They Ran an Extension Cord Next Door and Invented Google

Lex Fridman opens with the origin question: when did Eric Schmidt fall in love with technology? Schmidt grew up in the 1960s, when every boy wanted to be an astronaut. Behind the house was a literal cow pasture, and that is where they shot model rockets. He thinks that is the beginning. Today's equivalent, he says, would be video games and everything you can do online.

Fridman notes that eighth-grade math is where many people leave the mathematical universe, and quotes Edward Frenkel's *Love and Math*: if painting were taught like math, students would be asked to paint a fence and never see the art. Schmidt's path was the opposite of the fence. People like him get math-enabled early, then notice they can use it to find new insights. The great scientists, men and women, all tell a version of the same story. Somewhere in high school or college they discovered they could discover something themselves. That sense of building a thing you own, of having an impact with your name on it, is what drives knowledge acquisition.

For him the medium was programming, the feeling that he could make things that had not existed. This was before open source, though he thinks of it as the same instinct. A sixteen- or seventeen-year-old Schmidt today, he is sure, would be trying to contribute the way the open-source heroes do, failing and learning in public on GitHub. One lesson from Google is that the world is full of underused smart people. Repositories are a bank of that work, and a chance to contribute to something larger.

In 1975 Schmidt built a program called Lex with his supervisor Michael Lesk, a system still in use, which Fridman, whose name is Lex, treats as a personal thank-you. Could a 1970s engineer have predicted the personal-computer decades that followed? Schmidt is sure he could not. He was the beneficiary of people who saw more clearly. After Lex he worked at Xerox Palo Alto Research Center, where the Alto, predecessor of the modern PC and the Macintosh, was invented. The machines were rare. He drove an hour from Berkeley, skipped classes, and did whatever it took to get on one.

He knew they were consequential. What he did not understand was scaling, what happens when you have 100 million users instead of 100. Since learning that lesson he looks for things that will become platforms: mobile phones, Android, the rest. The world is numerous. People have needs. They will use the platforms, and you can build large businesses on top of them. When he sees a piece of technology now, Fridman summarizes, he asks what it looks like in the hands of a billion people. That is right, Schmidt says.

The market is competitive enough that if you cannot see a path to a million or a billion users, something more general will likely arrive and turn your idea into a lost one, or a specialized service with few users. Luxury goods can stay tiny and still be good businesses. Scale of impact is a different game. You look for common value, common pricing, common distribution, and common problems: information, medicine, health, education. People have lots of problems. He is a fan of the middle class because, by definition, there are so many of them. Starting at the high end is fine if you have a plan to get to that middle. Define yourself as only a narrow market and a general-purpose player can push you into a corner and shrink the impact you might have had.

Did writing programs feel like giving life to another entity, a dream of creating the gods? Schmidt wishes he could say yes. He found the platforms exciting, and that is what he focused on. Most of the technical people he worked with, Steve Jobs a notable exception, saw a great technological play. Relatively few understood the scale of the impact. He used MCP, a predecessor to TCP/IP, because connecting things made sense. They did not run the tape forward to the internet, then companies, then Facebook and Twitter and politics. A rare person can see compounding at scale. Ask people to predict the future and they give you six to nine months, or twelve, because that is about as far as imagination goes.

He likes a line he attributes to an old MIT professor: we overestimate what can be done in one year and underestimate what can be done in a decade. Hardware and software platforms take about that long. Self-driving cars were imagined in the 1990s. The first DARPA Grand Challenge was roughly 2004. Fifteen years later, cars were operating in a city in Arizona, with more time still needed before they would be generally available. That is the tempo he wants people to feel in their bones.

The practical version is a five-year plan whose core is a model of the underlying platforms. Moore's law, in the old shrinking-the-transistor sense, has largely halted because costs got too high. Algorithmic gains and specialized hardware still have room, and you need to know the shape of those gains. On connectivity he had long assumed the world would run mainly on fiber. It now looks like fiber plus very powerful, high-bandwidth, short-distance links for the last mile. Those networks are more symmetric, so they have different latency, and algorithms feel faster. If you know that, you design systems differently. Dreamers still matter. Science, he says, is always marked by people who were disagreeable in the useful sense: they assembled the knowledge of the time into a platform the zeitgeist had not ordered.

How do you let the crazy engineers run while still being a grown-up company? Schmidt's model is a predictable revenue stream plus the ability to do new things. Google is large enough, and managed well enough, to have a decent sense of the next year or two of revenue. That cash is the luxury of making bets. The corporate form followed. Google became Alphabet, organized around those bets, in areas of fundamental importance: artificial intelligence, medical technology, self-driving cars, connectivity through balloons, and more still coming. One way to say it is that the current business is successful enough to fund the bets. Another is that they had the wisdom to build a structure that made the bets more likely to succeed. The company turned itself into a conglomerate of bets, with an underlying Google that is itself still innovative.

Pulling that off requires a belief system with both bottoms-up and tops-down. Bottoms-up is 20 percent time: people spend a fifth of their week on whatever they want. Tops-down is founders with a keen eye on technology, reviewing constantly. Someone hears an idea, it sounds interesting, they go visit, they start assembling pieces to see if it is possible. Do that long enough and you get pretty good at predicting what is likely to work. Sergey Brin, about fifteen years before this conversation, had already written a budget rule: 70/20/10. Seventy percent of time on the core business, twenty on adjacent, ten on other. He proved mathematically, Schmidt says, that you needed the ten percent for the sum of the growth to work. It turned out he was right.

Fridman puts Schmidt in the AI-safety argument: Elon Musk, Stuart Russell, and Max Tegmark on one emphasis, Yoshua Bengio and Steven Pinker on another. What is a healthy way for the field and the public to think about mismanagement? Schmidt's first move is to name the public's curriculum: robot-killer movies, *Terminator*. He can assure you Google is not building those solutions. If they showed up, someone would notice and unplug them. Exciting films. Not this year's research agenda.

Timeframe is the whole disagreement, in his view. A hundred years from now, when the human brain is fully understood and generations of MIT scientists have worked through the implications, there will be a large pile of ethics questions about science, thinking, robots, and computers. That is that timescale. The next five to ten years are not those questions. They are how to spread a disruptive technology as broadly as possible and take the maximum benefit. The primary benefit, he says, should be healthcare and education.

Healthcare is the obvious one. Medically we are all the same, even though we somehow believe we are not. Big data about health will save lives, help with skin cancer and other cancers, ophthalmological problems, psychological disease. The promise is extraordinary: companies, startups, funds, solutions, and better lives. Education is the sibling case. Imagine every child, and even every adult, with an AI tutor, not a human, properly trained, that helps you get smarter and works on language or math trouble. The societal gain of making humans smarter and healthier is enormous, and it compounds for decades. Google has published AI-safety policies he supports, useful as a guide to research. The killer robots are not arriving this year. They are not even being built.

Has it ever been useful, on the business or intellectual side, to think fifty years out? In this industry, Schmidt says, there are essentially no examples of fifty-year predictions that were correct. Recap AI. Invented partly at MIT and a few other universities around 1956 to 1958, with original claims of a decade or two. When he was a Ph.D. student he studied it and walked into the AI winter, about thirty years, a whole generation of scientists who did not make much progress because the algorithms and the computers had not improved. Brilliant mathematicians around Geoff Hinton in Toronto and Montreal later built the deep-learning models that empower the present. The seminal work was twenty years before this conversation; popularization took another ten. That is how hard the clocks are to read.

If he goes out on a limb anyway, a few things about fifty years from now seem knowable. There will be more people alive. Platforms will have to be more sustainable because the earth is limited. The platforms that get built will still follow the principles he has been describing: more empowering of individuals, more sensitive to ecology because they have to be. He also thinks humans will be a great deal smarter because of the tools already discussed, and they will live longer. Life extension continues. A baby born today has a reasonable chance of living to 100, well past the 21st century, so we had better take care of them.

Today more than half the world lives in cities. One of the great stories of the last twenty years, he says, is the rural-to-urban migration, in the United States, Europe, Asia, and Africa. Cities get more crowded. Believe it or not, health gets better, productivity gets better, IQ and educational capability improve. People moving to cities is good news, provided the cities are livable and safe.

Fridman lists Steve Jobs, Elon Musk, Larry Page, Sundar Pichai, and Schmidt himself, from calm sages to mad geniuses, and asks what is comparable. Schmidt learned as a young executive that there is no single formula, even if schools try to teach one. Some people just know what to do and move fast; those are often entrepreneurs. Some are systems thinkers and planners, more his type, somewhat conservative, thorough in execution, a little more risk-averse. Some are slightly insane in the charismatic sense: they feel it and drive it. What unifies the names on the list is very high intelligence. They saw the world quicker, processed information faster, were not always right, and stayed on top of it. They also started young. Jobs at Apple around 18 or 19, Gates around 20 or 21, Zuckerberg at 19 or 20. By 30 they had a decade of people, products, shipments, press, and business. That is a different pile of experience from the people busy getting Ph.D.s.

Celebrate them, he says, because the extra life experience forms judgment. At the top of these organizations the easy questions are gone: how to design the building, where to put the colors, what the box should look like. What remains, which is why the rooms are interesting, are the hard problems of employees, customers, and innovation. The companies are demonstrably different culturally, not cut from one mold. Compensation, values, behavior under the same input, all differ. There is proof that diversity works.

Fridman then puts the 1998 search market on the table, Lycos through Yahoo, all good products, and asks whether a human-centered-AI researcher at MIT should copy Larry and Sergey, given the low odds. Schmidt's answer is that people in universities are always trying to study the chaotic nature of entrepreneurship. Larry and Sergey did not have that conversation. They just did it. They sensed a moment when a mass of data needed organizing and they had a better algorithm. Human-centered AI, he says, must have new approaches, including work at MIT on how children learn, Josh Tenenbaum and others, that is different from what most of the field is doing. Successful entrepreneurs are not as crazy as they sound. They see an opportunity in what has already happened. Travis Kalanick's version is sitting in Paris, unable to get a cab, holding a smartphone. The rest is history. Find the equivalent of that moment.

The Google beginning was simple and sharp. Two Stanford graduate students kept a search engine in their room, pulled power from next door with an extension cord, moved into a house that was world headquarters for five people, and raised $100,000 from Sun founder Andy Bechtolsheim, Dave Cheriton, and a few others. Simple start, powerful insight, a real innovation: PageRank, now one of the most cited papers in the world. That pattern, Schmidt says, is replicable. On money and meaning he is equally direct. Happiness research finds that above a relatively low threshold, money does not move the number. Happiness tracks meaning, family, impact. People are happiest in service to others, not themselves, which cuts against the press excitement about powerful leaders. Fortune is a responsibility: educate others, and use the wealth to advance society. In his case that means AI and machine learning on education, inequality, and the middle class. It does not matter what you do. It matters that you believe in it, that it is important to you, and that you spend a life on it.