Pamela McCorduck: AI Began With the Ancient Wish to Forge the Gods

Lex Fridman opens on a line he stumbled into from Pamela McCorduck's 1979 book *Machines Who Think*: artificial intelligence began with the ancient wish to forge the gods. That sentence, he says, draws a connecting line from day-to-day science and engineering to popular stories, science fiction, and myths of automatons that go back centuries. Through her literary work she spent time with the seminal figures of the field, including the founding fathers from the 1956 Dartmouth summer workshop where AI was launched. He reached out hoping to learn what those early days felt like, and how their dreams still reverberate.

McCorduck had already published two novels when she sat under a portal at a Stanford house rented for the summer and thought she should write a novel about these weird people in AI. Then the idea turned. Don't write a novel — write a history. Interview them, splice it together, voila, instant book. It was much harder than that. Nobody else was doing it, which she took as the opportunity. John McCarthy thought it was a nutty idea. The field had not evolved yet, and he had a mathematical topic he preferred she write instead.

Her reply became a small legend in the telling: she was not a woman in search of a project. This was what she wanted to do; she hoped he would cooperate. McCarthy muttered, then said fine, it was her time. The founders were alive and kicking, at the height of their research careers, and still willing to sit with a woman who said she thought she might write a book. That still humbles her.

Before she met her husband, Joe Traub, she was already immersed in the field. She had been Ed Feigenbaum's assistant at Stanford, and before that had worked on *Computers and Thought*, the first textbook of AI readings, edited by Feigenbaum and Julian Feldman. They compiled it for a practical reason: they were trying to teach AI at Berkeley, and there was no internet to fetch papers, so students were being sent from journal to journal. She was an English major. What did she know? She was fascinated anyway. That literary background, she argues, is part of the continuum AI grew out of — the same impulse, not a side dish.

The funding finally came as a private grant, without her knowing the source at first: Ed Fredkin at MIT. He was a wealthy man who liked what he called crackpot ideas, and he considered a history of AI to be one. He was willing to support it. She is ever grateful. Fridman notes that a history-of-science treatment of this kind has hardly been written since. McCorduck does not argue the point. She had seen, from the first, that the story was older and stranger than a decade of papers.

Even when she finished the book she did not know if the field would succeed. The final chapter, she admits, is wishy-washy. What had interested her was the idea of intelligence outside the human cranium — a phenomenal idea for an English major sitting among engineers. Feigenbaum, a dear friend she still sees every month or so, told her that when *Machines Who Think* first came out he thought all the front matter was baloney. Ten years later he said he saw what she was getting at: this had been a human impulse for thousands of years. It is hard, she adds, to step back to that picture when you are down at the algorithmic level just trying to make something work.

Asked who stands out from Dartmouth in 1956, she names four founding fathers: Allen Newell, Herbert Simon, Marvin Minsky, and John McCarthy. They were not only in the room. Newell and Simon arrived with a working program called the Logic Theorist. Everybody else had great ideas about how they might do it, she says, but they were not going to do it yet. Claude Shannon was there too; Feigenbaum was not — he would have been an undergraduate then. The four were extraordinary, and they did not walk the same road.

Newell and Simon were cognitive psychologists. They wanted to simulate aspects of human intelligence and found they could do it on a computer. Minsky just thought it was a really cool thing to do. McCarthy had gotten the idea in 1949 as a Caltech student, listening to a lecture she can no longer name off the top of her head, and thought it would be fun to figure out how. He took a very mathematical approach; Minsky was a hybrid. Over the years the definition of intelligence expanded past anything those rooms would have allowed. Biologists now talk comfortably about the intelligence of a cell, of any kind of brain, of cephalopods. An octopus is intelligent by any measure. Nobody would have thought that in the 1960s or even the 1970s.

She heard the primatologist Frans de Waal say that AI taught the field the questions to ask. Fridman agrees: you start asking when you try to build it; the attempt puts a mirror to ourselves. McCarthy, looking forward, saw the technology as a great boon for human beings, reason enough to do it. He had fables about how the mathematics showed the benefits, and a technological answer for every technological objection. One of his favorites, which he presented to her several times, was what he called the literary problem. Literature has conventions. One of them is a hero and a villain; the hero is usually human, the villain a machine. That, McCarthy said, is just not the way it is going to be — but it is the way we are used to telling the story. She thought he was right.

If AI is only a postwar math-and-engineering story, a long prehistory drops out. McCorduck splits that prehistory into two attitudes. The Hellenic view says robots are great: they help, they work in Hephaestus's forge, and people welcome them. The Hebraic view sits on the second commandment — thou shalt not make any graven image. Do not start imitating humans; it is forbidden. A lot of the reaction to artificial intelligence, she says, has been a sense that the work is somehow wicked, somehow blasphemous. There will be downsides. Blasphemy, she insists, is not one of them.

The prehistory runs further than the Golem and early twentieth-century robots. A classical scholar had recently told her that the Odyssey is full of robots. How does Odysseus's ship get from place to place? He does not have the crew to do it. It is magic. It is robots. The Iliad has them too. Toward the end of the nineteenth century and the beginning of the twentieth, scientists actually tried to make the thing happen and failed; they did not have the technology. Charles Babbage, in the 1850s and 1860s, already saw that what he was building was capable of intelligent behavior. When the British government finally said that was enough funding, he and Ada Lovelace discussed playing the ponies with the machine, among other schemes for raising money.

On her coffee table sits an annotated *Frankenstein*. She calls it a tremendous novel, beautifully perceptive. The creature never has a name; Frankenstein is the man who created him. The creature wants to be loved and accepted. Only when Frankenstein turns away — runs the other way — and the creature is without love does he become the later monster. Who the villain is, she says, is unclear. The people who drive him away bring out the worst. At one point he befriends a blind man, serves him, and they become close; when the sighted family returns, they see a monster, and the story breaks. What she had not known is that Mary Shelley and Percy Shelley were great readers of American abolitionist literature and picked it up wholesale: you are making monsters of these people because you will not give them the respect and love they deserve.

Fridman asks whether, once machines are a little more intelligent — even a Roomba — a darker habit of abusing the other will come out. She does not worry about it. She can imagine it happening. What AI has to offer the human race, she thinks, will be so attractive that people will be won over. The primal fear he is circling, she says more generally, is that a machine that can think will out-think us and replace us: a kind of mortality. Practical worries sit on top of that — bias, discrimination baked into algorithms, social networks whose recommenders steer the masses. The fear underneath still feels older than any of those systems.

Traub was not in AI. He worked on algorithmic complexity. There was, she says, a period when AI felt like an outcast inside the new discipline. In 1974, computer science — hardly ten years old — was asked by the National Science Foundation, or possibly the National Academies, she cannot remember which, to tell fellow scientists what the field was and what it meant. The drafters wanted to leave AI out. They put it in because Don Knuth said it was important and you could not just omit it. An AI person could not have made that argument and been believed. Knuth was believed. Traub, working on what some would call the real stuff, said in plain English, again and again, that the smartest people he knew were in AI.

Expert systems in the 1980s, Fridman notes, hit walls after a boom. McCorduck's reaction to the phrase AI winter is blunt: the whole thing is a crock. She looks at the basic research of the 1980s, which was supposed to be the freeze, and sees work that was really important, laying down things nobody had thought about before. It was basic research. You could not monetize it. Hence the winter. Scientific research, she says, goes in fits and starts. It is not a nice smooth sequence of this-follows-this. Fridman pushes: winters are often the fault of hype, not of researchers. She takes it back in part. Sometimes it is the researchers. Some of them believed they were only a few years away.

The conversation is 2019. Elon Musk, she has just heard, believes in mass deployment of autonomous vehicles in a year. A year? She believes autonomous vehicles will come. She does not believe it will be a year. Fridman fills in the commercial mechanism: companies promise 2021 or 2022, the date arrives, the cars do not, billions have been spent, no money has come back, and the field is abandoned intellectually except for a few basic researchers. She is more worried about science being killed by missing or untimely funding than by a metaphor. Britain's Lighthill Report, as she remembers it, said there was no use putting money into this, that it was going nowhere. Underneath were academic factions — Edinburgh against Cambridge, Cambridge against Manchester. It had a hard effect. They have come roaring back with DeepMind, she says, but that is one guy and the visionaries around him.

She admires science perhaps more than she admires commerce. Commerce is fine; everyone has to live. Science has a much longer view, and it continues almost regardless of what is saleable and what is not. The winter, in that telling, is something that happens on the commerce side. Fridman calls it a beautifully optimistic message. She agrees that key scientists keep working out of love for the problem, scraping up enough funding to stay alive. She also grants his worry: a huge amount of money is in deep learning now, and another round of over-promising could slow student pipelines. It does affect that. The distinction she will not drop is between work you can monetize and work that lays the road.

While she was writing *Machines Who Think*, Herbert Simon became a close friend. He walked past their house on Northumberland Street in Pittsburgh every day after work. She would be putting the cover on her typewriter, lean out the door, and ask if he wanted a sherry. He almost always did. He would stop in and they would talk for an hour, two hours; her journal records afternoons that ran to three. They did not talk much about AI. They talked about life. Both loved literature. Simon had read Proust in the original French twice, all the way through; she had read him in English. They talked about languages, about music, about art. He was enough of a painter that he had to give it up, afraid it was interfering with his research. It was chat. It was very warm.

One summer she complained. She was teaching in the English department at the University of Pittsburgh. Her students had all the really interesting conversations — the meaning of life. She had university meetings about the photocopying budget and whether the Romantic poetry course should be one semester or two. Simon laughed. He knew what she meant. You could do something about that, he said. He and his wife Dorothea, Dot, used to hold a salon at the University of Chicago every Sunday night, essentially an open house, and people knew it was not for small talk. It was for a topic of depth. His advice was to choose the topic ahead of time.

They exchanged mail over the summer — U.S. post, no personal email — and she organized it. Eight people: Allen Newell and his wife; Simon and Dorothea; a novelist newly in town, Mark Harris, most famous then for *Bang the Drum Slowly*, and his wife Josephine; and McCorduck and Traub. They met monthly and sank their teeth into hard topics. It was great fun. For a writer marooned in committee work, the salon was the repair. The field's founders were not only interview subjects. They were, at least once a month, a room where the conversation was allowed to be serious.

How did her own views change? She was a little skeptical that the whole thing would work out. It did not matter. The project was audacious. In some ways it has not worked out the way she expected. There is a wonderful lot of apps thanks to deep learning, she says, but those are algorithmic. In symbolic processing there is very little yet. That is the field that lies waiting for industrious graduate students. Expert systems will re-emerge, Fridman suggests, with some new breakthrough. She does not argue. What did surprise her was the center of gravity.

She had been away from AI long enough to get involved with complexity at the Santa Fe Institute. A young student told her what he was doing, and she thought: algorithms, yes, they are on the surface, but they are not the main event. They became the main event. That surprised her. Everyone knows the downside, she adds. If you use an algorithm to make decisions based on a gazillion human decisions, baked into it are all the mistakes humans make — the bigotry, the shortsightedness, and the rest. Symbolic AI, as she saw it in 2019, had really kind of dried up. The excitement in the room had moved.

Asked about timescales, she will not predict. Geoffrey Hinton came up with the deep-learning algorithm in 1986, and it took years for the technology to be good enough to apply. She would not even try to call the next leap. Young people think they have just found something special and will change the world tomorrow. The spirit of the people, the conversations she hears through the grapevine, feels the same as the roots she remembers: the same kind of excitement, the same sense that this is really going to make a difference. It will. It has. The schedule is what nobody owns.

One of the last projects she did on AI was the work of Harold Cohen, a painter who became deeply involved with the field. His lifelong program, AARON, was built to reflect his own cognitive processes. She wrote a book about it, and they had a lot of friction. The book was published and, in her phrase, fell into a ditch. She thought that was it: she was finished, time to do something different. By chance it was a sabbatical year for Traub. They spent two months at the Santa Fe Institute, two months at Caltech, and the spring semester in Munich. Those two months in Santa Fe were restorative.

The Institute was very small then, in an office complex on Old Santa Fe Trail. Everybody kept their door open. You could crack your head on a problem, fail to get it, and walk in on Stuart Kauffman or any number of people and say you did not get this, could they explain. She was talking about complex adaptive systems with Murray Gell-Mann and told him what Cohen had done. It sounded to her like a complex adaptive system. He said yes, it is. AARON, she realized, had kissing cousins all over science and economics. She was so relieved. Her instincts were okay. She was doing the right thing. She just had not had the vocabulary. That was one of the things Santa Fe gave her. If she could have rewritten the Cohen book — it had just come out, she could not — she would have had the words to explain what AARON was doing.

George Cowan, then head of the Institute, thought it might be a nice idea if she wrote a book about the place. She had her eye on some other project and said she would love to, but it was not going to work at that moment. He was right and she was wrong. She wishes she had done it: a road not taken. Fridman asks how she found the guts to walk into rooms of large scientific egos as a writer from literature. She never picked up that anybody was intimidated by her. People have enough patience with writers, she says, if they think a book will come of it, and they can tell if there is a sparkle in the eye, if the visitor gets it. The time she is describing is 1990, 1991, 1992. Traub later became external faculty; they bought a house in Santa Fe and spent summers there. She loved the audacity of the ideas. That is what appeals to her.

She is not a big fan of the singularity as Ray Kurzweil has presented it. If she understands his view, there is going to be a moment when machines are smarter than humans, and then it is game over — reservations, or whatever the ending is. Machines are already smarter than humans in some ways, and have been since adding machines were invented. It is not going to arrive like some Oedipal crossroads where they meet and one side is finished. Calculators already outdo us at arithmetic. That game is over, and we are still here. Human-like intelligence — a friend, a deep connection, something past the Turing test — is a different dream, and she has dreamt it.

In a book she wrote with Feigenbaum there is a little story called the geriatric robot. She dreamed it up to talk to college students about what AI might be, and they were rolling in the aisles as she elaborated. The robot does not just clean you up, feed you, and wheel you into the sun. Its great advantage is that it listens. Tell me again about the great coup of 1973. Tell me again how awful or how wonderful your grandchildren are. It is not hanging around to inherit your money. It is not hanging around because it cannot get another job. This is its job. When the story went into the book, the *New York Review of Books* took her hide off: this is just what we thought about these people in AI, they are inhuman. She thinks it would be a good thing for the world.

She is pushing 80 in this conversation. By the time she needs that kind of help, she also wants the robot to roll itself into a corner and shut up. Fridman lingers. Does she really? She has watched friends bring help into the house. Some were lucky. Some had people who wanted the television on all day, or talked on their phones all day. Human assistants, even when they are assisting, still care about themselves. An AI assistant can optimize for you. He pushes the other way: people do not necessarily want a perfect servant. There is push and pull, a little tension, a little mystery, and that is hard to get right. She grants a little backtalk, a little. He adds that social media has made people lonelier, especially the young, and that a listener might ease some of that longing. She does not put a date on human-level conversation. The field has been too full of surprises. Herb Simon was terrible about predicting when this or that would happen, and he was a sensible guy. His quotes are still used as legend.

Does she have concerns about existential threats — the Musk and Sam Harris register of worry? It takes up half a chapter in her book. She called it the male gaze. The term comes from film criticism: most movies made from the male point of view, women as objects not subjects, without agency. When Elon and his pals, Hawking and the rest, came out saying AI was going to eat our lunch, our dinner, and our midnight snack, she thought: what? She said to Feigenbaum that these guys have always been the smartest on the block, and here comes something that might be smarter, so let's stamp it out before it takes over. He laughed and said he had not thought about it that way. She had.

That, for her, is the male gaze. Suppose these things do have agency. Wait and see what happens. Can we imbue them with ethics, with a sense of empathy, or are they just going to be — she does not finish the insult. We have had centuries of guys like that. Fridman hears the ego threatened, unable to think patiently about what the technology could become. She does not deny the possibility of harm. She denies that the first move should be to kill the thing that might out-think you. Patience, ethics, and empathy are the experiment she wants, not a preemptive stamp.

With Nancy Ramsey she wrote a book on four possible futures for women in science and tech. The method is to take aspects of the present and push each to an extreme: backlash, one step forward and one step back, a golden age of equality. The golden age was the hardest chapter to write. She used a Santa Fe model, the sandpile effect. Sand drops very slowly onto a pile, which grows and grows until suddenly it breaks apart. In a way, she says, Me Too did that. It was the last drop of sand. A perfect example of the sandpile effect. It did not change all of society. It really woke a lot of people up, and that made her feel good.

Has enough changed? Not enough. The way women are ground down in computing is simply unbelievable. Look at backlash, she says, and you can point to fights over abortion and the rest; the present already contains the dark scenarios the book extrapolated. Asked whether she is optimistic after Me Too, she will not pretend that solving one kind of harm solves everything. She is, still, a congenital optimist. She cannot help it. The futures book was never prophecy so much as a way of looking hard at what was already on the table.

What does she worry about? The things we cannot anticipate. There is going to be something out of left field that makes us say we were not prepared. She is generally optimistic. When she first took an interest in AI, like most people in the field, more intelligence seemed like more virtue. What could be bad? In a way she still believes that. Her notion of intelligence has broadened. There are many kinds, and we need to imbue machines with those many kinds — not only the sort that lived in a 1960s definition of a clever man at a blackboard.

She is finishing a memoir. Looking at the entirety of it, the biggest thing — and it was not really a surprise — is how lucky she was: to have access to the beginning of a scientific field that is going to change the world. How did she luck out? Her view of things has widened. Then she returns to the feminist thread, which she says was subconscious. She wanted AI to succeed because she was so tired of hearing that intelligence was inside the male cranium. If something out there was not a male, thinking and doing, that would put a lie to the whole notion.

She did not know that until one night with Harold Cohen, a glass of wine, maybe two. He asked what drew her to AI. She gave the standard answers: the smartest people she knew, a great project, and so on. Then another sentence bubbled up. She wanted something besides male smarts. Fridman calls it brilliant. AI, she says, really humbles all of us, and humbles the people who need to be humbled the most. He thanks her. The hour ends on that unplanned confession, which in hindsight organizes the novelist in the lab, the book about women's futures, and the robot whose job is to listen.