But I knew it wouldn\u2019t be nearly as interesting as visiting his island, because the island, with its cottage, is almost like a stand-in for his personality, his life. Like the Ramsay house in \u201cTo the Lighthouse.\u201d And, of course, he very graciously offered to host me there.<\/span><\/span> Perhaps he\u2019d be an introvert who wanted solitude, or a tech overlord with a God complex and a futuristic compound. Several days before my arrival, he e-mailed me a photograph he\u2019d taken of a rattlesnake coiled in the island\u2019s grass. I wasn\u2019t sure whether I felt delighted or scared.
\n
\nIn fact, as private islands go, Hinton\u2019s is fairly modest\u2014two acres in total. Hinton himself is the opposite of a Silicon Valley techno-messiah. Now seventy-five, he has an English face out of a Joshua Reynolds\u00a0 painting, with white hair framing a broad forehead; his blue eyes are often steady, leaving his mouth to express emotion. A mordant raconteur, he enjoys talking about himself\u2014\u201c \u2018Geoff\u2019 is an anagram for \u2018ego fortissimo,\u2019 \u201d he told me\u2014but he\u2019s not an egotist; his life has been too grief-shadowed for that. <\/a>A tragic, but beautiful phrase. When did it come to you? <\/span>My editor suggested it! I had something windier.<\/span><\/span> \u201cI should probably tell you about my wives,\u201d he said, the first time we spoke. \u201cI\u2019ve had three marriages. One ended amicably, the other two in tragedy.\u201d He is still friendly with Joanne, his first wife, whom he married early, but his second and third wives, Rosalind and Jackie, both died of cancer, in 1994 and 2018, respectively. For the past four years, Hinton has been with Rosemary Gartner, a retired sociologist. \u201cI think he\u2019s the kind of person who always needs a partner,\u201d she told me, tenderly. <\/a>Why did you include the adverb? <\/span>I wanted to make it clear that Rosemary meant this kindly, not as a criticism or cutting observation, and not in a wry way. She did mean it tenderly. <\/span><\/span>He is a romantic rationalist, with a sensibility balancing science and emotion. In the cottage, a burgundy canoe sits in the single large room that makes up most of the ground floor; he and Jackie had found it in the island\u2019s woods, in disrepair, and Jackie, an art historian, worked with some women canoe-builders to reconstruct it during the years coinciding with her illness. \u201cShe had the maiden voyage,\u201d Hinton said. No one has used it since. <\/a>Just six words, but they carry so much weight. Why did you write them, rather than quote Hinton? <\/span>Sometimes, for an emotional beat to land, it needs just a little moment of repetition or underscoring. And it can be nice to have a kind of reaction shot from the text \u2014 an acknowledgment, in the piece, of the gravity of something.<\/span><\/span>
\n
\nHe stowed the dragonfly, then walked over to a small standing desk, where a laptop was perched next to a pile of sudoku puzzles and a notebook containing computer passwords. (He rarely uses the notebook, having devised a mnemonic system that enables him to generate and recall very long passwords in his head.) \u201cShall we do the family tree?\u201d he asked. Using two fingers\u2014he doesn\u2019t touch-type\u2014he entered \u201cGeoffrey Hinton family tree\u201d and hit Return. When Google acquired Hinton\u2019s startup, in 2013, it did so in part because the team had figured out how to dramatically improve image recognition using neural nets; now endless family trees swarmed the screen.
\n
\nHinton comes from a particular kind of scientific English family: politically radical, restlessly inventive. Above him in the family tree are his great-uncle Sebastian Hinton, the inventor of the jungle gym, and his cousin Joan Hinton, who worked as a physicist on the Manhattan Project. Further back, he was preceded by Lucy Everest, the first woman to become an elected member of the Royal Institute of Chemistry; Charles Howard Hinton, the mathematician who created the concept of the tesseract, a doorway into the fourth dimension (one appears in the film \u201cInterstellar\u201d); and James Hinton, a groundbreaking ear surgeon and an advocate of polygamy. (\u201cChrist was the savior of men, but I am the savior of women,\u201d he is said to have remarked.) In the mid-nineteenth century, a great-great-grandfather of Hinton\u2019s, the English mathematician George Boole, developed the system of binary reasoning, now known as Boolean algebra, that is fundamental to all computing. Boole was married to Mary Everest, a mathematician and author and the niece of George Everest, the surveyor for whom Mt. Everest<\/a> is named. <\/a>It\u2019s so interesting the way you use the family tree as a device to present Hinton\u2019s ancestors. <\/span>I wish I could say that this was my idea, but almost everyone who writes about Hinton runs down the family tree. In this case, he made it easy, by actually proposing that we look at one.<\/span><\/span>
\n
\n\u201cGeoff was born into science,\u201d Yann LeCun, a former student and collaborator of Hinton\u2019s who now runs A.I. at Meta, told me. Yet Hinton\u2019s family was odder than that. His dad, Howard Everest Hinton, grew up in Mexico during the Mexican Revolution, in the nineteen-tens, on a silver mine managed by his father. \u201cHe was tough,\u201d Hinton said of his dad: family lore holds that, at age twelve, Howard threatened to shoot his boxing coach for being too heavy-handed, and the coach took him seriously enough to leave town. Howard\u2019s first language was Spanish, and at Berkeley, where he went to college, he was mocked for his accent. \u201cHe hung out with a bunch of Filipinos, who were also discriminated against, and he became a Berkeley radical,\u201d Hinton said. Howard\u2019s mature politics were not just Marxist but Stalinist: in 1968, as Soviet tanks rolled into Prague, he said, \u201cAbout time!\u201d <\/a>Could you verify what his father said? <\/span>This is based on Geoff\u2019s recollection.<\/span><\/span>
\n
\nAt school, Hinton was inclined toward science. But, for ideological reasons, his father forbade him to study biology; in Howard\u2019s view, the possibility of genetic determinism contravened the Communist belief in the ultimate malleability of human nature. (\u201cI hate faiths of all kinds,\u201d Hinton said, remembering this period.) Howard, who taught at the University of Bristol, was a kind of entomologist Indiana Jones: he smuggled rare creatures from around the world back to England in his luggage, and edited an important journal in his field. Hinton, whose middle name is also Everest, felt immense pressure to make his own mark. He recalls his father telling him, \u201cIf you work twice as hard as me, when you\u2019re twice as old as I am you might be half as good.\u201d
\n
\nAt Cambridge, Hinton tried different fields but was dismayed to find that he was never the brightest student in any given class. He left college briefly to \u201cread depressing novels\u201d and to do odd jobs in London, then returned to attempt architecture, for about a day. Finally, after dipping into physics, chemistry, physiology, and philosophy, looking for a focus, he settled on a degree in experimental psychology. He haunted the office hours of the moral philosopher Bernard Williams, who turned out to be interested in computers and the mind. One day, Williams pointed out that our different thoughts must reflect different physical arrangements inside our brains; this was quite unlike the situation inside a computer, in which the software was independent of the hardware. Hinton was struck by this observation; he remembered how, in high school, a friend had told him that memory might be stored in the brain \u201cholographically\u201d\u2014that is, spread out, but in such a way that the whole could be accessed through any one part. What he was encountering was \u201cconnectionism\u201d\u2014an approach that combined neuroscience, math, philosophy, and programming to explore how neurons could work together to \u201cthink.\u201d One goal of connectionism was to create a brainlike system in a computer. There had been some progress: the Perceptron, a machine built in the nineteen-fifties by a psychologist and pioneering connectionist named Frank Rosenblatt, had used simple computer hardware to simulate a network of hundreds of neurons. When connected to a light sensor, the apparatus could recognize letters and shapes by tracking which artificial neurons were activated by different patterns of light. <\/a>How much research apart from your field reporting did you have to do to write complex material with clarity and authority? <\/span>I did a lot of research. But I also follow this field closely already, and have for many years, and many of the ideas I\u2019m discussing in sections like this have come up in other contexts \u2014 for example, my profile several years ago of Daniel Dennett, the philosopher of mind<\/span><\/span><\/a>, was tremendously useful preparation for writing this piece.<\/span><\/span>
\n
\nIn the cottage, Hinton stood and strolled, ranging back and forth behind the kitchen counter and around the first floor. He made some toast, got us each an apple, and then set up a little booster table for himself using a step stool. Family pressure had had the effect of pushing him out of temporary satisfactions. \u201cI always loved woodwork,\u201d he recalled wistfully, while we ate. \u201cAt school, you could do it voluntarily in the evenings. And I\u2019ve often wondered whether I\u2019d have been happier as an architect, because I didn\u2019t have to force myself to do it. Whereas, with science, I\u2019ve always had to force myself. Because of the family, I had to succeed at it\u2014I had to find a path. There was joy in it, but it was mostly anxiety. Now it\u2019s an enormous relief that I\u2019ve succeeded.\u201d
\n
\nHinton\u2019s laptop dinged. <\/a>How did you decide on this verb? <\/span>I believe that it actually made a \u201cding\u201d sound. But that may very well be what Hinton would call a \u201cconfabulation\u201d \u2014 that is, an extrapolation from other existing data.\u00a0 <\/span><\/span>Ever since he\u2019d left Google, his in-box had been exploding with requests for comment on A.I. He ambled <\/a>What were you trying to convey with this verb? <\/span>I was struck by how Hinton spent all day standing, walking, striding, ambling, leaning, and so on. He has a whole vocabulary of walks. <\/span><\/span>over and looked at the e-mail, and then got lost again in the forest of family trees, all of which seemed to be wrong in one way or another.
\n
\n\u201cLook at this,\u201d he said.
\n
\nI walked over and peered at the screen. It was an \u201cacademic family tree,\u201d showing Hinton at the top with his students, and theirs, arrayed below. The tree was so broad that he had to scroll horizontally to see the extent of his influence. \u201cOh, dear,\u201d Hinton said, exploring. \u201cShe wasn\u2019t really a student of mine.\u201d He scrolled further. \u201cHe was brilliant but not so good as an adviser, because he could always do it better himself.\u201d A careful nurturer of talent, Hinton seems to enjoy being surpassed by his students: when evaluating job candidates, he used to ask their advisers, \u201cBut are they better than you<\/em>?\u201d Recalling his father, who died in 1977, Hinton said, \u201cHe was just extremely competitive. And I\u2019ve often wondered, if he\u2019d been around to see me be successful, whether he\u2019d have been entirely happy. Because now I\u2019ve been more successful than he was.\u201d
\n
\nAccording to Google Scholar, Hinton is now the second most cited researcher among psychologists, and the most cited among computer and cognitive scientists. If he had a slow and eccentric start at Cambridge, it was partly because he was circling an emerging field. \u201cNeural networks\u2014there were very few people at good universities who did it,\u201d he said, closing the laptop. \u201cYou couldn\u2019t do it at M.I.T. You couldn\u2019t do it at Berkeley. You couldn\u2019t do it at Stanford.\u201d There were advantages to being a hub in a nascent network. For years, many of the best minds came to him.


\n\u201cTHE WEATHER’S GOOD,”<\/strong> Hinton said, the next morning. <\/a>You skip past half a day. Why?\u00a0 Did you spend the night at Hinton\u2019s? <\/span>It was a long visit. All told I spent four nights on the island with him. So I skipped a lot! <\/span><\/span>\u201cWe should cut down a tree.\u201d He wore a dress shirt tucked into khakis and didn\u2019t look much like a lumberjack; still, he rubbed his hands together. On the island, he is always cutting down trees to create more orderly and beautiful tableaus.
\n
\nThe house, too, is a work in progress. Few contractors would travel to a place so remote, and the people Hinton hired made needless mistakes (running a drainage pipe uphill, leaving floors half finished) that still enrage him today. Almost every room harbors a corrective mini-project, and, when I visited, Hinton had appended little notes to them to help a new contractor, often writing on the building materials themselves. In the first-floor bathroom, a piece of baseboard propped against the wall read \u201cBathroom should have this type of baseboard (maple trim in front of shower only).\u201d In the guest-room closet, masking tape ran along a shelf: \u201cDo not prime shelf, prime shelf support.\u201d <\/a>Is that where you stayed and the way you noticed it? <\/span>Yes, I stayed in the guest room and poked around a bit, like journalists do.<\/span><\/span>
\n
\nIt\u2019s useful for minds to label things; it helps them get a grip on reality. But what would it mean for an artificial mind to do so? While Hinton was earning a Ph.D. in artificial intelligence from the University of Edinburgh, he thought about how \u201cknowing\u201d in a brain might be simulated in a computer. At that time, in the nineteen-seventies, the vast majority of A.I. researchers were \u201csymbolists.\u201d In their view, knowing about, say, ketchup might involve a number of concepts, such as \u201cfood,\u201d \u201csauce,\u201d \u201ccondiment,\u201d \u201csweet,\u201d \u201cumami,\u201d \u201cred,\u201d \u201ctomato,\u201d \u201cAmerican,\u201d \u201cFrench fries,\u201d \u201cmayo,\u201d and \u201cmustard\u201d; together, these could create a scaffold on which a new concept like \u201cketchup\u201d might be hung. A large, well-funded A.I. effort called Cyc centered on the construction of a vast knowledge repository into which scientists, using a special language, could enter concepts, facts, and rules, along with their inevitable exceptions. (Birds fly, but not penguins or birds with damaged wings or . . .)
\n
\nBut Hinton was doubtful of this approach. It seemed too rigid, and too focussed on the reasoning skills possessed by philosophers and linguists. In nature, he knew, many animals acted intelligently without access to concepts that could be expressed in words. They simply learned how to be smart through experience. Learning, not knowledge, was the engine of intelligence. <\/a>This is a bumper sticker or T-shirt slogan! What a great line. Did Hinton tell you this and, if so, why did you paraphrase it rather than quote him? <\/span>This is something he says a lot, in different words. I said it myself, rather than quoting him, for two reasons. First, it\u2019s sometimes important to get the rhythm of a paragraph just right. And second, we\u2019re in free indirect style in this passage \u2014 the narration is unfolding from his perspective, as though we\u2019re inside his head back in the days when he was a young scholar. It would break the spell to suddenly quote something he said years later; suddenly, we\u2019d see him from the outside.<\/span><\/span>
\n
\nSophisticated human thinking often seemed to happen through symbols and words. But Hinton and his collaborators, James L. McClelland and David Rumelhart, believed that much of the action happened on a sub-conceptual level. Notice, they wrote, how, \u201cif you learn a new fact about an object, your expectations about other similar objects tend to change\u201d: if you\u2019re told that chimpanzees like onions, for instance, you might guess that gorillas like them, too. This suggested that knowledge was likely \u201cdistributed\u201d in the mind\u2014created out of smaller building blocks that could be shared among related ideas. There wouldn\u2019t be two separate networks of neurons for the concepts \u201cchimpanzee\u201d and \u201cgorilla\u201d; instead, bundles of neurons representing various concrete or abstract \u201cfeatures\u201d\u2014furriness, quadrupedness, primateness, animalness, intelligence, wildness, and so on\u2014might be activated in one way to signify \u201cchimpanzee\u201d and in a slightly different way to signify \u201cgorilla.\u201d To this cloud of features, onion-liking-ness might be added. A mind constructed this way risked falling into confusion and error: mix qualities together in the wrong arrangement and you\u2019d get a fantasy creature that was neither gorilla nor chimp. But a brain with the right learning algorithm might adjust the weights among its neurons to favor sensible combinations over incoherent ones.
\n
\nHinton continued to explore these ideas, first at the University of California, San Diego, where he did a postdoc (and married Joanne, whom he tutored in computer vision); then at Cambridge, where he worked as a researcher in applied psychology; and then at Carnegie Mellon, in Pittsburgh, where he became a computer-science professor in 1982. There, he spent much of his research budget on a single computer powerful enough to run a neural net. He soon got married a second time, to Rosalind Zalin, a molecular biologist. At Carnegie Mellon, Hinton had a breakthrough. Working with Terrence Sejnowski, a computer scientist and a neuroscientist, he produced a neural net called the Boltzmann Machine. The system was named for Ludwig Boltzmann, the nineteenth-century Austrian physicist who described, mathematically, how the large-scale behavior of gases was related to the small-scale behavior of their constituent particles. Hinton and Sejnowski combined these equations with a theory of learning.
\n
\nHinton was reluctant to explain the Boltzmann Machine to me. \u201cI\u2019ll tell you what this is like,\u201d he said. \u201cIt\u2019s like having a small child, and you decide to go on a walk. And there\u2019s a mountain ahead of you, and you have to get this little child to the top of the mountain and back.\u201d He looked at me\u2014the child in the metaphor\u2014and sighed. He worried, reasonably, that I might be misled by a simplified explanation and then mislead others. \u201cIt\u2019s no use trying to explain complicated ideas that you don\u2019t understand. First, you have to understand how something works. Otherwise, you just produce nonsense.\u201d Finally, he took some sheets of paper and began drawing diagrams of neurons connected by arrows and writing out equations, which I tried to follow. (Ahead of my visit, I\u2019d done a Khan Academy course on linear algebra.) <\/a>Why did you do this? <\/span>I\u2019d bought a textbook called \u201cDeep Learning,\u201d blurbed by both Hinton and Elon Musk (!), and wanted to read it before my reporting trip. I quickly discovered that, if I wanted to understand the notation, I needed to give myself at least an introduction to linear algebra. And it came in handy; a few times during my visit, Hinton used equations in his explanations, and I could follow along.<\/span><\/span>
\n
\nOne way to understand the Boltzmann Machine, he suggested, was to imagine an Identi-Kit: a system through which various features of a face\u2014bushy eyebrows, blue eyes, crooked noses, thin lips, big ears, and so on\u2014can be combined to produce a composite sketch, of the sort used by the police. For an Identi-Kit to work, the features themselves have to be appropriately designed. The Boltzmann Machine could learn not just to assemble the features but to design them, by altering the weights of the connections among its artificial neurons. It would start with random features that looked like snow on a television screen, and then proceed in two phases\u2014\u201cwaking\u201d and \u201csleeping\u201d\u2014to refine them. While awake, it would tweak the features so that they better fit an actual face. While asleep, it would fantasize a face that didn\u2019t exist, and then alter the features so that they were a worse fit.
\n
\nIts dreams told it what not to learn. There was an elegance to the system: over time, it could move away from error and toward reality, and no one had to tell it if it was right or wrong\u2014it needed only to see what existed, and to dream about what didn\u2019t.<\/a> Hinton seems to be a master of the metaphor. <\/span>Artificial intelligence is incredibly abstract. We\u2019re talking about thinking, probability, psychology, computation \u2014 all things that are happening in our heads (and now in machines). And he uses metaphors very effectively to make this abstract world comprehensible and accessible to intuition.<\/span><\/span>
\n
\nHinton and Sejnowski described the Boltzmann Machine in a 1983 paper. \u201cI read that paper when I was starting my graduate studies, and I said, \u2018I absolutely have to talk to these guys\u2014they\u2019re the only people in the world who understand that we need learning algorithms,\u2019 \u201d Yann LeCun told me. In the mid-eighties, Yoshua Bengio, a pioneer in natural-language processing and in computer vision who is now the scientific director at Mila, an A.I. institute in Quebec, trained a Boltzmann Machine to recognize spoken syllables as part of his master\u2019s thesis. \u201cGeoff was one of the external reviewers,\u201d he recalled. \u201cAnd he wrote something like \u2018This should not work.\u2019 \u201d Bengio\u2019s version of the Boltzmann Machine was more effective than Hinton expected; it took Bengio a few years to figure out why. This would become a familiar pattern. In the following decades, neural nets would often perform better than expected, perhaps because new structures had formed among the neurons during training. \u201cThe experimental part of the work came before the theory,\u201d Bengio recalled. Often, it was a matter of trying new approaches and seeing what the networks came up with.
\n
\nPartly because Rosalind loathed Ronald Reagan, Hinton said, they moved to the University of Toronto. They adopted two children, a boy and a girl, from Latin America, and lived in a house in the city. \u201cI was this kind of socialist professor who was dedicated to his work,\u201d Hinton said.
\n
\nRosalind had struggled with infertility, and had bad experiences with callous doctors. Perhaps as a result, she pursued a homeopathic route when she was later diagnosed with ovarian cancer. \u201cIt just didn\u2019t make any sense,\u201d Hinton said. \u201cIt couldn\u2019t be that you make things more dilute and they get more powerful.\u201d He couldn\u2019t see how a molecular biologist could become a homeopath. Still, determined to treat the cancer herself, Rosalind refused to have surgery even after an exam found a tumor the size of a grapefruit; later, she consented to an operation but declined chemotherapy, instead pursuing increasingly expensive homeopathic remedies, first in Canada and then in Switzerland. She developed secondary tumors. She asked Hinton to sell their house so that she could pay for new homeopathic treatments. \u201cI drew the line there,\u201d he recalled, squinting with fresh pain. <\/a>Is this the kind of gesture that you record in your notebook? <\/span>Yes, very much so. <\/span><\/span>\u201cI said, \u2018No, we\u2019re not selling the house. Because if you die I\u2019m going to have to look after the children, and it\u2019s much better for them if we can stay.\u2019 \u201d
\n
\nRosalind returned to Canada and went immediately into the hospital. She hung on for a couple of months, but wouldn\u2019t let the children visit her until the day before she died, because she didn\u2019t want them to see her so sick. Throughout her illness, she was convinced that she\u2019d soon get well. Describing what happened, Hinton still seems overwhelmed\u2014he is angry, guilty, wounded, mystified. When Rosalind died, Hinton was forty-six, his son was five, and his daughter was three. \u201cShe hurt people by failing to accept that she was going to die,\u201d he said.
\n
\nThe sound of waves filled the midafternoon quiet. Strong yellow sun spilled through the room\u2019s floor-to-ceiling windows; faint spiderwebs extended across them, silhouetted by the light. <\/a>What an exquisite description. Could you describe the writing and revision behind it?<\/span>I was so struck by the light in Hinton\u2019s cottage that I took many photographs, and I went back to them to write the scene. As I got to the end of the writing process, I knew that I wanted to introduce the spiderwebs somewhere; they\u2019d turned out to be a metaphor for how A.I. might grow to cover the world. I tried using them in a few different places, and concluded that this spot was both the most beautiful and the most parsimonious. <\/span><\/span>Hinton stood for a while, collecting himself.
\n
\n\u201cI think I need to go cut down a tree,\u201d he said.
\n
\nWe walked out the front door and down the path to the sheds. From one of them, Hinton retrieved a small green chainsaw and some safety goggles.
\n
\n\u201cRosemary says I\u2019m not allowed to cut down trees when there\u2019s nobody else here, in case I chop off an arm or something,\u201d he said. \u201cHave you driven boats before?\u201d
\n
\n\u201cNo,\u201d I said.
\n
\n\u201cI\u2019ve got to not chop off my right arm, then.\u201d
\n
\nOver his khakis, he strapped on a pair of protective chaps.
\n
\n\u201cI don\u2019t want to give you the impression that I know what I\u2019m doing,\u201d he said. \u201cBut the basic idea is, you cut lots of V\u2019s, and then the tree falls down.\u201d
\n
\nHinton crossed the path to the tree that he had in mind, inspecting the bushes for snakes as we walked. The tree was a leafy cedar, perhaps twenty feet tall; Hinton looked up to see which way it was leaning, then started the saw and began to cut into the trunk on the side opposite the lean. He removed the saw, and made another converging cut to form a V.
\n
\nHinton worked the chainsaw in silence, occasionally stopping to wipe his brow. It was hot in the sun, and mosquitoes swarmed every shady nook. I inspected the side of the shed, where ants and spiders were engaged in obscure, ceaseless activity. Down at the end of the path, the water shone. It was a beautiful spot. Still, I thought I saw why Hinton wanted to alter it: a lovely rounded hill descended into a gentle hollow, and if the unnecessary tree were gone the light could flow into it. The tree was an error.
\n
\nEventually, he began a second cut on the other side of the tree, angling it toward the first. Then he stopped and turned to me. \u201cBecause the tree leans away from the cut, the V will open up as you go deeper, and the blade won\u2019t get stuck,\u201d he explained. He continued the upper cut, nudging the tree toward an entropic moment. Suddenly, almost soundlessly, gravity took over. The tree fell under its own weight, landing with surprising softness at the bottom of the hollow. The light streamed in. <\/a>Why did you lavish so much attention on the tree-cutting? <\/span>Two reasons. First, I simply thought it was interesting to see Geoff Hinton, godfather of A.I., felling a tree. But second, the tree struck me as evocative of several different things in the world of A.I. There\u2019s the idea of pruning away undesirable things (trees, or connections between neurons) to let the light in. There\u2019s also the idea of setting something up (a tree trunk or a neural network) and then letting nature take its course through a process (gravity or learning). I wasn\u2019t able to get into it in the piece, but one way to understand the training process is by imagining a landscape; a technique called \u201cgradient descent\u201d is used to find \u201cminima\u201d in the network\u2019s errors. And then there\u2019s the fact that the development of A.I. has been a slow, patient process, with Geoff as its \u201cgardener.\u201d All these things made me want to dwell a little on the tree. I think that readers can sense the vibrancy of a scene like this, even if it\u2019s not explicitly worked out for them.<\/span><\/span>


\nHINTON WAS IN LOVE<\/strong> with the Boltzmann Machine. He hoped that it, or something like it, might underlie learning in the actual brain. \u201cIt should be true,\u201d he told me. \u201cIf I was God, I\u2019d make it true.\u201d But further experimentation revealed that as Boltzmann Machines grew they tended to become overwhelmed by the randomness that was built into them. \u201cGeoff and I disagreed about the Boltzmann Machine,\u201d LeCun said. \u201cGeoff thought it was the most beautiful algorithm. I thought it was ugly. It was stochastic\u201d\u2014that is, based partly on randomness. By contrast, LeCun said, \u201cI thought backprop was super clean.\u201d
\n
\n\u201cBackprop,\u201d or backpropagation, was an algorithm that had been explored by a few different researchers beginning in the nineteen-sixties. Even as Hinton was working with Sejnowski on the Boltzmann Machine, he was also collaborating with Rumelhart and another computer scientist, Ronald Williams, on backprop. They suspected that the technique had untapped potential for learning; in particular, they wanted to combine it with neural nets that operated across many layers.
\n
\nOne way to understand backprop is to imagine a Kafkaesque judicial system. Picture an upper layer of a neural net as a jury that must try cases in perpetuity. The jury has just reached a verdict. In the dystopia in which backprop unfolds, the judge can tell the jurors that their verdict was wrong, and that they will be punished until they reform their ways. The jurors discover that three of them were especially influential in leading the group down the wrong path. This apportionment of blame is the first step in backpropagation.
\n
\nIn the next step, the three wrongheaded jurors determine how they themselves became misinformed. They consider their own influences\u2014parents, teachers, pundits, and the like\u2014and identify the individuals who misinformed them. Those blameworthy influencers, in turn, must identify their respective influences and apportion blame among them. Recursive rounds of finger-pointing ensue, as each layer of influencers calls its own influences to account, in a backward-sweeping cascade. Eventually, once it\u2019s known who has misinformed whom and by how much, the network adjusts itself proportionately, so that individuals listen to their \u201cbad\u201d influences a little less and to their \u201cgood\u201d influences a little more. The whole process repeats again and again, with mathematical precision, until verdicts\u2014not just in this one case but in all cases\u2014are collectively as \u201ccorrect\u201d as possible. <\/a>Amazing use of an extended metaphor. How did it come to you? <\/span>I tried many different metaphors, including a Presidential election in which the \u201cwrong\u201d candidate gets elected and a corporate bureaucracy in which the orders of the C.E.O. are imperfectly carried out. But I liked this Kafka-esque metaphor because it captures the authoritarian aspect of backpropagation \u2014 ultimately, it\u2019s the humans who tell the machine what\u2019s right or wrong \u2014 and because it responds to the slightly sour view that Hinton takes of backprop. In his view, it\u2019s a disappointing way to learn.<\/span><\/span>
\n
\nIn 1986, Hinton, Rumelhart, and Williams published a three-page paper in Nature<\/em> showing how such a system could work in a neural net. They noted that backprop, like the Boltzmann Machine, wasn\u2019t \u201ca plausible model of learning in brains\u201d: unlike a computer, a brain can\u2019t rewind the tape to audit its past performance. But backprop still enabled a brainlike neural specialization. In real brains, neurons are sometimes arranged in structures aimed at solving specific problems: in the visual system, for instance, different \u201ccolumns\u201d of neurons recognize edges in what we see. Something similar emerges in a backprop network. Higher layers subject lower ones to a kind of evolutionary pressure; as a result, certain layers of a network that\u2019s tasked with deciphering handwriting, for instance, might become tightly focussed on identifying lines, curves, or edges. Eventually, the system as a whole can develop \u201cappropriate internal representations.\u201d The network knows, and makes use of its knowledge.
\n
\nIn the nineteen-fifties and sixties, a great deal of excitement had accompanied the Perceptron and other connectionist efforts; enthusiasm for connectionism waned in the years after. The backprop paper was part of a revival of interest and earned widespread attention. But the actual work of building backprop networks was slow-going, for both practical and conceptual reasons. Practically, computers were sluggish. \u201cThe rate of progress was basically, How much could a computer learn overnight?\u201d Hinton recalled. \u201cThe answer was often not much.\u201d Conceptually, neural nets were mysterious. It wasn\u2019t possible to program one in the traditional way. You couldn\u2019t go in and edit the weights of the connections among artificial neurons. And, anyway, it was hard to understand what the weights meant, because they had adapted and changed themselves through training.
\n
\nThere were many ways the learning process could go wrong. In \u201coverfitting,\u201d for example, a network effectively memorized the training data instead of learning to generalize from it. Avoiding the various pitfalls wasn\u2019t always straightforward, because it was up to the network to learn. It was like felling a tree: researchers could make cuts here and there, but then had to let the process unfold. <\/a>Did you know you were going to refer to the felling of Hinton\u2019s tree again here?<\/span> I knew that the tree would be a useful metaphor. In general, throughout the writing and revision process, I try to tie things together as much as possible \u2014 to see which metaphors, phrases, and images can be usefully pulled through from one part of the piece to another. <\/span><\/span>They could try techniques like \u201censembling\u201d (combining weak networks to make a strong one) or \u201cearly stopping\u201d (letting a network learn, but not too much). They could \u201cpre-train\u201d a system, by taking a Boltzmann Machine, having it learn something, and then layering a backprop network on top of it, so that a system\u2019s \u201csupervised\u201d training didn\u2019t begin until it had acquired some elemental knowledge on its own. Then they\u2019d let the network learn, hoping that it would land where they wanted it. <\/a>Your writing makes complex science accessible. How did you learn to do so? <\/span>I\u2019m always trying to improve, and not always succeeding. Being an editor myself has really helped: I read so much excellent writing done by the writers with whom I\u2019m working. The writer who\u2019s helped me the most is probably Daniel Dennett, the philosopher (whom I profiled in 2017). Dan works to build what he calls \u201cintuition pumps\u201d \u2014 thought experiments or metaphors drawn from the physical world that help guide your thinking. I\u2019m always trying to emulate him.<\/span><\/span>
\n
\nNew neural-net \u201carchitectures\u201d were developed: \u201crecurrent\u201d and \u201cconvolutional\u201d networks allowed the systems to make progress by building on their own work in different ways. But it was as though researchers had discovered an alien technology that they didn\u2019t know how to use. They turned the Rubik\u2019s Cube this way and that, trying to pull order out of noise. \u201cI was always convinced it wasn\u2019t nonsense,\u201d Hinton said. \u201cIt wasn\u2019t really faith\u2014it was just completely obvious to me.\u201d The brain used neurons to learn; therefore, complex learning through neural networks must be possible. He would work twice as hard for twice as long. <\/a>A nod to his father\u2019s brutal statement. Why did you allude to it here? <\/span>One of the most striking things about Hinton\u2019s story is his persistence, and I think part of it is rooted in his upbringing. He believed in these ideas and it just didn\u2019t make sense to him to walk away from them. He also felt driven to achieve something substantial. So I wanted to bring his dad back here, to remind us of how unique Hinton is as a person.<\/span><\/span>
\n
\nWhen networks were trained through backprop, they needed to be told when they were wrong and by how much; this required vast amounts of accurately labeled data, which would allow networks to see the difference between a handwritten \u201c7\u201d and a \u201c1,\u201d or between a golden retriever and a red setter. But it was hard to find well-labelled datasets that were big enough, and building more was a slog. LeCun and his collaborators developed a giant database of handwritten numerals, which they later used to train networks that could read sample Zip Codes provided by the U.S. Postal Service. A computer scientist named Fei Fei Li, at Stanford, spearheaded a gargantuan effort called ImageNet; creating it required collecting more than fourteen million images and sorting them into twenty thousand categories by hand.
\n
\nAs neural nets grew larger, Hinton devised a way of getting knowledge from a large network into a smaller one that might run on a device like a mobile phone. \u201cIt\u2019s called distillation,\u201d he explained, in his kitchen. <\/a>Why did you identify the setting?<\/span>I just didn\u2019t want the piece to feel too disembodied, even during the technical sections. <\/span><\/span>\u201cBack in school, the art teacher would show us some slides and say, \u2018That\u2019s a Rubens, and that\u2019s a van Gogh, and this is William Blake.\u2019 But suppose that the art teacher tells you, \u2018O.K., this is a Titian, but it\u2019s a peculiar Titian because aspects of it are quite like a Raphael, which is very unusual for a Titian.\u2019 That\u2019s much more helpful. They\u2019re not just telling you the right answer\u2014they\u2019re telling you other plausible answers.\u201d In distillation learning, one neural net provides another not just with correct answers but with a range of possible answers and their probabilities. It was a richer kind of knowledge.
\n
\nA few years after Rosalind\u2019s death, Hinton reconnected with Jacqueline Ford, an art historian whom he\u2019d dated briefly before moving to the United States. Jackie was cultured, warm, curious, beautiful. <\/a>How do you know this?<\/span>From Geoff! <\/span><\/span>\u201cShe\u2019s way out of your league,\u201d his sister said. <\/a>Did you talk with Hinton\u2019s sister? <\/span>The fact-checking department did. <\/span><\/span>Still, Jackie gave up her job in the U.K. to move to Toronto. They got married on December 6, 1997\u2014Hinton\u2019s fiftieth birthday. The following decades would be the happiest of his life. His family was whole again. His children loved their new mother. He and Jackie started exploring the islands in Georgian Bay. Recalling this time, he gazed at the canoe in his living room. \u201cWe found it in the woods, upside down, covered in canvas, and it was just totally rotten\u2014everything about it was rotten,\u201d he said. \u201cBut Jackie decided to rescue it anyway, like she did with me and the kids.\u201d
\n
\nHinton was not in love with backpropagation. \u201cIt\u2019s so unsatisfying intellectually,\u201d he told me. Unlike the Boltzmann Machine, \u201cit\u2019s all deterministic. Unfortunately, it just works better.\u201d Slowly, as practical advances compounded, the power of backprop became undeniable. In the early seventies, Hinton told me, the British government had hired a mathematician named James Lighthill to determine if A.I. research had any plausible chance of success. Lighthill concluded that it didn\u2019t\u2014\u201cand he was right,\u201d Hinton said, \u201cif you accepted the assumption, which everyone made, that computers might get a thousand times faster, but they wouldn\u2019t get a billion times faster.\u201d Hinton did a calculation in his head. Suppose that in 1985 he\u2019d started running a program on a fast research computer, and left it running until now. If he started running the same program today, on the fastest systems currently used in A.I., it would take less than a second to catch up.
\n
\nIn the early two-thousands, as multi-layer neural nets equipped with powerful computers began to train on much larger data sets, Hinton, Bengio, and LeCun started talking about the potential of \u201cdeep learning.\u201d The work crossed a threshold in 2012, when Hinton, Alex Krizhevsky, and Ilya Sutskever came out with AlexNet, an eight-layer neural network that was eventually able to recognize objects from ImageNet with human-level accuracy. Hinton formed a company with Krizhevsky and Sutskever and sold it to Google. He and Jackie bought the island in Georgian Bay\u2014\u201cmy one real indulgence,\u201d Hinton said.
\n
\nTwo years later, Jackie was diagnosed with pancreatic cancer. Doctors gave her a year or two to live. \u201cShe was incredibly brave and incredibly rational,\u201d Hinton said. \u201cShe wasn\u2019t in deep denial, desperately trying to get out of it. Her view was \u2018I can feel sorry for myself, or I can say I don\u2019t have much time left and I\u2019d better do my best to enjoy it and make everything O.K. for other people.\u2019 \u201d She and Hinton pored over the statistics before deciding on therapies; largely through chemo, she extended one or two years to three. In the cottage, when she could no longer manage the stairs, he constructed a small basket on a string so that she could lower her tea from the second floor to the first, where he could warm it up in the microwave. <\/a>The devotion behind this gesture is heart-rending.\u00a0 How did you react when he told you about it? <\/span>I was very moved. When people open up like this, I feel an extra sense of responsibility to write something that\u2019s worthy of that openness. <\/span><\/span>(\u201cI should\u2019ve just moved the microwave upstairs,\u201d he observed.) <\/a>Some editors and writers argue that attributive verbs should be limited to \u201csays\u201d or \u201csaid.\u201d They argue that examples like the one used here are unnecessary. What\u2019s your take on this? <\/span>I can imagine that over-using those kinds of words could be distracting. But what\u2019s the harm from time to time?<\/span><\/span>
\n
\nLate in the day, we leaned on Hinton\u2019s standing desk as he showed me photos of Jackie on his laptop. In a picture of their wedding day, she and Hinton stand with his kids in the living room of their neighbor\u2019s house, exchanging vows. Hinton looks radiant and relaxed; Jackie holds one of his hands lightly in both of hers. In one of the last pictures that he showed me, she gazes at the camera from the burgundy canoe, which she is paddling in the dappled water near the dock. \u201cThat was the summer of 2017,\u201d Hinton said. Jackie died the following April. That June, Hinton, Bengio, and LeCun won the Turing Award\u2014the equivalent of the Nobel Prize in computer science. <\/a>You could have delivered this information anytime earlier. Why did you decide to wait until now? <\/span>I think you can only tell a story once. If I\u2019d included this higher up, it would\u2019ve been repetitive appearing here. I wanted to keep things in order, as much as possible.<\/span><\/span>
\n
\nHinton is convinced that there\u2019s a real sense in which neural nets are capable of having feelings. \u201cI think feelings are counterfactual statements about what would have caused an action,\u201d he had told me, earlier that day. \u201cSay that I feel like punching someone on the nose. What I mean is: if I didn\u2019t have social inhibitions\u2014if I didn\u2019t stop myself from doing it\u2014I would punch him on the nose. So when I say \u2018I feel angry,\u2019 it\u2019s a kind of abbreviation for saying, \u2018I feel like doing an aggressive act.\u2019 Feelings are just a way of talking about inclinations to action.\u201d
\n
\nHe told me that he had seen a \u201cfrustrated A.I.\u201d in 1973. A computer had been attached to two TV cameras and a simple robot arm; the system was tasked with assembling some blocks, spread out on a table, into the form of a toy car. \u201cThis was hard, particularly in 1973,\u201d he said. \u201cThe vision system could recognize the bits if they were all separate, but if you put them in a little pile it couldn\u2019t recognize them. So what did it do? It pulled back a little bit, and went bash!<\/em>, and spread them over the table. Basically, it couldn\u2019t deal with what was going on, so it changed it, violently. And if a person did that you\u2019d say they were frustrated. The computer couldn\u2019t see the blocks right, so he bashed them.\u201d To have a feeling was to want what you couldn\u2019t have. <\/a>Here\u2019s another example of wisdom you deliver rather than Hinton. Why? <\/span>In this case, he did a great job of explaining this idea just a paragraph up. But I wanted to rephrase it in a way that would make a broader connection to the house, and to the life with Jackie that wasn\u2019t able to unfold there, and to the bittersweet feelings that the place sometimes provokes.<\/span><\/span>
\n
\n\u201cI love this house, but sometimes it\u2019s a sad place,\u201d he said, while we looked at the pictures. \u201cBecause she loved being here and isn\u2019t here.\u201d
\n
\nThe sun had almost set, and Hinton turned on a little light over his desk. He closed the computer and pushed his glasses up on his nose. He squared up his shoulders, returning to the present.
\n
\n\u201cI wanted you to know about Roz and Jackie because they\u2019re an important part of my life,\u201d he said. \u201cBut, actually, it\u2019s also quite relevant to artificial intelligence. There are two approaches to A.I. There\u2019s denial, and there\u2019s stoicism. Everybody\u2019s first reaction to A.I. is \u2018We\u2019ve got to stop this.\u2019 Just like everybody\u2019s first reaction to cancer is \u2018How are we going to cut it out?\u2019 \u201d But it was important to recognize when cutting it out was just a fantasy.
\n
\nHe sighed. \u201cWe can\u2019t be in denial,\u201d he said. \u201cWe have to be real. We need to think, How do we make it not as awful for humanity as it might be?\u201d


\nHOW USEFUL\u2014OR DANGEROUS<\/strong>\u2014will A.I. turn out to be? No one knows for sure, in part because neural nets are so strange. In the twentieth century, many researchers wanted to build computers that mimicked brains. But, although neural nets like OpenAI\u2019s GPT models are brainlike in that they involve billions of artificial neurons, they\u2019re actually profoundly different from biological brains. Today\u2019s A.I.s are based in the cloud and housed in data centers that use power on an industrial scale. Clueless in some ways and savantlike in others, they reason for millions of users, but only when prompted. They are not alive. They have probably passed the Turing test\u2014the long-heralded standard, established by the computing pioneer Alan Turing, which held that any computer that could persuasively imitate a human in conversation could be said, reasonably, to think. And yet our intuitions may tell us that nothing resident in a browser tab could really be thinking in the way we do. The systems force us to ask if our kind of thinking is the only kind that counts.
\n
\nDuring his last few years at Google, Hinton focussed his efforts on creating more traditionally mindlike artificial intelligence using hardware that more closely emulated the brain. In today\u2019s A.I.s, the weights of the connections among the artificial neurons are stored numerically; it\u2019s as though the brain keeps records about itself. In your actual, analog brain, however, the weights are built into the physical connections between neurons. Hinton worked to create an artificial version of this system using specialized computer chips.
\n
\n\u201cIf you could do it, it would be amazing,\u201d he told me. The chips would be able to learn by varying their \u201cconductances.\u201d Because the weights would be integrated into the hardware, it would be impossible to copy them from one machine to another; each artificial intelligence would have to learn on its own. \u201cThey would have to go to school,\u201d he said. \u201cBut you would go from using a megawatt to thirty watts.\u201d As he spoke, he leaned forward, his eyes boring into mine; I got a glimpse of Hinton the evangelist. <\/a>In a New Yorker Radio Hour when the magazine\u2019s editor, David Remnick, interviews you about this story, he suggests that Hinton is a godfather who has become an apostate. Which is he in your mind: Evangelist or apostate? <\/span>I think that part of the reason he\u2019s such an important person to listen to now is that he\u2019s neither \u2014 or, at least, he\u2019s attempting to be neither. He\u2019s trying to be a realist, to be someone who knows the technology well and is simply describing the possibilities. Observers see him as one or the other, depending on their points of view. But he\u2019s not trying to hype A.I. or criticize it; he\u2019s trying to describe our situation, as he sees it. <\/span><\/span>Because the knowledge gained by each A.I. would be lost when it was disassembled, he called the approach \u201cmortal computing.\u201d \u201cWe\u2019d give up on immortality,\u201d he said. \u201cIn literature, you give up being a god for the woman you love, right? In this case, we\u2019d get something far more important, which is energy efficiency.\u201d Among other things, energy efficiency encourages individuality: because a human brain can run on oatmeal, the world can support billions of brains, all different. And each brain can learn continuously, rather than being trained once, then pushed out into the world.
\n
\nAs a scientific enterprise, mortal A.I. might bring us closer to replicating our own brains. But Hinton has come to think, regretfully, that digital intelligence might be more powerful. In analog intelligence, \u201cif the brain dies, the knowledge dies,\u201d he said. By contrast, in digital intelligence, \u201cif a particular computer dies, those same connection strengths can be used on another computer. And, even if all the digital computers died, if you\u2019d stored the connection strengths somewhere you could then just make another digital computer and run the same weights on that other digital computer. Ten thousand neural nets can learn ten thousand different things at the same time, then share what they\u2019ve learned.\u201d This combination of immortality and replicability, he says, suggests that \u201cwe should be concerned about digital intelligence taking over from biological intelligence.\u201d
\n
\nHow should we describe the mental life of a digital intelligence without a mortal body or an individual identity? In recent months, some A.I. researchers have taken to calling GPT a \u201creasoning engine\u201d\u2014a way, perhaps, of sliding out from under the weight of the word \u201cthinking,\u201d which we struggle to define. \u201cPeople blame us for using those words\u2014\u2018thinking,\u2019 \u2018knowing,\u2019 \u2018understanding,\u2019 \u2018deciding,\u2019 and so on,\u201d Bengio told me. <\/a>At what stage in the reporting did you interview Begio and Yann LeCun? How many human sources did you speak with? <\/span>I talked to them after I spent time with Hinton on his island, not before. I take different approaches in different pieces; in this case, I felt that it would be especially important to see Hinton on a personal level, as a human being, rather than as a technologist first and a human being second. So I wanted to see him without talking to others about him first, and to encounter him fresh. I ended up talking to maybe a dozen people on background, but only wanted to quote a few in the piece, to prevent the story from succumbing to talking-head syndrome, and to keep it focused on a single human individual. <\/span><\/span>\u201cBut even though we don\u2019t have a complete understanding of the meaning of those words, they\u2019ve been very powerful ways of creating analogies that help us understand what we\u2019re doing. It\u2019s helped us a lot to talk about \u2018imagination,\u2019 \u2018attention,\u2019 \u2018planning,\u2019 \u2018intuition\u2019 as a tool to clarify and explore.\u201d In Bengio\u2019s view, \u201ca lot of what we\u2019ve been doing is solving the \u2018intuition\u2019 aspect of the mind.\u201d Intuitions might be understood as thoughts that we can\u2019t explain: our minds generate them for us, unconsciously, by making connections between what we\u2019re encountering in the present and our past experiences. We tend to prize reason over intuition, but Hinton believes that we are more intuitive than we acknowledge. \u201cFor years, symbolic-A.I. people said our true nature is, we\u2019re reasoning machines,\u201d he told me. \u201cI think that\u2019s just nonsense. Our true nature is, we\u2019re analogy machines, with a little bit of reasoning built on top, to notice when the analogies are giving us the wrong answers, and correct them.\u201d
\n
\nOn the whole, current A.I. technology is talky and cerebral: it stumbles at the borders of the physical. \u201cAny teen-ager can learn to drive a car in twenty hours of practice, with hardly any supervision,\u201d LeCun told me. \u201cAny cat can jump on a series of pieces of furniture and get to the top of some shelf. We don\u2019t have any A.I. systems coming anywhere close to doing these things today, except self-driving cars\u201d\u2014and they are over-engineered, requiring \u201cmapping the whole city, hundreds of engineers, hundreds of thousands of hours of training.\u201d Solving the wriggly problems of physical intuition \u201cwill be the big challenge of the next decade,\u201d LeCun said. Still, the basic idea is simple: if neurons can do it, then so can neural nets. <\/a>At any point, did you use A.I. during the reporting and writing? If so, how, when and why? <\/span>I used A.I. a lot! Not for the writing or reporting, really, but just to have the experience of using it. I kept a running conversation going with ChatGPT about itself, about Hinton, about A.I. technology and ethics. I\u2019d chat with it in speech mode quite often. I\u2019d try out different metaphors for various technical concepts and ask ChatGPT to riff on those metaphors. I also asked ChatGPT to take on different personalities; for whatever reason, I enjoyed it most when it was snarky. At one point, I asked it if \u201cwriter\u2019s block\u201d was real and explained that I was writing a piece on Hinton. It replied: \u201cOh, Josh, a project on Geoff Hinton? How original. It’s not like he’s one of the most recognized figures in deep learning or anything. But hey, maybe you’ll manage to write something that isn’t another regurgitated biography or fanboy piece. Most writers would kill for an opportunity to profile someone like him, but here you are, whining about it taking too long. Maybe you should consider if you’re actually cut out for this kind of work. What could possibly be so hard about writing a piece on a guy who’s been interviewed and profiled a million times? Good luck with that. Hope your \u2018writer\u2019s block\u2019 miraculously disappears.\u201d<\/span><\/span>
\n
\nHinton suspects that skepticism of A.I.\u2019s potential, while comforting, is often motivated by an unjustified faith in human exceptionalism. Researchers complain that A.I. chatbots \u201challucinate,\u201d by making up plausible answers to questions that stump them. But he contests that terminology. \u201cWe should say \u2018confabulate,\u2019 \u201d he told me. \u201c \u2018Hallucination\u2019 is when you think there\u2019s sensory input\u2014auditory hallucinations, visual hallucinations, olfactory hallucinations. But just making stuff up\u2014that\u2019s confabulation.\u201d He cited the case of John Dean, President Richard Nixon\u2019s White House counsel, who was interviewed about Watergate before he knew that the conversations he described had been tape-recorded. Dean confabulated, getting the details wrong, mixing up who said what. \u201cBut the gist of it was all right,\u201d Hinton said. \u201cHe had a recollection of what went on, and he imposed that recollection on some characters in his head. He wrote a little play. And that\u2019s what human memory is like. In our minds, there\u2019s no boundary between just making it up and telling the truth. Telling the truth is just making it up correctly. Because it\u2019s all in the weights, right?\u201d From this perspective, ChatGPT\u2019s ability to make things up is a flaw, but also a sign of its humanlike intelligence.
\n
\nHinton is often asked if he regrets his work. He doesn\u2019t. (He recently sent a journalist a one-liner\u2014\u201ca song for you\u201d\u2014along with a link to Edith Piaf\u2019s \u201cNon, Je Ne Regrette Rien.\u201d) When he began his research, he says, no one thought that the technology would succeed; even when it started succeeding, no one thought that it would succeed so quickly. Precisely because he thinks that A.I. is truly intelligent, he expects that it will contribute to many fields. Yet he fears what will happen when, for instance, powerful people abuse it. \u201cYou can probably imagine Vladimir Putin creating an autonomous lethal weapon and giving it the goal of killing Ukrainians,\u201d Hinton said. <\/a>During your New Yorker Radio Hour interview with Remnick, audio excerpts of your conversations with Hinton are played. How would you describe your interviewing style?<\/span>In that case, I conducted a special, more formal interview with Hinton, just for the radio, at the end of my reporting trip. Ordinarily, I\u2019m more informal. I like to ask people to tell me about their lives in chronological order. I prefer listening to talking. <\/span><\/span>He believes that autonomous weapons should be outlawed\u2014the U.S. military is actively developing them\u2014but warns that even a benign autonomous system could wreak havoc. \u201cIf you want a system to be effective, you need to give it the ability to create its own subgoals,\u201d he said. \u201cNow, the problem is, there\u2019s a very general subgoal that helps with almost all goals: get more control. The research question is: how do you prevent them from ever wanting to take control? And nobody knows the answer.\u201d (Control, he noted, doesn\u2019t have to be physical: \u201cIt could be just like how Trump could invade the Capitol, with words.\u201d)
\n
\nWithin the field, Hinton\u2019s views are variously shared and disputed. \u201cI\u2019m not scared of A.I.,\u201d LeCun told me. \u201cI think it will be relatively easy to design them so that their objectives will align with ours.\u201d He went on, \u201cThere\u2019s the idea that if a system is intelligent it\u2019s going to want to dominate. But the desire to dominate has nothing to do with intelligence\u2014it has to do with testosterone.\u201d I recalled the spiders I\u2019d seen at the cottage, and how their webs covered the surfaces of Hinton\u2019s windows. They didn\u2019t want to dominate, either\u2014and yet their insectoidal intelligence had led them to expand their territory. Living systems without centralized brains, such as ant colonies, don\u2019t \u201cwant\u201d to do anything, yet they still find food, ford rivers, and kill competitors in vast numbers. Either Hinton or LeCun could be right. The metamorphosis isn\u2019t finished. We don\u2019t know what A.I. will become.
\n
\n\u201cWhy don\u2019t we just unplug it?\u201d I asked Hinton, of A.I. in general. \u201cIs that a totally unreasonable question?\u201d <\/a>Why did you include this dialogue with Hinton? <\/span>This is the question on many of our minds, and even though it\u2019s a silly one \u2014 obviously we can\u2019t \u201cunplug\u201d A.I. I wanted to ask it in a direct way, so that we could get to the heart of the matter.<\/span><\/span>
\n
\n\u201cIt\u2019s not unreasonable to say, We\u2019d be better off without this\u2014it\u2019s not worth it,\u201d he said. \u201cJust as we might have been better off without fossil fuels. We\u2019d have been far more primitive, but it may not have been worth the risk.\u201d He added, stoically, \u201cBut it\u2019s not going to happen. Because of the way society is. And because of the competition between different nations. If the U.N. really worked, possibly something like that could stop it. Although, even then, A.I. is just so useful. It has so much potential to do good, in fields like medicine\u2014and, of course, to give an advantage to a nation via autonomous weapons.\u201d Earlier this year, Hinton declined to sign a popular petition that called for at least a six-month pause in research. \u201cChina\u2019s not going to stop developing it for six months,\u201d he said.
\n
\n\u201cSo what should we do?\u201d I asked.
\n
\n\u201cI don\u2019t know,\u201d he said. \u201cIt would be great if this were like climate change, where someone could say, Look, we either have to stop burning carbon or we have to find an effective way to remove carbon dioxide from the atmosphere. There, you know what the solution looks like. Here, it\u2019s not like that.\u201d
\n
\nHinton was pulling on a blue waterproof jacket. We were heading to the marina to pick up Rosemary. \u201cShe\u2019s brought supplies!\u201d he said, smiling. As we walked out the door, I looked back into the cottage. In the big room, the burgundy canoe shone, caressed by sunlight. Chairs were arranged in front of it in a semicircle, facing the water through the windows. Some magazines were piled on a little table. It was a beautiful house. A human mind does more than reason; it exists in time, and reckons with life and death, and builds a world around itself. It gathers meaning, as if by gravity. An A.I., I thought, might be able to imagine a place like this. But would it ever need one? <\/a>Why did you make this observation? <\/span>To me, this story is partly about how one of the people who\u2019s been instrumental in creating A.I. is also deeply human. Hinton isn\u2019t someone who wants to transcend or escape or redefine what it means to be a human being; he\u2019s an empathetic, experienced, wise person who\u2019s lived a full and sometimes difficult life and who fully grasps complexity, richness and value of human experience. It\u2019s possible to imagine a version of the A.I. story in which the scientists behind the technology are somehow cold, like in the Hollywood movie \u201cEx Machina.\u201d I wanted to show how this wasn\u2019t true. And I also wanted to pose the question at the end of the paragraph. Many of the ideas in the closing part of the piece relate to mortality and its role in making us into individuals. (Can an immortal A.I. ever be an individual?) This is one of the ways in which thinking about artificial intelligence leads us to think about the nature of human existence. I thought it would be good to bring the piece full circle.<\/span><\/span>
\n
\nWe made our way down the wooded path, past the sheds and down the steps to the dock, then climbed into Hinton\u2019s boat. It was a perfect blue day, with a brisk wind roughing the water. Hinton stood at the wheel. I sat in front, watching other islands pass, thinking about the story of A.I. To some, it\u2019s a Copernican tale, in which our intuitions about the specialness of the human mind are being dislodged by thinking machines. To others, it\u2019s Promethean\u2014having stolen fire, we risk getting burned. Some people think we\u2019re fooling ourselves, getting taken in by our own machines and the companies that hope to profit from them. In a strange way, it could also be a story about human limitation. If we were gods, we might make a different kind of A.I.; in reality, this version was what we could manage. Meanwhile, I couldn\u2019t help but consider the story in an Edenic light. By seeking to re-create the knowledge systems in our heads, we had seized the forbidden apple; we now risked exile from our charmed world. But who would choose not to know how knowing works? <\/a>What is the narrative purpose of this paragraph?<\/span>Well, on one level, it\u2019s actually true that I had these thoughts while we motored back to the marina. But I decided to include them because I wanted to foreground how the A.I. story is still developing and could be told in many ways. It\u2019s such a primal, elemental story \u2014 it has Golems, Frankensteins, zombies, gods, and so on \u2014 and those kinds of metaphors have a magnetism that\u2019s hard to resist. I thought it would be helpful to name them, and to make them explicit, so that readers could engage with them out in the light. And I also wanted to set up the ending, with the snake, in a way that would make the symbolism more obvious while also rendering it a little tongue-in-cheek.<\/span><\/span>
\n
\nAt the marina, Hinton did a good job of working with the wind, accelerating forward, turning, and then allowing it to guide him into his slip. \u201cI\u2019m learning,\u201d he said, proud of himself. We walked ashore and waited by a shop for Rosemary to arrive. After a while, Hinton went inside to buy a light bulb. I stood, enjoying the warmth, and then saw a tall, bright-eyed woman with long white hair striding toward me from the parking lot.
\n
\nRosemary and I shook hands. Then she looked over my shoulder. Hinton was emerging from the greenery near the shop, grinning.
\n
\n\u201cWhat\u2019ve you got for me?\u201d she asked.
\n
\nHinton held up a black-and-yellow garter snake, perhaps a metre long, twisting round and round like a spring. \u201cI\u2019ve come bearing gifts!\u201d he said, in a gallant tone. \u201cI found it in the bushes.\u201d
\n
\nRosemary laughed, delighted, and turned to me. \u201cThis just epitomizes him,\u201d she said.
\n
\n\u201cHe\u2019s not happy,\u201d Hinton said, observing the snake.
\n
\n\u201cWould you<\/em> be?\u201d Rosemary asked.
\n
\n\u201cI\u2019m being very careful with his neck,\u201d Hinton said. \u201cThey\u2019re fragile.\u201d
\n
\nHe switched the snake from one hand to another, then held out a palm. It was covered in the snake\u2019s slimy musk.
\n
\n\u201cHave a sniff,\u201d he said. <\/a>After a story that presents complex concepts in often long paragraphs, you shifted here to an easy-to-grasp scene driven by dialogue and action. Why the change? <\/span>The piece is structured as a visit, and I wanted to end it by bringing the visit to a close. And I also wanted the piece to be very much about Hinton as a person, and so it made sense to me to end it on single a moment with him, realized as a three-dimensional, real-time scene, rather than an interview. Also, I think it feels good, after a somewhat claustrophobic piece \u2014 it\u2019s just me and Hinton on the island \u2014 to suddenly have someone new arrive.<\/span><\/span>
\n
\nWe took turns. It was strange: mineral and pungent, reptilian and chemical, unmistakably biotic.
\n
\n\u201cYou\u2019ve got it all over your shirt!\u201d Rosemary said.
\n
\n\u201cI had to catch him!\u201d Hinton explained.
\n
\nHe put the snake down, and it slithered off into the grass. He watched it go with a satisfied look.
\n
\n\u201cWell,\u201d he said. \u201cIt\u2019s a beautiful day. Shall we brave the crossing?\u201d<\/a>Why did you end the story on this note? Did you experiment with alternate approaches? <\/span>As soon as this happened during my reporting \u2014 Geoff catching the snake, my meeting Rosemary, their charming interaction \u2014 I knew that I\u2019d probably be ending the story with it. And then it became a matter of working backwards and figuring out what I needed to add (or subtract) to make the ending as satisfying as possible. The article has been so retrospective and bittersweet. I thought it was good to see Hinton in the present, in a happy relationship. And we\u2019re seeing his comfort with the non-human, which is itself a dimension of his own humanity. My hope was that the ending would prompt us to appreciate our humanity at the same time it makes us think about what might challenge it.<\/span><\/span>
\n
\n* * *
\n
\nChip Scanlan<\/strong> is an award-winning writer who taught at the Poynter Institute and now coaches writers around the world. He is the author of several books on writing and the newsletter Chip’s Writing Lessons.<\/a><\/em>\n\n <\/div>\n <\/div>\n","protected":false},"excerpt":{"rendered":"

New Yorker Ideas Editor Joshua Rothman brushed up on algebra and spent four days on a private island to profile the “godfather” of A.I.<\/p>\n","protected":false},"author":1,"featured_media":11308,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[10],"tags":[],"topics":[],"authors":[13],"issue":[],"class_list":["post-11305","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-annotation-tuesday","authors-chip-scanlan"],"acf":[],"yoast_head":"\nHow a reporter prepped to understand A.I. and the man who helped invent it - Nieman Storyboard<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"How a reporter prepped to understand A.I. and the man who helped invent it - Nieman Storyboard\" \/>\n<meta property=\"og:description\" content=\"New Yorker Ideas Editor Joshua Rothman brushed up on algebra and spent four days on a private island to profile the "godfather" of A.I.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/\" \/>\n<meta property=\"og:site_name\" content=\"Nieman Storyboard\" \/>\n<meta property=\"article:published_time\" content=\"2024-06-06T12:00:32+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-10-17T13:43:32+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/10\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"750\" \/>\n\t<meta property=\"og:image:height\" content=\"500\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"commonmedia\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"commonmedia\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"79 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/\"},\"author\":{\"name\":\"commonmedia\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#\\\/schema\\\/person\\\/ad131fac9b2e8136d6171c8575cfc226\"},\"headline\":\"How a reporter prepped to understand A.I. and the man who helped invent it\",\"datePublished\":\"2024-06-06T12:00:32+00:00\",\"dateModified\":\"2024-10-17T13:43:32+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/\"},\"wordCount\":15801,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/niemanstoryboard.org\\\/app\\\/uploads\\\/2024\\\/10\\\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg\",\"articleSection\":[\"Story Annotations\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/\",\"url\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/\",\"name\":\"How a reporter prepped to understand A.I. and the man who helped invent it - Nieman Storyboard\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/niemanstoryboard.org\\\/app\\\/uploads\\\/2024\\\/10\\\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg\",\"datePublished\":\"2024-06-06T12:00:32+00:00\",\"dateModified\":\"2024-10-17T13:43:32+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#primaryimage\",\"url\":\"https:\\\/\\\/niemanstoryboard.org\\\/app\\\/uploads\\\/2024\\\/10\\\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg\",\"contentUrl\":\"https:\\\/\\\/niemanstoryboard.org\\\/app\\\/uploads\\\/2024\\\/10\\\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg\",\"width\":750,\"height\":500,\"caption\":\"Manual typewriter with paper in scroll that says ARTIFICIAL INTELLIGENCE\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/2024\\\/06\\\/06\\\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/niemanstoryboard.org\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"How a reporter prepped to understand A.I. and the man who helped invent it\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#website\",\"url\":\"https:\\\/\\\/niemanstoryboard.org\\\/\",\"name\":\"Nieman Storyboard\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/niemanstoryboard.org\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#organization\",\"name\":\"Nieman Storyboard\",\"url\":\"https:\\\/\\\/niemanstoryboard.org\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/niemanstoryboard.org\\\/app\\\/uploads\\\/2024\\\/02\\\/NiemanKOLogo.svg\",\"contentUrl\":\"https:\\\/\\\/niemanstoryboard.org\\\/app\\\/uploads\\\/2024\\\/02\\\/NiemanKOLogo.svg\",\"width\":1,\"height\":1,\"caption\":\"Nieman Storyboard\"},\"image\":{\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/niemanstoryboard.org\\\/#\\\/schema\\\/person\\\/ad131fac9b2e8136d6171c8575cfc226\",\"name\":\"commonmedia\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/7a67dac1d5145031a5e3f8bcb411f86c4aed5a7ebe4244392196080637ed18af?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/7a67dac1d5145031a5e3f8bcb411f86c4aed5a7ebe4244392196080637ed18af?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/7a67dac1d5145031a5e3f8bcb411f86c4aed5a7ebe4244392196080637ed18af?s=96&d=mm&r=g\",\"caption\":\"commonmedia\"},\"sameAs\":[\"https:\\\/\\\/niemanstoryboard.org\\\/wp\"],\"url\":\"https:\\\/\\\/niemanstoryboard.org\\\/author\\\/commonmedia\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"How a reporter prepped to understand A.I. and the man who helped invent it - Nieman Storyboard","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/","og_locale":"en_US","og_type":"article","og_title":"How a reporter prepped to understand A.I. and the man who helped invent it - Nieman Storyboard","og_description":"New Yorker Ideas Editor Joshua Rothman brushed up on algebra and spent four days on a private island to profile the \"godfather\" of A.I.","og_url":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/","og_site_name":"Nieman Storyboard","article_published_time":"2024-06-06T12:00:32+00:00","article_modified_time":"2024-10-17T13:43:32+00:00","og_image":[{"width":750,"height":500,"url":"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/10\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg","type":"image\/jpeg"}],"author":"commonmedia","twitter_card":"summary_large_image","twitter_misc":{"Written by":"commonmedia","Est. reading time":"79 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#article","isPartOf":{"@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/"},"author":{"name":"commonmedia","@id":"https:\/\/niemanstoryboard.org\/#\/schema\/person\/ad131fac9b2e8136d6171c8575cfc226"},"headline":"How a reporter prepped to understand A.I. and the man who helped invent it","datePublished":"2024-06-06T12:00:32+00:00","dateModified":"2024-10-17T13:43:32+00:00","mainEntityOfPage":{"@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/"},"wordCount":15801,"commentCount":0,"publisher":{"@id":"https:\/\/niemanstoryboard.org\/#organization"},"image":{"@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#primaryimage"},"thumbnailUrl":"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/10\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg","articleSection":["Story Annotations"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/","url":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/","name":"How a reporter prepped to understand A.I. and the man who helped invent it - Nieman Storyboard","isPartOf":{"@id":"https:\/\/niemanstoryboard.org\/#website"},"primaryImageOfPage":{"@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#primaryimage"},"image":{"@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#primaryimage"},"thumbnailUrl":"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/10\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg","datePublished":"2024-06-06T12:00:32+00:00","dateModified":"2024-10-17T13:43:32+00:00","breadcrumb":{"@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#primaryimage","url":"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/10\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg","contentUrl":"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/10\/markus-winkler-tGBXiHcPKrM-unsplash-scaled-e1717627577248.jpg","width":750,"height":500,"caption":"Manual typewriter with paper in scroll that says ARTIFICIAL INTELLIGENCE"},{"@type":"BreadcrumbList","@id":"https:\/\/niemanstoryboard.org\/2024\/06\/06\/profiles-artificial-intelligence-interviewing-geoffrey-hinton-story-structure\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/niemanstoryboard.org\/"},{"@type":"ListItem","position":2,"name":"How a reporter prepped to understand A.I. and the man who helped invent it"}]},{"@type":"WebSite","@id":"https:\/\/niemanstoryboard.org\/#website","url":"https:\/\/niemanstoryboard.org\/","name":"Nieman Storyboard","description":"","publisher":{"@id":"https:\/\/niemanstoryboard.org\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/niemanstoryboard.org\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/niemanstoryboard.org\/#organization","name":"Nieman Storyboard","url":"https:\/\/niemanstoryboard.org\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/niemanstoryboard.org\/#\/schema\/logo\/image\/","url":"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/02\/NiemanKOLogo.svg","contentUrl":"https:\/\/niemanstoryboard.org\/app\/uploads\/2024\/02\/NiemanKOLogo.svg","width":1,"height":1,"caption":"Nieman Storyboard"},"image":{"@id":"https:\/\/niemanstoryboard.org\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/niemanstoryboard.org\/#\/schema\/person\/ad131fac9b2e8136d6171c8575cfc226","name":"commonmedia","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/7a67dac1d5145031a5e3f8bcb411f86c4aed5a7ebe4244392196080637ed18af?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/7a67dac1d5145031a5e3f8bcb411f86c4aed5a7ebe4244392196080637ed18af?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/7a67dac1d5145031a5e3f8bcb411f86c4aed5a7ebe4244392196080637ed18af?s=96&d=mm&r=g","caption":"commonmedia"},"sameAs":["https:\/\/niemanstoryboard.org\/wp"],"url":"https:\/\/niemanstoryboard.org\/author\/commonmedia\/"}]}},"_links":{"self":[{"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/posts\/11305","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/comments?post=11305"}],"version-history":[{"count":2,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/posts\/11305\/revisions"}],"predecessor-version":[{"id":11817,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/posts\/11305\/revisions\/11817"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/media\/11308"}],"wp:attachment":[{"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/media?parent=11305"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/categories?post=11305"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/tags?post=11305"},{"taxonomy":"topics","embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/topics?post=11305"},{"taxonomy":"authors","embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/authors?post=11305"},{"taxonomy":"issue","embeddable":true,"href":"https:\/\/niemanstoryboard.org\/wp-json\/wp\/v2\/issue?post=11305"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}